doi
stringlengths 28
28
| title
stringlengths 19
311
| abstract
stringlengths 217
5.08k
| plain language summary
stringlengths 115
4.83k
| article
stringlengths 3.87k
161k
|
|---|---|---|---|---|
10.1371/journal.pcbi.1007374 | Heterogeneous responses to low level death receptor activation are explained by random molecular assembly of the Caspase-8 activation platform | Ligand binding to death receptors activates apoptosis in cancer cells. Stimulation of death receptors results in the formation of intracellular multiprotein platforms that either activate the apoptotic initiator Caspase-8 to trigger cell death, or signal through kinases to initiate inflammatory and cell survival signalling. Two of these platforms, the Death-Inducing Signalling Complex (DISC) and the RIPoptosome, also initiate necroptosis by building filamentous scaffolds that lead to the activation of mixed lineage kinase domain-like pseudokinase. To explain cell decision making downstream of death receptor activation, we developed a semi-stochastic model of DISC/RIPoptosome formation. The model is a hybrid of a direct Gillespie stochastic simulation algorithm for slow assembly of the RIPoptosome and a deterministic model of downstream caspase activation. The model explains how alterations in the level of death receptor-ligand complexes, their clustering properties and intrinsic molecular fluctuations in RIPoptosome assembly drive heterogeneous dynamics of Caspase-8 activation. The model highlights how kinetic proofreading leads to heterogeneous cell responses and results in fractional cell killing at low levels of receptor stimulation. It reveals that the noise in Caspase-8 activation—exclusively caused by the stochastic molecular assembly of the DISC/RIPoptosome platform—has a key function in extrinsic apoptotic stimuli recognition.
| Death receptors are targets of novel cancer therapeutics. Most of them signal through flexible multiprotein platforms to either activate apoptotic or necroptotic cell death, or propagate cell survival and pro-inflammatory signals. We focused our study on the role of dynamic assembly and composition of these platforms in the initiation of cell death at the single cell level. Since the assembly is slow through the competitive nature of protein binding within the platforms core we developed a stochastic mathematical model of the death inducing signalling platform. Our model provided an explanation for delayed cell death and fractional killing upon the death receptor stimulation. Additionally, we found that the variability in the cell death response arises through the random assembly initiates a slow noise-prone ramp activation of initiator Caspase-8 spontaneously triggering the apoptotic cascade. Our computational simulations predicted high variation in the time required for cell death induction at the single cell level and highlighted a significant role of death receptor clustering in effective Caspase-8 activation. Our knowledge and data driven model captures detailed processes governing the early events of cell death initiation and can be used to guide the development of more rational combinational treatments against cancer.
| Apoptotic signalling cascades are designed to irreversibly lead to cell death once specific death thresholds are overcome [1,2]. Activation of caspases plays a central role in this process. In certain scenarios, apoptotic cell death signalling is interrupted. This may lead to the activation of other forms of cell death or escape from cell death altogether.
Death ligands (DL) bind to death receptors (DR) at the plasma membrane and have been developed as novel cancer therapeutics. However, many cells in our body are exposed from time to time to endogenous DLs, such as TNF-α and TRAIL, without induction of cell death. Several studies have shown that while binding of DLs to DRs can induce apoptosis, not all cells will respond to DR stimulation with cell death, and only a fraction of the cell population will undergo apoptosis even if DLs bind at death-inducing concentrations [2–6] (Fig 1A). Interestingly, in vivo studies have shown that fractional death resistance has no direct association with the amount of DRs expressed on the plasma membrane [7,8]. Therefore, cell signalling activated by extrinsic ‘death’ signals is rather encoded downstream of receptor binding.
Binding of DLs to dedicated DRs triggers either the formation of receptor-associated Death-Inducing Signalling Complexes (DISC) (‘Complex I’) in proximity to the plasma membrane, or RIPoptosome complexes (‘Complex II’) in the cytosol [5,9–17]. Both complexes provide a platform for the activation of the initiator Caspase-8 (Casp8). For the activation of Casp8, the inactive pro-form of Casp8 (ProCasp8) must undergo autocatalytic activation. This is achieved through ProCasp8 dimerization and sequential inter- and intradimer cleavage, a process which results in the release of active Casp8 [18–20]. The dimeric ProCasp8 association-dissociation balance has been suggested to play a crucial role in the molecular control of apoptotic responses after DR activation [21]. However, as demonstrated by mutagenesis studies, ProCasp8 dimerization alone is not sufficient to enhance apoptotic responses in vivo [22]. Instead, formation of the DISC or RIPoptosome platforms are necessary for effective ProCasp8 dimerization and Casp8 activation [10,23,24].
Apart from apoptosis initiation, DR-induced complexes also initiate necroptosis by accumulating heterodimers of receptor-interacting proteins (RIPs), RIP1 and RIP3 (RIP1/3), and the formation of filamentous scaffolds [25–28]. Formation of such ‘Necrosome’ platforms activates the mixed lineage kinase domain-like (MLKL) pseudokinase. MLKL activation triggers necroptosis, a cell death distinct from apoptosis [29–31]. In theory, activation of DRs in individual cells could lead to both apoptosis and necroptosis signalling through the formation of different platforms. However, if RIP1/3 proteins are close to the site of Casp8 activation, RIP1/3 is cleaved by Casp8 [32]. This cleavage eliminates the kinase activity of RIP1/3, and consequently necroptosis activation is suppressed [9,33–36] (Fig 2B). This suggests that if one type of cell death is triggered in a given cell, the other type of cell death is suppressed, i.e., that the two types of cell death are mutually exclusive.
Previous studies of the apoptotic signalling network activated by DRs have identified that variability in death signalling arises from the process preceding the mitochondrial outer membrane permeabilization (MOMP). This process triggers Casp8-mediated cleavage of the pro-apoptotic Bid protein [2,4,37], which mediates MOMP and leads to cytochrome-C release, apoptosome formation and executioner caspase activation [38].
To understand cell death decision making in more detail, we created a mathematical model which incorporates the central events prior to Bid cleavage. The model was constructed to estimate apoptotic and necroptotic pathway initiation through the random assembly of the DISC/RIPoptosome platform. As a multiprotein platform with diverse functionality, we hypothesised that the random and stochastic process of its assembly may lead to the heterogeneous cellular responses (Fig 1A and 1B). Combining this model with experimentally derived sets of quantitative protein profiles and literature-based catalytic and binding rates, we simulated the heterogeneous responses of HeLa cells to DR activation. By modelling different conditions of DR stimulation and clustering, we investigated in particular how heterogeneous apoptotic responses arise, which role the random assembly of DR-induced platforms play in determining death delay at the single cell level, and how DR clustering facilitates death signalling. Our analysis reveals that the noise in Casp8 activation exclusively caused by the stochastic molecular assembly of the DISC/RIPoptosome platform has a key function in the low level extrinsic apoptotic stimuli recognition.
Apoptosis inducing DRs such as Tumour Necrosis Factor Receptor 1 (TNFR1) and Death Receptors 4 and 5 (DR4/5) are expressed at comparable protein levels in HeLa cells [39]. Additionally, it is known that their protein expression level is correlated with the receptor abundance on the cell surface [8]. High variation in TNFR1 surface abundance were estimated in previous studies ranging from 300 to 3000 molecules per single HeLa cell [40,41]. To get more accurate estimates, we performed the single cell quantification of TNFR1 membrane expression in HeLa cells employing the QuantiBRITE phycoerythrin beads based assay (see S1 File). We determined that the average number of TNFR1 does not exceed 905 receptors per cell. We further used this quantity as the reference in our comparative quantification of DR4/5 receptors based on MS data set (S1 File). Thus, we calculated that DR4 and DR5 receptors are present on HeLa cell surface in an average amount of 769 and 926 monomeric receptors, respectively (Table A in S1 File).
Next, we estimated the amount of the DR complexes associated with DL at the single cell level. Due to the fact that the DR-DL association is generally much quicker [42] than the downstream processes such as ProCasp8 dimerization and subsequent Casp8 activation [43], we applied the rapid equilibria approximation to calculate the amount of DL bound receptors. According to the law of mass action the time evolution of the amount of DR-DL complexes is
d[RL]dt=kon[R][L]-koff[RL],where[R]=[Rtotal]-[RL]
(1)
Where [Rtotal] is the total number of receptors per cell (Table A in S1 File), [RL] is the number of DR-DL complexes and [L] is death ligand concentration (Table B in S1 File).
Setting the RL to the rapid equilibrium
d[RL]dt=0
(2)
From (1) we calculated the average number of DR-DL complexes per cell as a function of L, Rtotal and the DL dissociation constant Kd
[RL]=[Rtotal](Kd[L]+1)
(3)
The minimal unit of the active DR-DL complex is the trimer [44]. The trimeric DR-DL complex gives birth to a single DISC platform which internalizes within the subsequent 10–15 minutes [45,46]. If the DISC has not bound to cellular Inhibitor of Apoptosis Proteins (cIAPs), cIAP1 or cIAP2, then it either releases active RIP1 protein into the cytosol [47] where it can form RIPoptosome or Necrosome platforms (Fig 2B) (as in case of TNFR1), or it makes active RIP1 protein accessible for further RIP1/3 and ProCasp8 proteins accumulation on the DISC itself (as in case of DR4/5 activation) [5,17]. Therefore, in the modelling routine each activated DISC was translated into a single RIP1 protein molecule which is available immediately after DL introduction to the cell culture.
Trimeric DR-DL complexes tend to organise high order clusters in cellular membranes [44,48] and bring several associated DISC/RIPoptosomes into close proximity. Such clustering stimulates more efficient signalling [49] and enables ProCasp8 activation not only by dimerization on the single DISC/RIPoptosome but also by synchronised binding of two ProCasp8 monomers with two independent DISC/RIPoptosomes within one cluster. To introduce DISC/RIPoptosomes clustering processes in the model, we estimated the number and the size of the DR-DL clusters based on the experimentally derived DR-DL probability distribution from a study published earlier by Fricke and co-workers [44]. We calibrated probability redistribution from the total pool of activated trimeric DR-DL complexes, calculated in the previous step, to the clusters of different size (see S1 File). Using these probabilities, we assigned for each random DISC/RIPoptosomes complex formed its associated cluster. The final algorithm assumes that DISC/RIPoptosomes complexes within one cluster are able to first encourage the activation of ProCasp8 by direct dimerization (cis-activation) and subsequently activate ProCasp8 via simultaneous binding within closed proximity (trans-activation) (Fig 2A). Thus, this information about the amount of the activated DR-DL complexes and their clustering conformation served as an important input for the model. The scenario of non-clustering DR signalling was studied as well by setting the probability of trans-activation of Casp8 within DISC/RIPoptosomes complexes cluster to zero. This scenario is hereafter referred to as disrupted clustering.
We have developed a core model capturing the cascade of intracellular reactions that are essential for the initiation of the apoptosis. The model reactions are partitioned into two modules: a stochastic and a deterministic module (Fig 2).
The first stochastic module represents the process of stochastic assembly of DR-induced DISC/RIPoptosome multiprotein platform which facilitates initiation of ProCasp8 dimerization and self-activation by cleavage (Casp8*; activated Casp8 dimer in Fig 2E). We implemented this module with the direct Gillespie stochastic simulation algorithm [50,51] which accounts for molecular fluctuations and slow association and dissociation rates following each component of the platform individually. It assigns the reaction propensities in probabilistic terms. The binding propensities of ProCasp8 together with its binding partner protein, Fas-associated death domain protein (FADD), and competitor protein RIP1/RIP3 that comprise the core scaffold of RIPoptosome are calculated from the concentrations that we quantified experimentally in HeLa cell culture (Table D in S1 File). FADD protein is crucial for apoptotic initiation [35,52]. This protein consists of both a Death Effector Domain (DED) and Death Domain (DD) which are specific motifs for ProCasp8 [53] and RIP1 [16,54] self-oligomerization respectively. Through these domains, ProCasp8 and RIP1 are bridged via FADD (as shown in grey in Fig 2B). RIP3 protein can form homo-oligomers, but can also associate with RIP1 scaffolds through the RIP homotypic interaction motif (RHIM), forming amyloid structures [27,28] (Fig 2B). Intensive recruitment of RIP3 molecules to the amyloid triggers transphosphorylation of RIP3 by RIP1 with consequent transmission of phosphate groups to the MLKL pseudokinase. Phosphorylated MLKL executes necroptosis [25,30]. Therefore, in the absence of FADD and joint Casp8 activation platforms these structures spontaneously trigger necroptosis [35,55,56] (necrosome complex; purple in Fig 2B). Additionally, we quantified the concentrations of the cellular FLICE (FADD-like IL-1β-converting enzyme)-inhibitory protein (c-FLIP). As a DED-containing protein, cFLIP in its short (cFLIPs) and long (cFLIPl) form, can be recruited to the ProCasp8 platform abrogating or restricting activation of Casp8 [53,57,58] (cFLIP molecules; light and dark blue in Fig 2B). In addition to this suppression, Casp8 activation can be disrupted by binding its own processed DEDs which may remain in the cytosol (DED1-DED2; white in Fig 2B).
The second deterministic module mimics the activation of two effector caspases, Caspase 3 (Casp3) and Caspase 6 (Casp6) which is triggered by stochastically activated Casp8. Pro-forms of both caspases form stable dimers at physiological concentrations [59]. By cleavage, Casp8 activates Casp3 (Casp3*; activated dimer of Casp3 in Fig 2E) [60]. Casp3* activates Casp6 (Casp6*; activated dimer of Casp6 in Fig 2E) [61,62] and has autocatalytic function cleaving ProCasp3 [63,64]. Finally, Casp6* can cleave free ProCasp8 (Casp8*; cleaved monomer of Casp8 in Fig 2E) [64–67] however Casp8 becomes active only after a very slow dimerization (Casp8*) [19,21]. Previous models suggest that this effector caspase feedback upon weak DR stimulation probably can accelerate Casp8 activation which was initially started at the DISC or RIPoptosome platform [68]. However, the feedback can be inhibited by X-linked IAP (XIAP) which tightly binds Casp3 and, further, marks Casp3 with ubiquitin that leads to its proteasomal degradation [69,70]. The overall dynamics of Casp8 activation can be tracked quantitatively with a Casp8-specific FRET cleavage probe (FRET, Fig 2E). The fixed threshold rate of this FRET probe cleavage accurately determines the moment of MOMP in HeLa cells [2]. Based on the mass action and conservation laws, the time evolution of the variables that comprise this module were modelled by a deterministic system of ordinary differential equations (ODE) (details in S1 File).
All protein concentrations and parameters used in the model are provided in Tables D and E of Materials and Methods file (S1 File).
The estimated weight of the RIPoptosome after short DR-targeted stimulation may exceed 2MDa [10,24,29]. To reproduce the RIPoptosome growth and composition we first employed the stochastic modelling module simulating the assembly of the individual RIPoptosomes at the single cell level. RIP1 on its own forms unlimited filaments in vitro [28], however, in the cell the long-term RIPoptosomal filament growth is limited by the cell volume and the stiffness of the cellular components. We followed unlimited filament growths without implementation of these physical limits, focusing on the initial dynamics of RIPoptosome progression. Fig 3 and S2 Fig illustrates the simulated molecular composition of RIPoptosome in the single HeLa cell treated with a dose of 5 ng/mL of the DL (rhTRAIL).
The composition and the time evolution of individual RIPoptosomes within single cell differed from one to another. Consequently, the size and, therefore, molecular weight of those RIPoptosomes varies as well. As an example, we display the composition change in a few randomly chosen RIPoptosomes over the first 20 min with 1 min step interval (Fig 3A, S2 Fig). Next, we calculated the progression of the molecular weight of a complete cellular pool of RIPoptosomes as simulated by the model. Interestingly, we found a high degree of variation between the RIPoptosomes formed within the same cell (Fig 3B). Our simulations confirmed that in HeLa cells, the most populated protein within each RIPoptosome is RIP1 through its highly stable association mechanism. This is explained by the RHIM domain binding property that shares homology with β-amyloids assembly domains. Simulation of the model revealed that the RIP1 filaments formation is triggered immediately after DR stimulation (Fig 3A). The model also predicts that it would be possible to observe RIPoptosomes of size 2 MDa only 5 minutes after DR stimulation (Fig 3B).
FADD recruitment to the fraction of the high molecular weight complexes is persistently increasing with post treatment time [29]. Our simulations show as well that the abundance of FADD within a single RIPoptosome increases linearly with time progression (Fig 3C) in conjunction with the filament growth. As a result, the abundance of FADD on average will not exceed the amount of 10 molecules per origin within the first two hours. Moreover, this abundance is independent of DL dose. Thus, a low dose of 5 ng/mL of the DL and a high dose of 50 ng/mL will result in similar FADD abundance (S4A and S4B Fig).
On the contrary, ProCasp8 recruitment in the single cell is most abundant in the RIPoptosome of the lower molecular weight (Fig 3B). The binding of the ProCasp8 or its DEDs domain to the end of the filament blocks the RIP1 recruitment and therefore also blocks intensive filament growth by competition. The population average over 600 cells shows that ProCasp8 abundance per RIPoptosome (origin) saturates after 2 hours of stimulation (S4C and S4D Fig). This relative abundance does not vary significantly for doses of 5 or 50 ng/mL of the DL and is unaffected by the clustering or non-clustering assumption in the model. These rapid saturation dynamics of ProCasp8 compared to linear FADD translocation has been observed earlier in experiments where no co-binding of FADD and Casp8 has been observed after 1 hour of stimulation but has become apparent at the second hour [29].
Molecular fluctuations in the RIPoptosome composition within single cells cause the fluctuations in the active Casp8 abundance (Fig 3D). Stochastic single cell Casp8 activation traces for 5 ng/mL dose simulation with the corresponding per cell accumulation of Casp8 Pro domain (DED1-DED2) are shown in Fig 3D and 3E. Interestingly, limited expression of RIP3 [28] protein in HeLa cell gives rise to very low and therefore heterogeneous distribution of RIP1-RIP3 heterodimers among the cells (Fig 3F) making the spontaneous event of the necroptosis less probable to overtake the apoptotic course of the cell death.
Averaged over the population the Casp8 activation time course demonstrated high dependence on the dose of the DL as well as the clustering capacity (S3 Fig). Thus, even low doses of the DL with enhanced clustering property can activate Casp8. This result confirms the established success in the application of combinational therapeutics where the DL has been combined with the ligand specific cross-linking antibodies that enhance receptor clustering [49].
As expected, the overall variability in the Casp8 activation is a function of the treatment dose (Fig 3G). Despite the coefficient of variation being within the limits of low-variance (less than 1), the early Casp8 initiation dynamics can bring significant stochasticity into triggering the downstream death pathway. Interestingly, the enhanced receptor clustering did not reduce the variability in the individual HeLa cell Casp8 activation dynamics significantly. We observed only a minor decrease in the coefficient of variation over all tested conditions (Fig 3G).
Next we studied the downstream caspase cleavage cascade, the second deterministic modelling module (Fig 2E), which feedbacks to the DISC/RIPoptosome based Casp8 production and is potentially capable of boosting cell apoptotic capacity especially following treatment of low doses of DL [68]. As an input we used the population average of the stochastic traces (Fig 2D, S3 Fig) we simulated for the first module of the DISC/RIPoptosome based network initiation assuming DR clustering (Fig 2B). Thus, we merged two modules into one complete deterministic system (Fig 2C) which enabled us to adjust undetermined parameters and estimate parameter sensitivity, hence avoiding computationally expensive parameter scans of the full stochastic formalism (see Materials and methods, S1 File).
The first undetermined parameter is the rate constant of Casp3 ubiquitin dependent degradation (kcat). Ubiquitination of active Casp3, which is set by XIAP, will attract proteasomal complex leading to Casp3 degradation. However, application of proteasome inhibitors does not stabilise the pool of active Casp3 and consequently does not result in reduced Casp3 proteasomal degradation. Instead, Casp3 catalytic activity is absolutely required for its own proteasomal degradation [71,72]. Therefore, dynamics of Casp3 degradation triggered by XIAP will not match the general degradation dynamics triggered by ubiquitin ligases for other types of proteins and this specific rate constant needed to be identified. We estimated that kcat needs to be significantly higher (1.75 min-1) from the general (basal) ubiquitin-dependent degradation rate (0.04 min-1) [73] (Table E in S1 File). Again, low doses of the DL bring into play a switch-like sensitive response to the change in kcat value (Fig 4A). In this case the cell death delay can be initiated in a spontaneous fashion if the Casp3 degradation mechanism is perturbed.
Furthermore, the similar steep ultra-sensitive response can be also initiated by the mild fluctuations in the XIAP concentration. We found that slight deviations from the mean XIAP level, 63 nM, quantified earlier for HeLa [1] could speed up the cell death by more than 3-fold in the case of low DL doses (Fig 4B). This decrease could be very sudden through this switch-like type of response. Indeed, XIAP specific inhibitors such as Embelin, Mithramycin A are able to overcome the DL resistance in different cancer types [74,75].
Finally, with the fully identified parameter set we formulated the new semi-stochastic hybrid model of apoptotic pathway initiation in a single cell with the fixed partitioning of the whole network into discrete (Fig 2B) and continuous reactions (Fig 2E). The slow discrete reactions are the DISC/RIPoptosome assembly. The fast continuous reactions capture the caspase cleavage cascade.
The simulation results for a single cell response on the addition of low and high amounts of the DL are demonstrated in Fig 5. We observed a prolonged ramp effect for all variables of the network before the system switched to the rapid response. The ramp duration for the displayed example exceeded 10 hours after treatment with the low dose of the DL (Fig 5A). Whereas the high dose treatment stimulates the ramp for shorter times, around one hour for a shown example (Fig 5C). In a similar manner to the simulations with our entirely deterministic model, the delay for the switch in the single cell response is a function of DL dose (Fig 4).
However, for both high and low doses we also observed very high dynamic noise in the ramp (Fig 5A and 5C). This noise characterises the time course of dimeric Casp8 and active Casp3 accumulation. In experiments both proteins are very unstable and hardly detectable in the pre MOMP period of apoptotic initiation [19,71]. As we have shown earlier initial formation of new Casp8 dimer species can be limited by the vulnerability in molecular assembly of the DISC/RIPoptosome platform (Fig 3). Moreover, active Casp8 dimer is unstable due to high dissociation rate in the cytoplasm [19,20]. Indeed, Casp8 under physiological concentrations is found mainly in monomeric form [18,20,59] (Fig 5B and 5D). Therefore, this process prevents accumulation of the excess catalytically active pool of Casp8 for further downstream apoptotic signalling in the pre MOMP period.
Casp3, as the main Casp8-dependent effector caspase [60], follows the noise in the dynamic course of Casp8 dimer during the ramp. Besides, Casp3 is sacrificed in the pre MOMP period due to the excess amount of XIAP which effectively [1,76] blocks Casp3 activity by binding and subsequent ubiquitination which leads to Casp3 degradation.
To study how the ramp noise property in individual cells influences the cell death delay we have performed 600 independent simulations of the semi-stochastic model mimicking the overall cell culture response. These simulations were repeated for four different scenarios: low and high dose treatment scenarios with or without receptor clustering order. The coefficient of variation in Casp8 dependent FRET probe cleavage calculated over ramp period for each cell was considered as the measure of the noise strength. As earlier, the moment of the individual cell death was recorded once the rate of FRET probe cleavage exceeded the expected experimental threshold rate [2] (Fig 5E and 5H). For the individual cells treated with low dose the cell death delay varied from 1 to 10 hours if we integrated the receptor clustering order. Even higher variability was observed when the clustering was absent. In this case the cell death time could vary from 1 to 22 hours. Examples of FRET time traces for five individual cells are shown in Fig 5E. By visualising the relationship between the single cell death delay and dynamic ramp noise strength over a population, we found out that noise was an important determinant of the delay. For both clustering and non-clustering scenarios this relationship follows the same trend (Fig 5F and 5G). Moreover, this trend was independent of the treatment dose (Fig 5I and 5J). Furthermore, for all tested scenarios coefficient of variation higher than 0.5 strictly characterised early dying cells which commit apoptosis within the first 2 hours. Interestingly, receptor clustering enhanced ramp noise resulting in higher values of coefficient of variation (Fig 5F and 5I).
Fractional cell killing was observed in DR-targeted treatments especially when applied in low amounts [2]. As we have shown, the high dispersion of the death delays was the main reason for fractional cell killing. What we found more interesting is that dispersion of the delays could exhibit strong bimodality clearly distinguishing between the fraction of early and late dying cells. Clear bimodality was predicted by our model particularly for the low ligand dose upon receptor clustering order (Figs 6 and 5A). Taking this fact together with the ramp noise analysis (Fig 5F) we can conclude that the high noise in the ramp sensitises cells for early death which will take place within the first five hours at the latest. This fluctuation-enhanced sensitivity has been called ‘stochastic focusing’ and allows quicker system relaxation to the stationary state when the noise is high. The bimodality breaks when receptor clustering is interrupted (Fig 6C, see S1 File) and most of the cells would die only after 10 hours. On the population average cell dynamics receptor clustering provides slightly quicker Casp8 activation for the low dosage of the DL (Fig 6E). This may enable better coupling of this stochastic process with the continuous positive caspase feedback loop. Thus, stochastic focusing coupled with the positive feedback facilitates a more robust bimodal response without the need of multi-stability encoded in the system itself. Finally, the overall cell survival can be dramatically reduced by enhancing the receptor clustering mechanisms (Fig 6G).
The roles of multiprotein signalling platforms assembled upon DR stimulation have been broadly discussed in the context of the programmed cell death initiation [11,17,29,31] as well as proliferation and proinflammatory signalling [5,77] over the last decades. The effect of DR and apoptotic inhibitors targeting on the structure and function of these platforms were investigated in different experimental models. However, the mechanism through which these platforms give rise to distinct functions is still poorly understood. Particularly, the mechanism through which the heterogeneous apoptotic response to the DR targeted therapeutics is initiated and how it can explain fractional cell death remains unclear. Our study shows that the noise exclusively caused by the stochastic molecular assembly of the DISC/RIPoptosome platform is able to explain fractional cell killing at low receptor level engagement. Furthermore, this noise in conjunction with receptor clustering facilitates a more rapid apoptotic response.
Most of the variability in cell death delay raised upon DR stimulation originates from the pre-MOMP phase. Individually, none of the proteins involved in the apoptosis activation prior to MOMP can explain variation in cell death delays. Casp8 activation rate and consequently the rate of Casp8-dependent BID cleavage are the only determinants of the process [2,4,78]. Casp8 activation is entirely dependent on the assembly of the multiprotein signalling platform such as RIPoptosome. Though there have been a few models developed none have explicitly accounted for the stochastic nature of the signalling platform assembly [79]. Hence in this study, we developed a novel mathematical model of the stochastic assembly of the RIPoptosome in the single cell together with downstream effector-caspases cascade. Two of these processes are paired together in the pre-MOMP phase of apoptotic pathway initiation. By incorporating the absolute protein concentrations that we have measured in HeLa cells experimentally, and using kinetic parameters derived from the literature we have simulated the Casp8 activation dynamics in the single cell for various conditions: different DL doses, full and disrupted DR clustering propensity. Our modelling simulations have shown that the random and competitive multiprotein assembly of RIPoptosome allows prolonged and slow activation of Casp8 in a ramp-like fashion which is prone to high stochastic fluctuations. Such fluctuations in conjunction with downstream positive feedback loop of effector caspases after certain delay can lead to the spontaneous acceleration of Casp8 accumulation. Because of these fluctuations each cell behaves differently. We have found that the time the single cell will commit to apoptosis depends on the amount of intrinsic noise level in the initial ramp Casp8 activation. The higher ramp noise favours quicker cell death. By that we provide the evidence that the random assembly of RIPoptosome on its own, without any contribution of extrinsic noise in protein expression may explain the heterogeneous cell death response.
Our modelling predictions confirm that the receptor clustering process is critical in the extrinsic apoptotic response initiation [80]. Furthermore, a lower DL treatment dose will benefit the most from the enhanced clustering capacity over all. However, the significant fraction of the cell population will remain in the delayed apoptotic state. This new finding is clearly reflected in the bimodality of the distribution of death delays initiated by low DL dose where we demonstrated the clear split of the cell population into early and late responders (Fig 6A).
Despite the high affinity of XIAP to Casp3, their concentration balance in HeLa cell does not ensure robust Casp3 inhibition prior to MOMP [76]. Additionally, XIAP stimulated Casp3 ubiquitination that leads to Casp3 degradation is critical to keeping the downstream executioner caspases cascade shut till the MOMP is set. We have shown that for the fixed XIAP level in HeLa, Casp3 will play an important role in determination of cell death delay. Thus, suppression of the Casp3 ubiquitination/degradation rate at some point can trigger an ultra-sensitive switch from late to early cell death (Fig 4A). This response is characteristic for the low doses of DL and has been suggested in previous modelling studies [68]. However, experimentally Casp3 proteasomal degradation is hard to inhibit unless the catalytic activity of Casp3 is suppressed [71,72]. Instead, XIAP inhibition can initiate the same effect (Fig 4B) and as we showed very minor suppression is needed to return rapid cell death response initiated by subminimal DL doses. We believe that this ultra-sensitivity serves the best explanation for established success in the application of XIAP specific inhibitors for DL dependant cell death amplification [74,75,81,82]. Strikingly, we found that at the low DL doses an increase in XIAP level exclusively would cause a tremendous linear increase in the time of cell death delay. Indeed, exceptionally only XIAP overexpression, not cIAP1/2 or Smac up and down regulation respectively, is the apoptosis resistance mechanism which can be developed in cancer cells in response to the chemotherapeutics [83].
The content and dynamics of the RIPoptosome assembly predicted by model conform the general knowledge that RIP1 is the most abundant protein among all that are comprising the core RIPoptosome scaffold [10,23,24,54,84–86]. The engraftment of ProCasp8 molecules into RIP1 oligomer can happen when the RIP1 filament growth is interrupted by binding of single FADD molecule that occasionally can lead to the sequential binding of ProCasp8. Our simulations have showed that this event is very rare for a given level of RIPoptosome proteins in HeLa cell and we do not see strong oligomerization of ProCasp8 or its DEDs in HeLa cell. Despite, overexpressed truncated form of ProCasp8 which includes only DED1-DED2 domain is prone to form filamentous structure by oligomerisation [12,53,87,88], the full length protein do not oligomerize [87–89]. Overall, the quantitative balance between the components may dictate the structure of the RIPoptosome that vary between different cell types [12,84,85]. Therefore, we can conclude that the RIPoptosome formation in HeLa is a competitive process of RIP1, FADD and ProCasp8 assembly and the structure and function of this assembly varies due to the noisy nature of the core protein binding and dissociation events.
In this context, the slow and probabilistic nature of Casp8 activation explained in our current study by the random RIPoptosome assembly serves as the basis for caution mechanism of kinetic proofreading. This mechanism needs to be in place to verify weak or temporal apoptotic stimuli. The cells which succeeded to assemble the pool of RIPoptosomes that can sustain efficient Casp8 activation will proceed further down the apoptotic pathway triggering MOMP. The high noise in the ramp of Casp8 activation, in this case, will signify the high RIPoptosome efficiency showing that each moment the Casp8 activity is sacrificed the next moment it can be reconstituted or even amplified (Fig 7).
The vulnerability of the apoptotic pathway and its susceptibility to adaptation are currently the key limitation of therapeutics designed to kill cancer cells through the DR targeting therapeutics. In this paper, we have uncovered the original mechanism that explains inefficient cell death stimulation through stochastic activation of apoptosis initiating caspase signalling, leading to heterogeneous responses. We believe that detailed understanding of basic principles of early events of cell death initiation may also stimulate more rationalised approaches in the development of combinational treatments against cancer.
We quantified TNFR1 in HeLa cells by QuantiBRITE phycoerythrin beads based assay. The amount of DR4/5 was calculated from TNFR1 level by comparative MS data analysis (Table A in S1 File). Receptor clustering conformation was calculated from experimentally derived cluster size probability distributions (S1 Fig). Initial protein concentrations were taken from the literature (Table D in S1 File). Except FADD and RIP1, which we quantified with recombinant protein comparative Western Blot and ProCasp6 concertation that we adjusted using the complete deterministic model (S1 File). Most binding kinetics and catalytic enzymes activity parameters were retrieved from the literature (Table E in S1 File). Hence FRET probe cleavage rate and Casp3 degradation rate were adjusted in the simulations.
Modelling formalism of Gillespie stochastic simulation algorithm (SSA) and ODE integration as well as semi-stochastic hybrid model was implemented in the MATLAB 2017b environment (see also S1 File).
|
10.1371/journal.pcbi.1004894 | Nucleosome Presence at AML-1 Binding Sites Inversely Correlates with Ly49 Expression: Revelations from an Informatics Analysis of Nucleosomes and Immune Cell Transcription Factors | Beyond its role in genomic organization and compaction, the nucleosome is believed to participate in the regulation of gene transcription. Here, we report a computational method to evaluate the nucleosome sensitivity for a transcription factor over a given stretch of the genome. Sensitive factors are predicted to be those with binding sites preferentially contained within nucleosome boundaries and lacking 10 bp periodicity. Based on these criteria, the Acute Myeloid Leukemia-1a (AML-1a) transcription factor, a regulator of immune gene expression, was identified as potentially sensitive to nucleosomal regulation within the mouse Ly49 gene family. This result was confirmed in RMA, a cell line with natural expression of Ly49, using MNase-Seq to generate a nucleosome map of chromosome 6, where the Ly49 gene family is located. Analysis of this map revealed a specific depletion of nucleosomes at AML-1a binding sites in the expressed Ly49A when compared to the other, silent Ly49 genes. Our data suggest that nucleosome-based regulation contributes to the expression of Ly49 genes, and we propose that this method of predicting nucleosome sensitivity could aid in dissecting the regulatory role of nucleosomes in general.
| The nucleosome—a large protein complex with DNA wound around it—is the fundamental unit of genomic organization in the eukaryotic cell. More than just a DNA organizer, however, nucleosomes may control gene expression by interfering with the cell’s ability to access the wound-up DNA, as shown by recent research. In this report, we demonstrate a computational method for predicting which elements of the genome are sensitive to regulation by nucleosomes. As a proof-of-concept, we identify AML-1a binding sites—important sequences in DNA regulation—as being specifically nucleosome sensitive. We then show that AML-1a sites are specifically depleted of nucleosomes when a gene is expressed, indicating the ability for nucleosomes to suppress the expression of that gene. This finding confirms that nucleosomes are likely involved in genome regulation, and provides a method for predicting which areas of the genome are probably affected most by nucleosomes. This paper also highlights the usefulness of the Ly49 gene family in testing computer-derived genomic predictions, and is of interest to anyone studying how gene expression is regulated from cell to cell.
| The nucleosome, comprising an octameric protein core surrounded by ~147 bp of DNA, is the fundamental unit of genomic organization in the eukaryotic cell [1,2]. This organizational paradigm lays the groundwork for higher-order chromatin compacting, ultimately allowing the dozens of centimetres of DNA in a typical mammalian cell to be packaged into a micrometer-sized nucleus [3]. Beyond this impressive organizational role, however, the nucleosome—a bulky and ubiquitous protein structure tightly bound to DNA [4]—is believed to occupy a central role in the regulation of gene transcription [5,6]. The histone core of the nucleosome contains some of the most heavily targeted proteins for post-translational modification. These modifications, or histone marks, encode the epigenome: a highly plastic, inheritable set of expression instructions that are responsible for the incredible diversity of cellular morphology and function that can arise from a single genome [7–9].
The role of histone modifications in regulating gene expression has received considerable attention since the discovery of the epigenome, and considerably more attention is still required to understand this complex regulatory network. Alongside these histone modifications, however, nucleosomes are also believed to regulate gene expression simply by their steric effects on the accessibility of a given sequence of DNA [10]. Originally, nucleosomes were believed to be evenly spaced every ~200 bp of DNA, emphasising the structural role they play [11]. Now, however, it is known that, while many nucleosomes do obey this simple spacing model, many others are specifically positioned within the genome, either by the action of chromatin remodeling enzymes [12] or by the basal affinity of a given stretch of DNA for a nucleosome [13–15]. The identity and target sequences of all chromatin remodeling enzymes are not yet known [16]; however, work from our lab and others have identified DNA sequence patterns that allow the modeling of basal nucleosome affinity for a given genetic sequence, allowing for in silico mapping of the ‘default’ nucleosome landscape of the genome [11,17]—that is, the nucleosome landscape before it is changed by remodelling enzymes.
In more specific terms, nucleosomes are understood to be positioned either statistically or specifically [18]. Statistical positioning means the nucleosome position is limited only by the adjacent nucleosome; if there is no adjacent nucleosome, the statistically positioned nucleosome could reside on any sequence of DNA. Conversely, specific nucleosome positioning means the nucleosome position is closely regulated by the underlying DNA sequences [11,13,14,19] or by nucleosome remodeling factors [12,20,21]. Various eukaryotic nucleosome positioning sequence (NPS) patterns have been proposed. These NPS are characterized by repeating nucleotide sequences in yeast [11,14,22], fly [23], and C. elegans [24]. Specifically, the periodic appearance of dinucleotides, such as AA or TT, in every 10 bp is a commonly observed sequence pattern that allows the tight bending of DNA around the histone octamer core [25–27]. Many statistical methods incorporating various parameters, such as Support Vector Machine [13], DNA deformation energy [8], a thermodynamic model including interactions between adjacent nucleosomes [28], or the hidden Markov Model [29], were used to propose the NPS patterns which predict nucleosome positions from the underlying genomic sequences.
Our goal has been to study how changes to this ‘default’ sequence-determined nucleosome landscape correlate with gene expression. To that end, we have investigated the effects of nucleosome positioning on the expression of a family of immune genes, the Ly49 receptors. Ly49 receptors—and their human analogues, the killer-cell immunoglobulin-like receptors (KIR)—are expressed on natural killer (NK) and other immune cells. These receptors interact with the class-I major histocompatibility complex (MHC-I). This sensitivity to MHC-I expression levels allows NK cells to distinguish healthy cells from cancer or virus-infected cells, and is vital to the innate immunosurveillance performed by NK cells [30,31].
Aside from its importance in immunology, the Ly49 family has several traits that make it an ideal model for our investigation of transcriptional regulation by nucleosomes. The Ly49 genes all reside in a 500,000 bp region of chromosome 6, and have very similar transcription factor requirements. However, expression of an individual Ly49 gene is stochastic, such that each NK cell acquires a unique repertoire of Ly49 receptors during development and then maintains this repertoire throughout its life [32–34]. This therefore presents a model system in which nucleosome differences in expressed genes can be compared to their silent neighbours, which require the same transcription factors and would be equally impacted by large-scale events such as chromosome looping, the actions of locus-control regions, and any technical errors. Additionally, a common NK-T cell line—called RMA—naturally expresses only Ly49A and none of the other Ly49, providing a convenient model for this line of study. While interesting, the stochastic expression of Ly49 in primary NK cells provides a number of technical challenges for an analysis such as this, favouring RMA as our model for this study.
In mice, this stochastic expression is achieved in part by the coordinated activity of two or three promoters for each gene [35]. Immature NK cells use the first promoter (Pro1), which is a bidirectional promoter consisting of a conserved transcription factor binding platform: a 5’ and a 3’ TATA box each flanked by C/EBP binding sites, separated by a central AML-1 and NFκB binding site [35]. The forward and the reverse directions for Pro1 compete for the same transcription complex, which by chance may assemble and transcribe in only one of the two directions. Forward transcription is believed to dislodge an inhibitory complex around the downstream second and/or third promoter [35,36], which then drives expression of that Ly49 for the rest of the NK cell’s life. Conversely, reverse transcription means the inhibitory complex is present at a key developmental moment, forever barring the NK cell from expressing that Ly49 [35]. The relative strength of the forward and reverse promoters is believed to be modulated by the C/EBP binding sites. In highly expressed Ly49 receptors, their Pro1 has a ‘forward’ sequence with higher affinity for these transcription factors than their ‘reverse’ sequence, while receptors with low expression rates have a high-affinity ‘reverse’ and low-affinity ‘forward’.
While the above model nicely describes the observed Ly49 expression patterns on NK cells, a recent report has shown that Pro1 affinity does not always correlate with that Ly49’s expression level [37]. This finding suggests that some other, as-yet unknown factor may be at play in regulating Ly49 expression, giving rise to our interest in the effects of nucleosome positioning on Ly49 expression. While the transcription factor requirements for Ly49 expression have been described, we propose that certain necessary transcription factor binding sites within the Ly49 promoters or the enhancer are sensitive to the steric effects of nucleosome coverage. We expect that these sensitive binding sites will be preferentially enriched within predicted nucleosome-bound regions of DNA. Additionally, we and others have previously shown that some transcription factors are arranged ‘in-phase’ with nucleosomes, and so possess a noticeable pattern of 10 bp periodicity in nucleosome covered regions [38]. Since 10 bp corresponds to one turn of a DNA double helix [39], factors with this periodic pattern would be able to orient themselves to face ‘outward’ from the nucleosome even when covered and therefore be available for target protein binding [38]. Indeed, some TATA box binding sites have been shown to require this 10 bp phasing; disrupting the phasing of the binding site relative to local nucleosomes was shown to severely impact promoter function [40].
In this report, we identify AML-1a as a transcription factor with binding sites displaying both this preferential nucleosome coverage and lack of 10 bp periodicity. We then confirm, in a cell line that naturally expresses Ly49, that AML-1a sites are preferentially depleted of nucleosomes throughout the promoter/enhancer regions of expressed Ly49 genes when compared to the unexpressed genes within the same population, implicating nucleosome positioning as a possible mechanism to affect Ly49 expression in vivo.
Based on sequences generated previously in our lab, we predicted the basal nucleosome affinity of the C57BL/6 mouse Ly49 gene family. Using a hidden Markov model of nucleosome occupancy, we determined the probable nucleosome map for these regions (S1A Fig). These predictions show only a slight bias toward GC-rich regions (S1B Fig), and when tested against a published MNase-Seq dataset showing nucleosome occupancy in mouse hepatocytes, gave an overall accuracy of 48.8% (S1C Fig). That the accuracy of this prediction is so low is not surprising—the prediction only accounts for the genomic affinity for nucleosome binding, and cannot account for the action of chromatin remodelling factors.
Having generated this sequence-based map of the default genomic nucleosome affinity, we next asked whether nucleosome positions, and their interaction with transcription factor binding sites, might contribute to Ly49 gene transcriptional regulation. For each of the Ly49 genes in the C57BL/6 genome with promoter annotations, we determined whether a pattern emerged when examining whether the promoter was predominantly nucleosome-bound or nucleosome-free (Fig 1). We also performed this analysis for Ly49 families from Balb/c, 129Sv, and NOD mouse strains, with similar results (S2 Fig). In most cases, the sequence at Pro1 reverse favours a free configuration, while Pro1 forward favours a bound configuration. We also noticed an AML-1a binding site located between the two promoters in virtually every case, agreeing with previous results for the B6 Ly49G promoter region [35].
We next sought to determine whether certain TF binding sites within the Ly49 gene cluster are particularly sensitive to nucleosome coverage. First, since Ly49 receptors are thus far known only to be expressed in immune cells, and predominantly in natural killer lymphocytes, we selected 17 transcription factors with well-known functions in regulating lymphocyte gene expression (S3 Fig). Factors with known Ly49 interactions—including AML-1 and C/EBPβ [35]—were included, but to avoid bias, other common lymphocyte transcription factors were included. We also included TATA as a control, which is expected to be positioned away from nucleosome centers. For the whole Ly49 region, we plotted the distance from the center of each of these TF binding sites to the center of the predicted nucleosome binding site. Of the 18 factors, 10 gave sufficient signal for the Ly49 cluster to generate histograms plotting the spatial relationship between TF and nucleosome binding sites (Figs 2 and S4A). We observed three distinct patterns of TF-nucleosome spacing, based both on visual assessment and the calculated kurtosis—the mathematical measurement of a curve’s pointedness—of the histograms. First, some factors—namely, NF-AT and TATA—show a bimodal distribution and a negative kurtosis (< -0.5, indicating a flat curve) about the nucleosome center, with peaks existing at or near the nucleosome boundary. This suggests that these factor binding sites preferentially avoid nucleosomes, and are unlikely to be sensitive to nucleosome interference. Second, AML-1a, AP-1, Lyf-1, Sp-1, and MZF-1 show a tightened, monomodal distribution and positive kurtosis (> 0.5, indicating a pointed curve), with most factor binding sites being preferentially covered by the nucleosome. The rest of the factors analyzed were found to be normally distributed about the nucleosome dyad (kurtosis between -0.5 and 0.5), and so show no preference for nucleosome coverage.
Above, we identified five transcription factor binding sites which are predicted to be covered by nucleosomes in the Ly49 gene family; these transcription factors are the ones we hypothesize to be most sensitive to nucleosome interference. However, as mentioned previously, factors which display a strong 10 bp periodicity are able to exist in-phase with the nucleosome—and therefore be accessible even when covered—while those specifically lacking 10 bp periodicity will be disrupted by the turning of the DNA in the nucleosome (Fig 3A). To measure this effect in the Ly49 gene family, we analyzed the periodicity of each transcription factor, taking an unbiased approach using a leave-one-out analysis. In this analysis, the periodicity of all 17 factors pooled together is determined by Fourier transformation, and then each factor individually is removed from the other 16 to measure the effect its removal has on the overall periodicity. Factors displaying 10 bp periodicity will reduce the overall 10 bp periodicity when removed from the period histogram, while factors without 10 bp periodicity will increase the overall 10 bp periodicity when removed (Figs 3B and 3C and S4B). Based on these results, AML-1a is the only transcription factor that specifically lacks 10 bp periodicity within the Ly49 gene family, as its removal from the pooled factors dramatically increases the signal at 10 bp. Taken together, these results indicate that, of the factors analyzed, AML-1a is the one most sensitive to the surrounding nucleosome environment, as it is the only factor that displays both a tendency to nucleosome coverage (Fig 2) and a specific lack of 10 bp periodicity (Fig 3).
To determine whether AML-1a’s pattern of nucleosome co-occupancy and a lack of 10 bp periodicity is unique to the Ly49 family, we extended these two analyses to the entirety of mouse chromosome 6 (S5 and S6 Figs). To our surprise, while AML-1a retained its nucleosome preference and displayed a very marginal lack of 10 bp periodicity, many of the other transcription factors analyzed displayed much more pronounced coverage (S5 Fig) and an extreme loss of 10 bp periodicity (S6 Fig). This suggests that what we report with regard to AML-1a in Ly49 is one example of a more general paradigm of gene regulation via nucleosomal interference with one or several key transcription factor binding sites.
The Ly49 gene family presents a powerful and convenient tool to test whether these patterns can be detected in an experimental setting. It is a large gene family, whose members all exist in the same region of chromosome 6, and which have closely related promoter sequences and similar transcriptional regulation in terms of transcription factor binding sites. Despite these similarities, individual Ly49 genes can display remarkably different expression patterns. Indeed, RMA, a mouse NK-T cell line with a C57BL/6-derived Ly49 gene cluster, expresses the inhibitory Ly49A receptor, but no other Ly49 for which there is a detecting antibody (Fig 4A). We used this model to determine whether the promoter for Ly49A will have a nucleosome landscape more divergent from the sequence-based predictions than any of the other, inactive Ly49 promoters. We generated the nucleosome sequence map for RMA using MNase-Seq (Fig 4B), and generated a list of nucleosomes that were predicted to be present and confirmed by the MNase-Seq map (‘true positives’) and a list of nucleosomes predicted to be present but found absent in the map (‘false positives’) using the UCSC table browser. Using these lists of true and false positive nucleosome predictions, we generated a heatmap showing the specific accuracy of our predictions for each Ly49 promoter region, across the whole region of interest (‘Whole’) or at all of the sites for each indicated transcription factor (Fig 4C). In this analysis, the promoter region selected includes the areas corresponding to promoters 1, 2, and 3, as well as approximately 8000 bp upstream of Pro1 to capture any distal enhancer elements. This map was coloured to show those Ly49 genes and transcription factor binding sites with a relatively high degree of convergence (blue) or divergence (red) with the prediction.
Next, we asked whether specific TF binding sites, for each promoter individually, diverge significantly in nucleosome occupancy, compared to the rest of that promoter. Taking the overall predictive convergence for each Ly49 promoter (the ‘Whole’ column in Fig 4C) as the expected value, we performed a chi-square analysis on the divergence of the TF binding sites’ nucleosome status from this expected result. Only two Ly49 promoters were significantly divergent in nucleosome occupancy at the tested TF binding sites: the pseudogene Ly49K, and the expressed Ly49A (Fig 4D). All other Ly49 genes are not expressed in RMA, and were not found to significantly differ at TF binding sites from the rest of their nucleosome landscape. As indicated by the heatmap, AML-1a, c-ETS-1, Lyf-1, and MZF-1 were the factors most responsible for the significant divergence in Ly49A (Fig 4D).
With the exception of c-ETS-1, these divergent factors are all preferentially covered by nucleosomes (Fig 2), and so exist in close proximity to each other. It is therefore possible that nucleosome divergence at one factor was artificially driving divergence at the others (Fig 5A). To determine whether this could be the case, we performed logistic regression analysis on each pair among these four factors, using nucleosome state (true or false positive) of the predictor to predict the nucleosome state of the nearest binding site for the reporter variable, after controlling for distance between the two sites (specifically, whether the two sites were within 147 bp of each other or not). Knowledge of the state of an AML-1a site was not found to be any better or worse at predicting the state of the nearest c-ETS-1 site than knowledge of a c-ETS-1 site could predict the nearest AML-1a, suggesting that neither of these sites impact the other unequally (Fig 5B and 5C). However, knowledge of the state of an AML-1a site was a much stronger predictor of the state of the nearest MZF-1 or Lyf-1 site than either of these sites was for predicting AML-1a, suggesting that AML-1a-based divergence may be preferentially influencing the Lyf-1- and MZF-1-based divergence (Fig 5B and 5C). Significantly, c-ETS-1 was not one of the factors enriched in nucleosome-bound regions, while AML-1a, Lyf-1, and MZF-1 were preferentially covered factors.
While AML-1 is well-known as a required factor for Ly49 expression in mouse NK cells [35], we performed a chromatin immunoprecipitation to validate the presence of AML-1 at the Ly49A promoter in RMA cells (Fig 5D). Precipitation with anti-AML-1 resulted in a 10-fold enrichment of the Ly49A promoter versus precipitation with an isotype, indicating the presence of AML-1 at this site. Conversely, a genomic region 6 kbp upstream of the Ly49A promoter was not enriched following anti-AML-1 enrichment.
Finally, we analyzed the regions of each Ly49 gene immediately upstream of exon 1 (corresponding to Pro2) and upstream of exon 2 (corresponding to Pro3). As these are much smaller genetic elements with only a few nucleosomes and AML-1a binding sites, statistical analysis similar to the one performed above was not possible. However, for each of Pro2 and Pro3, we identified each AML-1a binding site and whether it was preferentially covered or not, again determining the degree of convergence at these sites (Table 1). Surprisingly, Ly49A displays a higher degree of convergence at AML-1a sites in Pro2 and Pro3 than many of the other, non-expressed Ly49 genes, indicating that the nucleosome divergence for AML-1a in expressed genes is restricted to Pro1 and the enhancer elements.
In this report, we have shown that nucleosome positioning may be involved in regulating Ly49 expression, not just by condensing the gene complex, but by specifically interfering with AML-1a transcription factor binding sites. This conclusion is based on two complementary studies. In silico, we have shown that, within the Ly49 gene cluster, AML-1a is predicted to be exquisitely sensitive to steric hindrance by nucleosome occupancy. The AML-1a binding sites tend to be located within a nucleosome boundary, meaning that AML-1a cannot simply avoid nucleosomes. Moreover, unlike other covered TF binding sites, AML-1a binding sites lack 10 bp periodicity, meaning it must exist ‘out-of-phase’ with the nucleosomes and so cannot adopt a conformation where many AML-1a sites are exposed on the surface of the nucleosome. These predictions were confirmed in a cell line, where we have shown that expressed Ly49 genes show preferential nucleosome depletion specifically at AML-1a binding sites compared to the inactive Ly49 genes. Furthermore, this depletion at AML-1a sites is not likely to be the result of depletion at some confounding TF binding site, since knowledge of the nucleosome depletion of AML-1a was as good or better at predicting the depletion of all other highly depleted TF binding sites than those sites were at predicting AML-1a. Taken together, these results suggest that Ly49 expression is partially regulated by nucleosome occupancy at AML-1a binding sites.
We chose the Ly49 gene family for this analysis for both immunological and bioinformational reasons. Immunologically, Ly49 gene expression on NK cells is an incredibly complex phenomenon, with its mostly stochastic gene expression giving rise to myriad functionally distinct subpopulations of NK cells. This heterogeneity allows a population of NK cells to have members able to respond to virtually any alteration to homeostasis, giving rise to phenomena like missing-self rejection of foreign or diseased cells [41]. As this complex expression of Ly49 receptors is central to NK function, understanding Ly49 transcriptional regulation is central to understanding the NK cell itself, and while studies have focused on cis-elements and transcription factors involved in Ly49 regulation, there has been little attention to date given to the role of nucleosome positioning.
From a bioinformatics viewpoint, the Ly49 family presents an attractive subject of study thanks to its genetic isolation, homology, and independent expression. This provides a model system where comparisons can be made between related expressed and silent genes, using one group to control for the other, while any large-scale genetic factors, such as the action of locus control regions or chromosome looping, are likely to impact the entire family equally, regardless of expression state. Additionally, we have taken advantage of this fact to compare expressed and silent Ly49 genes within the same dataset, effectively mitigating the impact of technical errors on our results. We suggest that other predictive informatics endeavours would benefit from performing their analyses on the Ly49 (or KIR, in human datasets) gene family.
Unfortunately, this complex expression of Ly49 does also introduce hindrances into the present study. Not only are the Ly49 genes expressed randomly in a pool of NK cells, but most of them—the inhibitory receptors—display a mono-allelic expression pattern [42,43]. Therefore, any analysis of Ly49 genes from primary NK cells first requires enrichment for cells expressing the Ly49 in question, and then must contend with a sample in which only half of the genomic material includes an Ly49-expressing component. For this reason, we are fortunate that the RMA cell line naturally expresses Ly49A on 100% of the population, as this presents a convenient model for testing our predictions in a natural setting. Future work may focus on performing this analysis across a heterogeneous pool of natural killer cells ex vivo, which among other things would allow for the study of the differences between inhibitory and activating Ly49 receptors. Unfortunately, such a complex study goes beyond the scope of the present report.
That our analysis identifies and confirms AML-1a binding sites—specifically within the Pro1/enhancer region—as the most significant in terms of nucleosomal regulation is intriguing, thanks to AML-1’s known role in Ly49 transcriptional control. One of the features of the Pro1 region in many Ly49 genes is a central AML-1a binding site, identified as being absolutely required for Pro1 to function in either direction [35]. Thus, disruption of this AML-1a binding site represents a potent mechanism of Ly49 expression suppression, and may explain why nucleosomes appear to target AML-1a binding sites within unexpressed Ly49 genes. Interestingly, the requirement of the AML-1a site for both forward and reverse transcription at this site suggests that nucleosome regulation is unlikely to directly impact the stochastic expression of an Ly49 gene, but rather interferes with any gene activity at all, either forward or reverse. Whether this loss of nucleosome occupancy in expressed Ly49 genes is a cause or an effect of Ly49 gene transcription remains to be determined. That the nucleosome regulation is found within the Pro1/enhancer region may also indicate that nucleosomes best exert their regulatory effects during cell development and maturation, as Pro1 is active only in developing NK cells.
Finally, it will be of interest to study the nucleosome sensitivity of AML-1a binding sites at other, non-Ly49 genes, to determine whether AML-1a is ubiquitously sensitive to nucleosome occupancy, or whether this sensitivity is a selected trait for Ly49 regulation only. Our whole-chromosome analysis of nucleosome-transcription factor interactions revealed the surprising result that, on this scale, many other transcription factors display the nucleosome coverage and lack of 10 bp periodicity that identified AML-1a as a sensitive factor in the Ly49 cluster. This suggests that, on a genome-wide scale, many different transcription factors have the potential to be nucleosome-sensitive. However, within a given gene or gene family, only a small number of factors—such as AML-1a in Ly49—retain this sensitivity, and may act as nucleosome ‘linchpin’ factors, while the other factors lose their sensitivity either by avoiding nucleosomes or by getting in-phase with them. A genome-wide characterization and identification of these linchpin factors and their associated genes could provide a novel, informative method of functionally grouping genes.
It is our hope that this analysis of TF distribution and periodicity will assist in achieving a greater understanding of how chromatin compaction and nucleosome positioning shapes the gene expression landscape.
The genomic DNA sequence of the Ly49 gene cluster from C57BL/6 mice was previously available [44]; sequences for 129, BALB/c, and NOD mice were generated previously by A.P.M. [45–47]. These genomic DNA sequences were used to predict the nucleosome affinity and the probability of nucleosome binding by NuPoP [48]. The prediction parameter was set as the 4th order-Hidden Markov Model, and the species was set as mouse. NuPoP produces nucleosome occupancy score, affinity score, and Viterbi prediction of the position. The nucleosome dyad was set as the middle of the nucleosome position determined by the Viterbi prediction.
The same four genomic DNA sequences of the Ly49 gene clusters of the four mouse strains were used to predict transcription factor binding sites. The Position Weight Matrix (PWM) of the 18 transcription factors (AML-1a, AP-1, C/EBPβ, Egr-1, Egr-2, Egr-3, GATA-3, IRF-1, Ik-3, Lyf-1, MZF1, NF-AT, NF-κB, Oct-1, STAT3, Sp1, Tal-1α/E47, c-Ets-1(p54)) were retrieved from the TRANSFAC or JASPAR public databases [49,50]. Detection of predicted TF binding sites was performed using either the MATCH or FIMO algorithm [51,52], with cut-off selection set to minimize false positives. The TATA binding site position was also predicted as a control.
The promoter architecture was presented around the Pro 1 region including Pro1 and exon -1a. The landscape of the nucleosome binding sites and the TF binding sites in the promoters were examined to find relationships. Nucleosome predictions were aligned to exon -1a for each Ly49 gene in the C57BL/6 mouse. For genes without an exon -1a annotation, the average distance between the Pro1 and exon -1a for each annotated gene was used to approximate the location of exon -1a. If a transcription factor binding site was overlapped with a nucleosome position, then the transcription factor binding site was marked as closed. Otherwise, it was marked as open.
The distribution of transcription factor binding sites around the proximal nucleosome was explored for each transcription factor. First, the output files of the transcription factor binding sites and the Viterbi prediction of the nucleosome positions were converted to BED format. Then for each transcription factor binding site, the proximal nucleosome was identified by the BEDtools suite [53] using the CLOSEST command. The distance was calculated between the middle position of the transcription factor binding site and the dyad position of the proximal nucleosome. The distances from all pairs of the transcription factor binding site and the proximal nucleosome were counted to generate the histogram. The density of the distribution was estimated using Kernel Density Estimation by R software. The distribution was centered at the nucleosome dyad, with boundaries marked at 73 bp away from the dyad on either side.
For all transcription factor binding sites located within the predicted nucleosome-bound regions, we examined the periodicity of the distances between the transcription factor binding sites and the proximal nucleosome by Fourier transform. For each nucleosome-bound region of the Ly49 cluster, the distance between each transcription factor and its proximal nucleosome center was selected and pooled. The spectral density of the pooled distance counts from the 17 transcription factors was generated as a baseline diagram. Then, the spectral density of the distance counts from leave-one-out samples, which are the counts from 16 transcription factors by excluding one transcription factor at a time, were computed. The periodicities from the leave-one-out samples were compared with the baseline periodicities. Deviation from the baseline indicated that the left-out factor contributed to the periodicity of the baseline—with the factor gone, a reduction in periodicity at the 10 bp peak indicates that the left-out factor contributed to the overall 10 bp periodicity.
The predicted binding positions were compared with the experimental positions by MNase-seq results obtained on the mouse NK-T cell line, RMA, as previously described [54], with the following changes: library preparation was performed by Génome Quebec at McGill University, and samples were prepared and analyzed on an Illumina MiSeq (Illumina, San Diego, CA). MNase-seq results have been deposited in the GEO database, under accession GSE71863. The intersections of predictions with nucleosome-bound-regions (true predictions) or nucleosome-free regions (false predictions) were found using the UCSC table browser. The same approach was repeatedly used for each transcription factor binding site to identify the number of sites correctly or incorrectly predicted to be covered by a nucleosome. The degree of accuracy of the prediction was calculated for nucleosomes in general and for each TF binding site, performed for each Ly49 gene. The results are presented as a heatmap between the nucleosome predictions and the transcription factors.
A chi-square analysis was performed to detect specific transcription factors enriched for nucleosome deviancy compared to the rest of the gene. Based on each gene’s overall true-positive rate, expected true-positive counts were generated for each Ly49 region of interest. These expected values were compared to the observed using the chi-square test. To highlight individual TF contributions to the region of interest’s overall chi-square statistic, individual chi-square statistics were presented in a heat-map.
Odds ratios were estimated by logistic regression for each of the indicated TFs and AML-1a. Regression was performed using R by comparing the state (true or false positive coverage) at each predictor site to the nearest reporter site in the Ly49 cluster, after accounting for whether the two sites were within 147bp or not. Higher estimated odds ratios indicate that knowledge of the state of the predictor site is more able to predict the state of the reporter site.
RMA cells or isolated mouse splenocytes were analyzed by flow cytometry after staining with antibodies against NK1.1, TCRβ, and the Ly49 receptors such as Ly49A/D (clone 4E5), Ly49D, Ly49C/I (5E6), Ly49E/F, Ly49G (4D11), and Ly49H (3D10). Antibodies were purchased from eBioscience (San Diego, CA) or Becton Dickinson (San Jose, CA), and samples were acquired using a Beckman Coulter CyAN-ADP and analyzed using FlowJo (FlowJo LLC, Ashland, OR).
Fragmented chromatin from RMA cells was prepared by MNase digest as above, except that 150 bp fragments were not selected. Chromatin was then incubated with a rabbit polyclonal antibody raised against AML-1 or against immunoglobulin M (as an isotype control) overnight (Abcam). Antibody complexes were collected using protein A agarose beads, and DNA was purified using a high pH chelating solution, as previously described [55,56]. The following primers were used:
Ly49A promoter: AGGCCAGGGAAACCTGGTGTA
AAGAGGTGGGGCACTGGACTG
Ly49A 6 kb up: ACAGAACTCAGAGGGCAAAGGAAA
TGGGCCACTTGGCCATTTATCT
Real-time PCR was performed using an Eppendorf Realplex2 Mastercycler thermal cycler.
|
10.1371/journal.pgen.1005385 | MoSET1 (Histone H3K4 Methyltransferase in Magnaporthe oryzae) Regulates Global Gene Expression during Infection-Related Morphogenesis | Here we report the genetic analyses of histone lysine methyltransferase (KMT) genes in the phytopathogenic fungus Magnaporthe oryzae. Eight putative M. oryzae KMT genes were targeted for gene disruption by homologous recombination. Phenotypic assays revealed that the eight KMTs were involved in various infection processes at varying degrees. Moset1 disruptants (Δmoset1) impaired in histone H3 lysine 4 methylation (H3K4me) showed the most severe defects in infection-related morphogenesis, including conidiation and appressorium formation. Consequently, Δmoset1 lost pathogenicity on wheat host plants, thus indicating that H3K4me is an important epigenetic mark for infection-related gene expression in M. oryzae. Interestingly, appressorium formation was greatly restored in the Δmoset1 mutants by exogenous addition of cAMP or of the cutin monomer, 16-hydroxypalmitic acid. The Δmoset1 mutants were still infectious on the super-susceptible barley cultivar Nigrate. These results suggested that MoSET1 plays roles in various aspects of infection, including signal perception and overcoming host-specific resistance. However, since Δmoset1 was also impaired in vegetative growth, the impact of MoSET1 on gene regulation was not infection specific. ChIP-seq analysis of H3K4 di- and tri-methylation (H3K4me2/me3) and MoSET1 protein during infection-related morphogenesis, together with RNA-seq analysis of the Δmoset1 mutant, led to the following conclusions: 1) Approximately 5% of M. oryzae genes showed significant changes in H3K4-me2 or -me3 abundance during infection-related morphogenesis. 2) In general, H3K4-me2 and -me3 abundance was positively associated with active transcription. 3) Lack of MoSET1 methyltransferase, however, resulted in up-regulation of a significant portion of the M. oryzae genes in the vegetative mycelia (1,491 genes), and during infection-related morphogenesis (1,385 genes), indicating that MoSET1 has a role in gene repression either directly or more likely indirectly. 4) Among the 4,077 differentially expressed genes (DEGs) between mycelia and germination tubes, 1,201 and 882 genes were up- and down-regulated, respectively, in a Moset1-dependent manner. 5) The Moset1-dependent DEGs were enriched in several gene categories such as signal transduction, transport, RNA processing, and translation.
| This paper provides two major contributions to the field of genetics. First, we systematically studied the biological roles of eight histone lysine methyltransferase (KMT) genes in the phytopathogenic fungus Magnaporthe oryzae. We investigated their roles, especially focusing on their involvement in infection-related morphogenesis and pathogenicity. The results showed that the eight KMTs were involved in various infection processes to varying degrees, and that MoSET1, one of the KMTs catalyzing methylation at histone H3 lysine 4 (H3K4), had the largest impact on the pathogenicity of the fungus. Second, we focused on the role of MoSET1 in global gene regulation. H3K4 methylation is generally believed to be an epigenetic mark for gene activation in higher eukaryotes. However, in Saccharomyces cerevisiae, SET1 was originally characterized as being required for transcriptional silencing of silent mating-type loci. We addressed this apparent discrepancy by examining genome-wide gene expression and H3K4 methylation during infection-related morphogenesis in M. oryzae. RNA-seq analysis of a MoSET1 deletion mutant revealed that MoSET1 was indeed required for proper gene activation and repression. ChIP-seq analyses of H3K4 methylation and MoSET1 suggested that MoSET1 could directly play a role in gene activation while MoSET1-dependent gene repression may be caused by indirect effects.
| In eukaryotic cells, DNA-dependent processes can be regulated by covalent modifications of histones such as methylation, acetylation, phosphorylation, sumoylation, and ubiquitination [1]. Long amino-terminal tails of histones protruding from nucleosome cores are especially subject to post-translational modifications. The combination of histone modifications to regulate cellular processes is a dynamic language, and is referred to as the histone code [1]. Histone modifications serve as marks for the recruitment of various chromatin proteins or protein complexes to modulate diverse chromatin functions including gene expression, silencing, repair, and replication [2]. Numerous “writing” enzymes (methylases, acetylases etc) and “erasing” enzymes (demethylases, deacetylases etc) are involved in the histone code.
Histone methyltransferases are a group of enzymes catalyzing the transfer of methyl groups from S-adenosyl methionine to histones. They can be divided into two groups based on their target amino acid residues: protein arginine methyltransferases (RMTs) and histone lysine methyltransferases (KMTs) [3–5]. A nomenclature system for the KMT family has recently been proposed, in which KMTs are classified into eight major subclasses, KMT1 to KMT8, based on their phylogenetic relationships and domain structure/organization [6]. For example, KMT1 proteins, exemplified by Drosophila melanogaster Su(Var)3-9, Schizosaccharomyces pombe Clr4, and Neurospora crassa DIM-5, specifically methylate H3K9, which leads to gene silencing and heterochromatin formation [7–9]. The KMT2 proteins, typified by Saccharomyces cerevisiae SET1 and D. melanogaster Trithorax, specifically catalyze methylation at H3K4, a mark for gene activation [10, 11]. Since all KMTs except the KMT4 class contain a SET domain, named after three Drosophila lysine methyltransferases: Su(var)3-9, Enhancer of zeste, and Trithorax, they are also often referred to as SET proteins. It is to be noted that there are also known possible KMT proteins that are not included in the nomenclature system such as SET3 and SET4 in S. cerevisiae.
KMTs are conserved in a wide range of eukaryotes, playing roles in cellular signaling pathways related to the cell cycle, cell motility, transcription, apoptosis, and cancer [12, 13]. In filamentous fungi, KMT-related gene regulation has been investigated mainly with regard to gene silencing and secondary metabolite (SM) production [9, 14–21]. In N. crassa, H3K9me3 catalyzed by DIM-5 belonging to the KMT1 class directs DNA methylation and heterochromatin formation by recruiting a protein complex containing heterochromatin protein-1 (HP1) and DIM-2 DNA methyltransferase through interaction of the chromo shadow domain of HP1 and PXVXL-like motifs in DIM-2 [9]. In Fusarium graminearum, H3K27me3 catalyzed by KMT6 was required for normal fungal development and contributed to regulating the “cryptic genome” including SM gene clusters [18]. Gene repression by H3K9 and H3K27 methylation was also recently shown to be involved in fungal symbiosis and pathogenicity through production of SM and effectors [19, 20].
In Aspergillus nidulans, H3K4me2 and H3K4me3, marks for gene activation play a role in chromatin-level regulation of SM gene clusters [21]. A loss-of-function mutation of the CclA gene, a member of the H3K4 methylating COMPASS (Complex Proteins Associated with Set1), resulted in a reduction of H3K4me2 and H3K4me3 at the SM gene clusters [22]. Surprisingly, cryptic SM gene clusters are activated in the ΔcclA mutant despite H3K4me2 and H3K4me3 being considered marks for gene activation [22]. While it is generally believed that H3K4 di- and tri-methylation are epigenetic marks for gene activation in higher eukaryotes, involvement of H3K4 methylation in gene repression is also reported in fungi and other organisms [21–25]. To date, it is not clearly known to what extent genes are up- or down-regulated in a H3K4 methylation-dependent manner, and what is the underlying mechanism for this apparent discrepancy.
Rice blast caused by Magnaporthe oryzae (Pyricularia oryzae) is one of the most devastating worldwide rice (Oryza spp.) diseases. This fungal species consists of several host-specific pathotypes that cause blast disease on a wide range of gramineous hosts including wheat, oat, finger millet and etc. Owing to their economic importance and genetic tractability, rice and M. oryzae have emerged as a model system for studying fungi-plant interactions [26]. M. oryzae displays dramatic morphological changes during infection [27]. When a fungal spore lands on a plant’s surface it germinates and forms a melanized dome-shaped infection structure, called an appressorium, at the tip of the germ tube. The appressorium generates enormous turgor pressure and physical force to breach the host cuticle, and the fungus eventually develops invasive hyphae to colonize host cells. These morphological changes are accompanied with global transcriptional alterations [28–30]. However, the involvement of genome-wide histone modifications in infection-related transcriptional alterations is poorly understood in M. oryzae. We recently demonstrated that MoSET1 catalyzing H3K4 methylation was required for substrate-induced transcriptional activation of the MoCel7C cellulase gene in M. oryzae [31]. Here we report a reverse genetics study of the KMT gene family in the M. oryzae genome, and examine their roles in global gene regulation related to the formation of infection structures in M. oryzae, especially focusing on that of MoSET1.
Eight putative KMT genes were identified in the M. oryzae genome based on sequence similarity and domain structure of known KMTs in the KEGG database (http://www.genome.ad.jp/kegg). Moset1 (MGG_15053) belonging to the KMT2 family, was previously named after S. cerevisiae Set1 [31]. The other seven genes were designated in this study as Mokmt1 (MGG_06852), Mokmt3 (MGG_01661), Mokmt4 (MGG_05254), Mokmt5 (MGG_07393), Mokmt6 (MGG_00152), Mokmt2h (MGG_02937), and Moset6 (MGG_15522) (Table 1). Mokmt1, Mokmt3, Mokmt4, Mokmt5, and Mokmt6 likely belong to corresponding KMT families [6]. MoKMT2H and MoSET6 showed amino acid sequence similarity with N. crassa SET-3 and Schizosaccharomyces pombe SET6, respectively.
To determine biological functions of M. oryzae KMT genes, deletion mutants were constructed using the split-marker recombination method [32] (S1–S7 Figs). Ectopic transformants, which had an insertion of a disruption construct somewhere in the genome other than the target locus, and gene complementation strains, which were deletion mutant-derived strains complemented by random insertion of a plasmid carrying the corresponding wild-type locus, were also created. These strains were used in further studies alongside the deletion mutants (S1–S7 Figs and S1 Table). Deletion mutants and a complementation strain of Moset1 were previously made and used in this study [31].
Histone lysine methylation levels in the KMT deletion mutants were assessed using western blotting with specific antibodies (Fig 1). In the Δmoset1 mutant, the levels of H3K4me2 and H3K4me3 were strongly reduced, while H3K4me1 moderately decreased. Abundance of signals for H3K9me3, H3K27me3 and H4K20me3 was also significantly reduced in theΔmokmt1, Δmokmt6, and Δmokmt5 mutants, respectively. These target sites for Magnaporthe KMTs were consistent with known target sites for corresponding KMT family proteins in other organisms. Western blotting was also used to test antibodies against H3K14me2 (active motif #39350), H3K36me3 (active motif #61102), and H3K79me2 (active motif #39144), however, no specific signal reduction in any KMT deletion mutant was detected (S8 Fig). H3K36 and H3K79 methylation are known to be catalyzed by Set2 (KMT3 family) and Dot1 (KMT4 family), respectively, in S. cerevisiae. Other KMT proteins might be involved in these histone marks in M. oryzae, or the specificity of the antibodies might not be strict enough to distinguish between marked and non-marked histones in M. oryzae.
The decreased levels of histone methylation in the deletion mutants were completely recovered in gene complementation strains (S9 Fig). These results indicated that MoKMT1, MoSET1, MoKMT5, and MoKMT6 catalyze methylation of H3K9, H3K4, H4K20, and H3K27 respectively, in M. oryzae.
The rates of conidiation, germination, and appressorium formation were assessed for phenotypical characterization of the KMT mutants with regards to infection. In addition, their growth rates were examined on rich media. The assay used two deletion mutants and one ectopic transformant for each KMT gene, with a complement strain employed when phenotypic defects were observed. The growth rates of the KMT-mutants were generally lower than that of the wild-type strain (Fig 2A). Especially, Δmoset1 exhibited the most severe reduction in vegetative growth, conidiation and appressorium formation but not in germination (Fig 2B–2D). TheΔmokmt3, andΔmokmt2h mutants showed moderate defects in all phenotypic traits investigated in Fig 2. TheΔmokmt6 mutants also showed severe reduction in conidiation and slight defects in appressorium formation (Fig 2B and 2D). Compared with the wild-type strain, the rates of conidiation and appressorium formation were reduced to less than 10% in the Δmoset1 mutants, and to 20–50% in theΔmokmt3 andΔmokmt2h mutants (Fig 2B–2D). The rate of germination was decreased by 40–50% in theΔmokmt3 andΔmokmt2h mutants. Interestingly, while the Δmoset1 mutants germinated at levels comparable to the wild-type, the conidia of Δmoset1 mutants often appeared to be malformed. The wild-type strain produced three-celled, tear-drop-shaped spores. Conidia of the Δmoset1 mutants were also three-celled, but were more elongated than the wild-type spores. All phenotypic defects observed in theΔmoset1, Δmokmt3, Δmokmt6, andΔmokmt2h mutants recovered to wild-type levels in the corresponding complement strains with an exception (conidiation inΔmokmt6), indicating that the KMT mutant phenotypes were caused by the corresponding KMT genes.
Infection assays of the KMT mutants were performed using three wheat and two barley cultivars with different levels of resistance/susceptibility to the wild-type wheat-infecting M. oryzae strain (Br48) used in this study. The order of susceptibility of the cultivars to Br48 was as follows: barley, Nigrate (super susceptible) > barley, Russian No. 74 ≈ wheat, Norin 4 (susceptible) > wheat, Chinese spring (moderate susceptible) > wheat, Thatcher (moderate resistant) [33].
Consistent with the rates of appressorium formation, pathogenicity to the wheat and barley cultivars was most severely impaired in the Δmoset1 mutants (Fig 3 and Table 2). The Δmoset1 mutants produced no visible symptoms on most host plants tested. Interestingly, the Δmoset1 mutants caused disease, albeit with fewer lesions, on the super susceptible barley cultivar Nigrate (S10 Fig), indicating that mutants did not completely lose their ability to infect plants. That fewer lesions were produced by the Δmoset1 mutants could largely be attributed to the low rates of appressorium formation.
Δmokmt1, Δmokmt3, Δmokmt6 andΔmokmt2h mutants showed significant reduction in pathogenicity on all tested plant cultivars except Nigrate (Table 2). Δmokmt2h mutants failed to cause compatible lesions on any plants other than Nigrate, indicating that mutants became non-pathogenic with certain host plants susceptible to the parent strain Br48. Δmokmt3 mutants also become non-pathogenic with the wheat cultivar Chinese spring, which is moderately susceptible to Br48. The other strains, including theΔmokmt4, Δmokmt5 andΔmoset6 mutants, ectopic transformants, and complement strains were infectious to all plant cultivars at levels comparable to the wild-type strain.
To further examine infection types of the KMT mutants, cytological analysis of inoculated leaves of the wheat cultivar Norin 4 was performed. At least 100 spores with an appressorium were assessed, and cytological interactions classified into four types: A, B, C, and D [34]. In Type A, no reaction of the host cells was observed. In Type B, inhibition of fungal growth was associated with papilla, a cell wall apposition at the penetration site. Type C represents the hypersensitive reaction of epidermal cells. Types A to C are resistance responses of host cells. Type D describes a susceptible response where infection hyphae were observed in infected cells.
In leaves infected with the wild type strain, the incidence of susceptible response Type D was predominant (63.9%) (Table 3). Similarly, Type D was predominant in leaves infected with theΔmokmt4, Δmokmt5, andΔmoset6 mutants at levels similar to the wild-type strain. In contrast, the rate of resistant responses (Types A + B + C) was the majority in leaves infected with theΔmokmt1, Δmoset1, Δmokmt3, Δmokmt6 andΔmokmt2h mutants (Table 3). In leaves inoculated with theΔmoset1 mutant, Type A (no reaction) was predominant, (77.3%), suggesting that this mutant mostly failed to penetrate plant cuticle and/or cell walls. TheΔmokmt3 andΔmokmt2h mutants induced cytological responses at very similar rates in the host cells. Infection by the two mutants was mostly prevented by the HR (~60%), and partly blocked at the papilla (~20%). Only a small percentage of germlings successfully formed invasive hyphae in infected cells (Table 3). TheΔmokmt1 andΔmokmt6 mutant showed a slight reduction in compatible interaction rate (Type D), and slight increases in incompatible interaction rates (Type A–C) (Table 3).
The order of the degrees of reduction in KMT-mutant pathogenicity was as follows: Δmoset1 >Δmokmt2h >Δmokmt3 >Δmokmt1≈Δmokmt6. The KMT mutantsΔmokmt4, Δmokmt5, andΔmoset6 showed no detectable differences from the wild-type strain in all infection assays in this study. Therefore, we concluded that MoSET1 played the most important role in infection-related morphogenesis in M. oryzae, and focused on MoSET1 for our further studies.
Pharmacological examination was performed to gain an insight into which stage in the signaling pathway leading to appressorium formation was blocked in Δmoset1 mutants. Chemical and physical signals from host plants can trigger infection-related morphogenesis in M. oryzae. One such chemical signal is 1, 16-hexadecanediol, a plant cutin monomer released from the plant cuticle by degradation enzymes produced by the fungus [35]. After perception of external signals, the secondary messenger cyclic AMP (cAMP) plays a crucial role in the signaling pathway leading to appressorium formation in M. oryzae [36, 37]. Therefore, the effects of 1, 16-hexadecanediol and cAMP on infection-related morphogenesis in the Δmoset1 mutant were examined. In the presence of 5 mM cAMP or 5 μM 1, 16-hexadecanediol, the rates of appressorium formation were greatly restored (to over 80%) in the Δmoset1 mutant, though rates were still lower than seen in the wild-type strain (Fig 4B and 4C). It is noteworthy that there was no significant difference in the rate of appressorium formation between treatments with cAMP and 1, 16-hexadecanediol, suggesting that Δmoset1 mutants may have defects in the production and/or perception of external signals from host plants.
Inoculation assays were performed to determine whether the addition of exogenous 1, 16-hexadecanediol and cAMP restored the pathogenicity of Δmoset1 mutants. 5 mM cAMP or 5 μM 1, 16-hexadecanediol was added to conidia suspension of Δmoset1 mutants, and then spotted on leaves of the susceptible wheat cultivar Norin 4. No symptoms were observed on leaves inoculated with the chemical treatments (S11A Fig), suggesting that defects in appressorium formation were not the only cause making the Δmoset1 mutant noninfectious in wheat. To further examine this finding, wound inoculation tests were performed on Norin 4. The Δmoset1 mutant failed to cause disease on the wounded leaves (S11B Fig), suggesting that the Δmoset1 mutant had some deficits in its ability to develop disease, even after entering into plant tissues. Overall, these results suggested that MoSET1 is involved in the regulation of genes required for external signaling perception and disease development in plant cells.
Chromatin immunoprecipitation sequencing (ChIP-seq) and RNA sequencing (RNA-seq) analyses [38], using chromatin and RNA samples extracted from vegetative mycelia and germination tubes of the wild-type andΔmoset1 strains, were performed to examine genome-wide H3K4 methylation and MoSET1 distribution patterns during infection-related morphogenesis, and to determine their relationships with gene expression (S2 Table). Germination tubes were collected after 6 h incubation, when appressoria had begun to form; genes involved in appressorium formation were expected to be at their most active at this time point. To carry out ChIP-analysis of MoSET1, N-terminal FLAG-tagged MoSET1 was constructed and introduced into theΔmoset1 mutant. The phenotypic defects of theΔmoset1 mutant recovered when FLAG-tagged MoSET1 was introduced in the mutant (S12 Fig), indicating that FLAG-tagged MoSET1 was functional.
ChIP- and RNA-seq data were visualized by showing a representative chromosomal region (Fig 5). DNA immunoprecipitated with H3K4me2 and H3K4me3 antibodies predominantly localized to coding regions in the M. oryzae genome. As a control to show overall H3 levels, ChIP-seq data with antibodies against a C-terminal peptide of histone H3 was also presented (Fig 5). H3K4me3 accumulated relatively more in the 5′ gene regions as reported in other organisms (Figs 5 and S13) as reported in other organisms [39–41]. The number of genes showing differences in normalized mean coverage of H3K4me2 and H3K4me3 enrichment between vegetative mycelia and germination tubes is presented in Table 4. Genes with altered ChIP enrichment (p < 0.01), and over-represented Gene Ontology (GO) categories in the gene sets, are listed in S3 and S4 Tables. Changes in the H3K4me3 ChIP coverage were more frequently detected than those in the H3K4me2 coverage, suggesting that H3K4me3 could be a dynamic mark for gene regulation during infection-related morphogenesis in M. oryzae.
MoSET1 ChIP-seq reads were also largely mapped to gene regions. In contrast to H3K4 methylation that showed specific enrichment patterns among genes, MoSET1 appeared to be distributed rather ubiquitously to almost every gene, albeit at varying levels. In Table 4, the number of genes showing differences in MoSET1 enrichment between vegetative mycelia and germination tubes is presented. In some cases such as MGG_11148 and MGG_11149, relative enrichment of H3K4 methylation in mycelia than in germination tubes was associated with levels of MoSET1 enrichment at the loci (Fig 5). Consistently, in 91 of 133 (68.4%) and 268 of 399 (67.2%) genes showing significant enrichment of H3K4me2 and H3K4me3, respectively (Table 4), normalized mean coverage of MoSET1 was concomitantly increased in germination tubes. However, it is not likely that different levels of H3K4 methylation among genes were simply attributed to levels of MoSET1 localization to their loci. For instance, while similar levels of MoSET1 coverage were observed at the MGG_01752 and MGG_01753 loci in germination tubes, much higher H3K4me2/me3 coverage was detected at the MGG_01752 locus than at the MGG_01753 locus (Fig 5). These results implied that MoSET1 could principally distribute throughout the genome but might not be always enzymatically active.
To gain a global view of the relationship between H3K4 methylation and gene expression, the M. oryzae genes were categorized into groups of 100 genes based on their expression levels, and the levels of H3K4 methylation in the gene groups were analyzed. Levels of H3K4me2 and H3K4me3 decreased as RNA levels decreased in mycelia and germination tubes (Fig 6A and 6B), indicating that H3K4 methylation was associated with active gene expression in M. oryzae in a similar way as reported in other organisms [39–41].
Next, we examined whether changes in the H3K4me2/me3 patterns were associated with gene activation or silencing during infection-related morphogenesis in M. oryzae at a global scale. RNA-seq analysis revealed that a total of 4,077 genes showed significant increases (1,936 genes) or decreases (2,141 genes) (p < 0.001) in expression levels in germination tubes (Table 4). These were sorted into five up-regulated and five down-regulated gene groups, and changes in H3K4me2/me3 ChIP enrichment in the groups were plotted (Fig 6C). The median H3K4me2 levels slightly decreased as the magnitude of transcript reductions increased. The H3K4me3 profile showed a more dynamic correlation with the transcript levels compared with the H3K4me2 levels, and the medians of the H3K4me3 levels were higher in the up-regulated gene groups and lower in the down-regulated gene groups.
It is noteworthy that the transcriptional activity of a gene was not always associated with local enrichment of H3K4 methylation as reported previously [18]. Higher H3K4me2/me3 coverage in mycelia than in germination tubes was accompanied with higher gene expression in some cases such as MGG_11149 in Fig 5. The up-regulation of the MGG_11149 gene in mycelia was significantly diminished in theΔmoset1 mutant, supporting the idea that H3K4me2/me3 contributes to gene activation. However, transcriptional activation of the neighbor genes (MGG_01755, MGG_01756, and MGG_01757) in germination tubes did not concomitantly occur with apparent H3K4me2/me3 enrichment at their loci (Fig 5). In addition, with the MGG_11148 gene, H3K4me2/me3 enrichment in the wild-type strain appeared to be accompanied with gene activation but a similar change in gene expression also occurred in theΔmoset1 mutant (Fig 5). Such apparent discrepancies were observed in many other cases. For example, H3K4me3 were significantly enriched in 398 genes and depleted in 223 genes in germination tubes compared to in mycelium (Table 4). Increase and decrease in RNA-seq read coverage were not accompanied with the H3K4me3 enrichment and depletion in 70 (17.6%) and 59 (26.5%) genes, respectively. Thus, H3K4 methylation tends to be overall associated with transcriptionally active genes, but the mechanism of gene regulation by H3K4 methylation is fairly complex, and possibly affected by other histone modifications.
The roles of MoSET1 in gene regulation were investigated by RNA-seq analysis of the Δmoset1 mutant during infection-related morphogenesis. A total of 2,572 genes were differentially expressed in Δmoset1 mycelia compared with the wild-type strain (p < 0.01), with 1,491 genes up-regulated and 1,081 down-regulated. Similarly, in germination tubes, 1,388 genes were up-regulated and 1,044 genes down-regulated in the Δmoset1 mutant. These results indicated that a significant amount of M. oryzae genes were affected by the Moset1 mutation. Interestingly, the number of genes up-regulated in the Δmoset1 mutant was comparable to, or even more than, the number of down-regulated genes, suggesting that MoSET1 directly or indirectly plays a role in gene repression, as well as in gene activation.
To analyze the characteristics of differently expressed genes between the wild-type and Δmoset1 strains, we examined the frequency distribution of genes belonging to these gene groups in mycelia (Fig 7A) and germination tubes (Fig 7B), based on the expression levels in the wild-type strain. In both mycelia and germination tubes, genes down-regulated in the Δmoset1 strain were highly biased to the high expression gene groups in the wild-type strain, while those up-regulated in the Δmoset1 strain were more frequently distributed in medium and low expression level gene groups (Fig 7A and 7B).
We next addressed how the moset1mutation affected gene regulation by comparing the fold change (FC) in gene expression during infection-related morphogenesis between the wild-type andΔmoset1 strains. A log2-scale scatter plot showed a positive linear correlation between the FC values in the wild-type andΔmoset1 strains (p = 6.8E-06) (Fig 7C). However, regression analysis gave the equation, y = 0.51x-0.04 with the correlation coefficient, r2 = 0.49, indicating that the correlation was only moderate. The slope lower than one indicated that gene expression changes were generally more marked in the wild-type strain (x-axis) than in theΔmoset1 mutant (y-axis) for both up- and down-regulated genes (Fig 7C). This supported the conclusion that MoSET1 contributed to bilateral and global gene regulation during infection-related morphogenesis in M. oryzae.
To assess the effect of MoSET1 on gene induction and repression during infection-related morphogenesis, we focused on a subset of 4,077 genes that showed a significant change in expression levels between wild-type mycelia and germination tubes in the RNA-seq analysis (Table 4). There were less up-regulated genes in the subset (1,936 genes) than down-regulated genes (2,141 genes). To understand the dependency of their gene expression on Moset1, we defined the criterion of “Moset1-dependent genes” based on a comparison of FC values (germination tubes/mycelia) between wild-type and Δmoset1 strains. When the rate of FC increase or decrease of a gene in the Δmoset1 strain was less than 50% of the wild-type strain, the gene was categorized as a Moset1-dependent gene. Genes not meeting the criterion were classified as “Moset1-independent genes”. Based on these criteria, 1,201 and 735 genes were categorized as Moset1-dependent and -independent up-regulated genes, respectively; and 883 and 1,258 genes were grouped as Moset1-dependent and -independent down-regulated genes, respectively. Therefore, approximately half of the transcriptional changes during infection-related morphogenesis were directly or indirectly dependent on Moset1 in M. oryzae. Dependency on Moset1 was more evident with the up-regulated genes. Lists of the Moset1-dependent and -independent genes and over-represented GO categories in the gene sets were given in S5 and S6 Tables, respectively.
Moset1-dependent and -independent genes were further classified using euKaryotic Orthologous Group (KOG) functional categories (Fig 8). The KOG category “signal transduction mechanisms” was highly over-represented in the MoSET1-dependent up-regulated gene set. Consistently, several GO categories related to “signal transduction mechanisms” were significantly over-represented in the gene set (S6 Table). In this category, forty one kinases and thirteen GTPase regulators were detected, indicating that a large number of key signal mediators were transcriptionally regulated by MoSET1, either directly or indirectly. Interestingly, twenty-four active transmembrane transporters, including MgAPT2 (MGG_02767), MgAPT3 (MGG_04066), and MgAPT4 (MGG_04852) [42] were regulated by MoSET1.
The KOG categories over-represented in the Moset1-dependent down-regulated gene set were different from those in the upregulated gene sets, and included “translation, ribosomal structure and biogenesis” and “RNA processing and modification”. Interestingly, sixty-four structural constituents of the ribosome were found in this criterion (S6 Table), indicating that MoSET1 was associated with down-regulation of a significant portion of ribosome-related genes. In addition, various nucleic acid binding proteins, especially those that bind RNA, were down-regulated in a Moset1-dependent manner. These included proteins homologous to nuclear ribonucleoprotein, RNA helicase, tRNA synthetase, rRNA biogenesis protein, poly(A) polymerase, and poly(A)-binding protein.
Finally, we addressed whether Moset1-dependent gene regulation is directly related to H3K4 methylation. Levels of H3K4me2 and H3K4me3 enrichment in mycelia and germination tubes were plotted separately by the four gene criteria indicated in Fig 9. In the up-regulated gene groups, levels of H3K4 methylation were generally higher in the Moset1-dependent genes than in the Moset1-independent genes. In addition, the Moset1-dependent genes showed stronger enrichment of H3K4me2 in germination tubes, where they were up-regulated, than did the Moset1-independent genes (Fig 9). Thus, in the up-regulated gene groups, changes in H3K4 methylation were more dynamic in the Moset1-dependent genes than in the Moset1-independent genes, suggesting direct contribution of H3K4 methylation to gene activation.
In contrast, in the down-regulated gene groups, while both H3K4me2 and H3K4me3 levels decreased in germination tubes, where the genes in these criteria were down-regulated, only slight difference was observed in H3K4me2 and H3K4me3 enrichment patterns between the Moset1-dependent and -independent genes. This may suggest that the Moset1-dependency in the down-regulated genes was not directly resulted from H3K4 methylation.
The gene knockout studies of the eight KMT genes in M. oryzae revealed that MoKMT1, MoSET1, MoKMT3, MoKMT6, and MoKMT2H played significant roles in infection-related morphogenesis and/or pathogenicity to varying degrees, while MoKMT4, MoKMT5, and MoSET6 did not. Δmokmt1 mutants did not display detectable defects in infection-related morphogenesis, but showed a slight reduction in vegetative growth and of pathogenicity on host plants. MoKMT1 belongs to the KMT1 family responsible for methylation at H3K9 and is paralogous to N. crassa DIM-5. In N. crassa, H3K9me3 catalyzed by DIM-5 is recognized by HP1 that forms a complex with DIM-2 DNA methyltransferase [9]. HP1 is a structural protein essential for heterochromatin formation, and leads to gene repression [43]. Since H3K9me3 is a conserved epigenetic mark for gene repression, MoKMT1 is likely involved in gene repression in M. oryzae. Interestingly, Δmokmt6 mutants showed a reduction in pathogenicity at levels similar toΔmokmt1 (Tables 2 and 3). The KMT6 family enzymes catalyze H3K27 methylation, a mark for gene repression. In F. graminearum, a KMT6 deletion mutant exhibited developmental defects and reduced pathogenicity as didΔmokmt6 mutants [18]. Similarly, in the plant pathogenic fungus, Leptosphaeria maculans, silencing of LmDIM5 belonging to the KMT1 family resulted in a reduction in pathogenicity [20]. Thus, gene repression itself or proper switching from gene repression to expression may be required for the full pathogenicity of the fungi, or these KMT genes may have functions other than gene repression.
Δmokmt3 andΔmokmt2h mutants showed a significant reduction in vegetative growth, germination, appressorium formation, conidiation, and pathogenicity to host plants. Thus, these mutants had defects in every phenotypic assay performed in this study. The reduction in pathogenicity was more severe inΔmokmt2h mutants than inΔmokmt3 mutants. MoKMT3 is paralogous to N. crassa SET-2, which belongs to the KMT3 family responsible for methylation at H3K36. H3K36me3 levels peak within the body of active genes, and may be associated with transcription elongation through contributing to the maintenance of chromatin architecture [44, 45]. In N. crassa, SET-2 loss-of-function mutants show various defects, including slow vegetative growth, low conidiation, and female sterility [16]. This is consistent with our results. Thus, MoKMT3 possibly affected the correct expression of a number of genes involved in phenotypic defects.
While possible paralogs of MoKMT2H are widely conserved in ascomycete fungi, their biological roles have not been well-characterized. Based on phylogenetical analysis, the most closely related KMT to MoKMT2H in mammals is ASH1L, which is implicated in H3K4 and H3K36 methylation, and in transcriptional activation of certain genes, including Hox genes [46, 47]. Therefore, if MoKMT2H is a functional homolog of ASH1L, it follows that MoKMT2H may also contribute to gene activation in M. oryzae.
MoSET1 was among the most crucial KMTs for infection-related morphogenesis and symptom development on host plants in M. oryzae. Δmoset1 mutants showed severe defects in vegetative growth, appressorium formation, conidiation, and pathogenicity to host plants, but not in the rate of germination. MoSET1 catalyzes methylation at H3K4, an evolutionary conserved epigenetic mark for gene activation. Our data suggested that this epigenetic mark could be the most important histone methylation for infection-related gene expression in M. oryzae. RNA-seq analysis of the Δmoset1 and wild-type strains suggested that approximately half of the genes induced or repressed during infection-related morphogenesis were dependent on MoSET1.
Moset1-dependent genes appeared to be involved in various infection-related processes. One such process is appressorium formation. cAMP signaling is crucial for appressorium formation in M. oryzae [36, 37]. Several genes involved in the cAMP signaling pathway leading to appressorium formation, including MacI (MGG_09898), MacI-interacting protein (MGG_05531), Mck1 (MGG_00883), and Sum1 (MGG_07335) were categorized as Moset1-dependent up-regulated genes [48]. The deficiency in transcriptional up-regulation of these genes in the germination tubes may be the cause of severe reduction in appressorium formation in the Δmoset1 mutants. This assumption is consistent with appressorium formation in the mutants being restored by exogenous cAMP.
Pathogenicity of Δmoset1 mutants to host plants was not recovered by cAMP addition, indicating that Δmoset1 mutants had defects in pathogenicity other than appressorium formation. A large number of transporters were up-regulated in a Moset1-dependent manner in germination tubes (Fig 8). One such gene, MgApt2 is a P-type ATPase involved in exocytotic processes during plant infection [42]. Exocytotic mechanisms are involved in the delivery of proteins into plant cells to suppress plant defenses.
Signal transducers were highly enriched in the Moset1-dependent gene set. MgATG1 (MGG_06399), a serine/threonine-protein kinase found in this group, is involved in autophagy and the generation of normal turgor pressure in the appressorium, and is thus essential for successful infection [49]. However, many important protein kinases for fungal pathogenesis, including Pmk1 (MGG_09565), CpkA (MGG_06368), Msp1 (MGG_05344), and MST7 (MGG_00800) were not Moset1-dependent genes. Thus, not all signal transducers responding environmental stimuli required MoSET1 for their activation.
Cell-wall degradation enzymes (CWDEs) are possible Moset1-dependent contributors to the full pathogenicity of the fungus. We previously reported that CWDEs, such as GH7 and GH8 cellulases and GH10 and GH11 xylanases, were greatly activated during infection [50, 51]. Their gene activation was often induced by presence of relevant cellulose or xylan substrates. Substrate-dependent gene activation of the cellulases was severely compromised in the Δmoset1 [31]. Therefore, lack of CWDE activation during infection may be the cause of the severe reduction observed in the pathogenicity of the Δmoset1 mutants on the host plants. Gene activation of CWDEs was, however, barely detectable in the RNA-seq analysis; this is a consequence of RNA being extracted from germination tubes on slide glasses, thus having no available enzyme substrates.
It should be note that, sinceΔmoset1 mutants were also impaired in vegetative growth, the role of MoSET1 in gene regulation was not infection specific. Thus, the down-regulation of such general genes could also contribute to the loss of pathogenicity inΔmoset1 mutants.
In eukaryotes, H3K4 methylation is an epigenetic mark for gene activation. ChIP-seq or ChIP-chip analysis together with transcriptome analysis in human, Arabidopsis, and S. cerevisiae revealed a global positive correlation between H3K4me2/me3 and active transcription [39–41]. In M. oryzae, substrate-induced gene expression of GH6 and GH7 cellulases was associated with enrichment of H3K4me2 [31]. Interestingly, however, expression levels of GH6 and GH7 cellulases under non-inducing conditions increased in the Δmoset1 mutant, suggesting a possible role of H3K4 methylation in gene repression [31]. This is consistent with the observations in Aspergillus nidulans and A. fumigatus, where a deletion mutant of the CclA gene, which encodes a component of the COMPASS complex catalyzing H3K4 methylation, results in a reduction in H3K4me2 and H3K4me3, and also causes increased gene expression of cryptic SM gene clusters [21, 22]. Thus, CclA-mediated H3K4 methylation appears to contribute to gene silencing of SM clusters. It is noteworthy that SET1 in S. cerevisiae was originally identified as a gene required for transcriptional silencing of silent mating-type loci in the subtelomeric region [10]. Subsequently, SET1 was demonstrated to play a role in silencing rDNA and the retrotransposon Ty1 in S. cerevisiae [23, 24]. Consistently, contribution of MoSET1 to the repression of ribosome-related genes was shown in this study (Fig 8). Recently, SET-1 was reported to have a role in DNA methylation of the frq promoter in N. crassa [52]. Thus, SET1 orthologs in fungi are involved in gene silencing in addition to gene activation.
Our RNA-seq analysis revealed that significant numbers of M. oryzae genes were up- or down-regulated in the Δmoset1 mutant in comparison with the wild-type strain, supporting that H3K4 methylation is directly or indirectly involved in both gene activation and repression. Moset1-dependent gene up-regulation was largely detected in highly-expressed genes in the wild-type strain, whereas Moset1-dependent gene repression was more frequently observed in genes with middle or low expression levels (Fig 7). It is to be noted that the roles of Moset1-dependent gene repression in the pathogenicity of the fungus were so far unclear while Moset1-dependent gene activation, most likely, indeed contributed to infection-related morphogenesis in M. oryzae as discussed above.
Apparent differences in the roles of KMT2 proteins among organisms can be attributed to the experimental approaches. In fact, this study demonstrates that results obtained by ChIP-related techniques in fungi are not much different from those obtained in higher eukaryotes. The role of SET1 orthologs in gene repression has mainly been revealed by gene knock-out approaches that are often used in fungi but seldom in higher eukaryotes. In higher eukaryotes, several SET1 homologs redundantly serve as catalytic enzymes for methylation at H3K4. At least six Set1 homologs (Set1A, Set1B, MLL1, MLL2, MLL3, and MLL4) have been identified in mammalian cells, making gene knock-out strategies ineffective. Thus, it might be possible that the complete loss of SET1 homologs in the genome uncovered their additional functions that were hard to find by other approaches. However, it also should be noted that, in gene knockout approaches, it is difficult to distinguish direct and indirect effects of the loss of the target gene. Since SET1 homologs positively regulate global gene expression, their knockout mutants might fail to activate genes required for proper gene repression. Thus, de-repression of genes in theΔmoset1 mutant may not arise directly from the loss of the gene but may come from secondary causes, for example, insufficient expression of repressor genes in the mutant. Our results showed that changes in H3K4 methylation during germination tube formation were more dynamic in the Moset1-dependent gene set than in the Moset1-independent gene set among the up-regulated genes but not among the down-regulated genes (Fig 9), suggesting that the Moset1-dependency in the down-regulated genes could not be directly related to changes in H3K4 methylation. The data favors the hypothesis that the Moset1-dependent gene repression arose from indirect effects of the loss of MoSET1 even though other hypotheses are not completely eliminated. For example, recently, it has been reported that H3K4 monomethylation functioned as a mark for gene repression in several types of mammalian cells [25]. Thus, it might be possible that the lack or severe depletion of H3K4 monomethylation in the set1 or cclA deletion mutants resulted in activation of genes that were repressed under the wild-type background.
The wheat-infection M. oryzae isolate, Br48 [53] and its transformants constructed in this study (S1 Table) were kept on barley seeds media at 4°C for long-term storage [54]. For working culture, a barley grain from the stock culture was placed on a PDA (potato dextrose agar) slant media and cultured at 25°C. Fungal plugs were transferred to flasks containing complete medium (5% sucrose, 3% casamino acids, and 3% yeast extract) and incubated in a shaker at 120 rpm at 25°C for 4 days. To prepare conidial suspension, fungal strains were cultured on oatmeal agar plates (40g of oatmeal, 17g of agar in 1000 ml water) in the darkness at 25°C for 5 days. Then, aerial mycelia were removed by rubbing surface of mycelia with a sterile microtube, and further incubated under BLB light for 3 days at 25°C to induce conidiation.
In this study, a split-marker gene disruption strategy [32] was used to obtain a gene knock-out mutant in M. oryzae (see S1–S7 Figs). First, PCR products of the upstream and downstream of a targeted gene were cloned separately into the multiple cloning site of pSP72-hph that carries the Hygromycin resistance gene cassette [55]. Primers used in this study are given in S7 Table. PCR fragments amplified from the resulting 5’ and 3’ constructs were mixed and introduced into fungal spheroplasts by a polyethylene glycol (PEG)-mediated method as previously described [56]. For initial screening, colonies PCR were performed with appropriate sets of primers for each gene. The candidate strains were further examined by Southern blot analysis. Fungal genomic DNA was extracted using Plant Genomic DNA Extraction Miniprep System (Viogene) following the manufacturer’s instruction. Southern blot analysis was performed using the DIG DNA Labeling and Detection Kit (Roche Applied science). Ten to twenty micrograms of genomic DNA were digested by appropriate restriction enzymes. The digests were separated by agarose gel electrophoresis and transferred to Hybond N+ (Amersham biosciences). The hybridization procedures were carried out according to the manufacturer’s instructions. The positions of DIG-labeled probes used in Southern blot analysis are given in S1–S7 Figs.
Genetic complementation of KMT deletion mutants was performed by introducing the corresponding native genomic fragment to them. Genomic DNA fragments containing KMT genes with their 5’flanking and 3’flanking regions were amplified with pairs of specific primers (S7 Table) using KOD FX Neo (Toyobo), and cloned into pBluescript SK(+). Each of the resulting plasmids was introduced into the corresponding KMT mutant with pII99 carrying the geneticin-resistance gene.
Growth rates (colony diameter) of M. oryzae mutants on PDA media were measured up to 14 days with three replications. For conidiation assay, conidia were harvested 3 days after BLB induction by suspending them with 20ml sterile distilled water per plate. Spore concentration was estimated by microscopic observation of at least 20 visual fields using a hemocytometer. For conidial germination and appressorium formation assays, conidial suspension (105 spores per ml) dropped on slide glasses was incubated in a humidity box at 25°C. The rates of conidia germination and appressorium formation were counted by microscopic observation of at least 200 spores after 5 and 24 h incubation at 25°C, respectively.
Infection assay was performed as described previously [57]. Wheat and barley seedlings were grown in vermiculite supplied with liquid fertilizer in plastic pots (5.5cm×15cm×10cm) at 22°C in a controlled-environment incubator with a 12h photoperiod for 8 days. The plant cultivars used were wheat cultivars “Norin 4”, “Chinese spring” and Thatcher, and barley cultivars “Russian No.74” and “Nigrate”. Conidia suspension (1–2☓105 spores/ml) containing 0.01% Tween 20 was sprayed to the primary leaves of 8-day-old wheat and barley seedlings. The inoculated seedlings were maintained under high humidity and dark conditions for 24 hours then moved to an incubator at 22°C with a 12h photoperiod for 5 days. Symptoms appeared were assessed based on the size and color of lesions to determine infection type. The size of a lesion was rated from 0 to 5: 0, no visible evidence if infection; 1, pinpoint spots; 2, small lesion (<1.5 mm); 3, lesion with an intermediate size (<3 mm); 4, large and typical lesion; 5, complete blighting of leaf blades. Green (G) and brown (B) lesions were regarded as susceptible and resistant responses, respectively [58].
For microscope observation of host cell response to M. oryzae, inoculated leaves were picked up at 48 h post inoculation (hpi) and deeply boiled in alcoholic lactophenol (lactic acid/phenol/glycerol/distilled water/ethanol = 1:1:1:1, v/v/v/v) for 2 min as descried previously [57]. Samples were observed using an epifluorescence microscope under bright and fluorescent fields. Host response was classified into four types: no reaction, papilla formation, hypersensitive reaction (HR), and hyphal growth.
Fungal mycelia powder was suspended in 1×TBS buffer (50 mM Tris-HCl [pH 7.5], 150 mM NaCl) containing 1% Nonidet P-40. The homogenates were centrifuged (12,000 rpm, 2 minutes), and supernatants were collected. Proteins in the supernatants were then heat-treated at 80°C for 10 min to precipitate contaminating proteins. The supernatant was recovered by centrifugation, and subjected to 15% SDS_polyacrylamide gel electrophoresis. After blotted to PVDF membrane, proteins were probed with the following primary antibodies; anti-H3K4me1 (Active motif #39298), anti-H3K4me2 (Active motif #39141), anti-H3K4me3 (Active motif #39159), anti-H3K9me3 (Active motif #39161), anti-H3K27me3 (Active motif #39535), anti-H4K20me3 (Active motif #39181) and a C-terminal peptide of histone H3 (Active motif #39163). Washing was performed three times with TBS-T buffer containing a higher concentration of NaCl (50 mM Tris-HCl [pH 7.5], 190 mM NaCl, 0.05% Tween 20). Proteins reacting with the primary antibodies were visualized by appropriate peroxidase (HRP)-conjugated secondary antibodies and ECL plus western blotting detection regents.
An N-terminal 2xFLAG-tagged MoSET1 construct was made by PCR amplification with the primers, FLAG-MoSET1-F and MoSET1-TGA-R (S7 Table), and then fused to the native MoSET1 promoter and terminator sequences by In-Fusion HD cloning kit (Clontech Laboratories) at EcoRV site in pBluescript SK(+) with the primers, IF-PMoSET-F, IF-PMoSET-R, IF-TMoSET-F, and IF-TMoSET-R (S7 Table). The resulting construct was introduced into the Δmoset1 mutant and used in chromatin immuno-precipitation (ChIP) analysis.
ChIP experiments were performed with germinating conidia and vegetative mycelia using the ChIP-IT Express kit (Active motif #53008) according to manufacturer's instructions using sonication as a method for chromatin shearing. In addition to the antibodies used in western blot analysis, Anti-DDDDK-tag mAb-magnetic beads (Medical & Biological Laboratories, Japan) was used in ChIP experiments. Briefly, samples (100mg) were treated with 1% formaldehyde by shaking gently (100rpm) for 30 minutes at room temperature. Chromatin was sheared on ice by sonication using a Bioruptor apparatus (Diagenode) for 3 cycles of 1 min on at high intensity (200 W) and 30 sec off, followed by 4 cycles of 1 min on at medium intensity (160 W) and 30 sec off. The size of the sheared chromatin was around 200 to 1,000 bp as determined by agarose gel electrophoresis. After immunoprecipitation with an appropriate antibody, DNA fragments were recovered by Proteinase K treatment. Indexed ChIP-seq libraries were prepared with the NEBNext ChIP-Seq Library Prep Master Mix Set for Illumina (New England Biolabs) according to the manufacturer’s instructions. Fragment size selection of ChIP-seq libraries was done using Agencourt AMPure XP beads (Beckman Coulter). The products were purified and enriched with PCR to create the final double stranded cDNA library. The MiSeq system (Illumina) was used to sequence the cDNA library.
RNA extraction and cDNA preparation were carried out as described previously with a few modifications [31]. Total RNA was isolated using Sepasol RNA I Super (Nakalai Tesque), and used for cDNA synthesis using ReverTra Ace qPCR RT master mix with genomic DNA remover KIT (Toyobo). Depletion of rRNA was performed using the Ribo-Zero rRNA removal kit for human/mouse/rat (Epicentre). Indexed RNA-seq libraries were prepared with the NEBNext Ultra™ RNA Library Prep Kit for Illumina kit (New England Biolabs) or NEXTflex™ Directional RNA-Seq Kit (BIOO Scientific Corp.) according to the manufacturer’s instructions. Fragment size selection of RNA-seq libraries was done using Agencourt AMPure XP beads. The products were purified and enriched with PCR to create the final double stranded cDNA library. The MiSeq system (Illumina) was used to sequence the cDNA library.
The RNA-seq and ChIP-seq reads (75–120 bp) were mapped to the genome of the Magnaporthe oryzae strain 70–15 (release 8.0, http://www.broadinstitute.org/) using TopHat v2.0.10 [59] and bwa v0.6.2-r126 [60], respectively. RNA-Seq and ChIP-seq data were visualized in the Integrative Genomics Viewer genome browser [61]. The edgeR package [62] for R v3.0.1 [63] was used for TMM normalization [64], identification of differentially expressed genes (DEGs) from RNA-seq data, and detection of genes differentially enriched for histone modifications from ChIP-seq data with the corrected p-value [65] cutoffs of 0.01 (ChIP-seq analysis) or 0.001 (RNA-seq analysis). The GOstats package [66] was used to identify statistically significant enriched Gene Ontology (GO) categories.
|
10.1371/journal.ppat.1004777 | The Impact of Host Diet on Wolbachia Titer in Drosophila | While a number of studies have identified host factors that influence endosymbiont titer, little is known concerning environmental influences on titer. Here we examined nutrient impact on maternally transmitted Wolbachia endosymbionts in Drosophila. We demonstrate that Drosophila reared on sucrose- and yeast-enriched diets exhibit increased and reduced Wolbachia titers in oogenesis, respectively. The yeast-induced Wolbachia depletion is mediated in large part by the somatic TOR and insulin signaling pathways. Disrupting TORC1 with the small molecule rapamycin dramatically increases oocyte Wolbachia titer, whereas hyper-activating somatic TORC1 suppresses oocyte titer. Furthermore, genetic ablation of insulin-producing cells located in the Drosophila brain abolished the yeast impact on oocyte titer. Exposure to yeast-enriched diets altered Wolbachia nucleoid morphology in oogenesis. Furthermore, dietary yeast increased somatic Wolbachia titer overall, though not in the central nervous system. These findings highlight the interactions between Wolbachia and germline cells as strongly nutrient-sensitive, and implicate conserved host signaling pathways by which nutrients influence Wolbachia titer.
| Many invertebrate organisms carry bacterial endosymbionts within their cells. In many cases, this ensures host access to resources provided by the endosymbionts, and reciprocally, a rich source of host-supplied nutrients supports bacterial growth and reproduction. However if bacterial reproduction is uncontrolled, an over-abundance of bacteria will ultimately destroy the host cell. Here we explore the factors that regulate endosymbiont abundance in host cells. We focused on Wolbachia endosymbionts that are carried naturally in the germ cells of fruit flies. Specifically, we determined whether dietary nutrients affect the amount of Wolbachia bacteria carried by female flies. We found that yeast-enriched diets strongly depleted Wolbachia in fly ovarian cells. By contrast, sucrose-enriched diets doubled the amount of Wolbachia in ovarian cells. In addition, we found that this response to diet is mediated through highly conserved TORC1 and insulin signaling pathways in the fly. Recent studies have revealed that host diet dramatically influences the types and abundance of gut microbes. Our study informs how host diet affects endosymbiotic bacteria housed within specific types of host cells.
| Microbial endosymbionts have a profound impact on host metabolism and there are numerous examples in which microbes provide essential nutrients to the host [1–14]. In contrast, considerably less is known regarding how host metabolism and nutrition affect resident endosymbionts. To date, there is evidence that restricting the supply of host carbon, nitrogen and phosphorous significantly limits the number of Chlorella endosymbionts of green hydra and dinoflagellate endosymbionts of cnidarians [1]. Researchers have also observed that exposure to high levels of exogenous thiamine monophosphate suppresses the titer of Sodalis and Wigglesworthia endosymbionts in tsetse flies [15,16]. In this largely unexplored area, many outstanding questions remain: What are the host and endosymbiont metabolic and signaling pathways involved in nutrient sensing? To what extent do endosymbionts exhibit tissue-specific responses to nutrient availability? How are the rates of endosymbiont replication and cell death influenced by host metabolism and nutrients?
The symbiosis between Wolbachia and Drosophila is an excellent system to experimentally address these issues. Wolbachia are obligate intracellular endosymbionts carried by an estimated 40% of all insect species, including the established model organism Drosophila melanogaster [17–20]. Though Wolbachia endosymbionts are naturally carried within germline cells of both male and female insects, Wolbachia are ultimately removed from sperm prior to completion of spermatogenesis [17,18,21–25]. Thus, Wolbachia rely upon transmission through the maternal germline for their success. In addition to its functional importance in Wolbachia transmission, the well-characterized molecular and cell biology of Drosophila oogenesis has provided considerable contextual information and experimental tools that can be applied to studies of Wolbachia-host interactions [18,26–30].
The primary developmental units of the ovary that carry Wolbachia are referred to as egg chambers [27,28]. In each egg chamber, an outer layer of somatic follicle cells encapsulates an interconnected cyst of germline cells, comprised of 15 nurse cells and an oocyte. Wolbachia are initially loaded into these developing cysts during the first mitotic division from a Wolbachia-infected germline stem cell [18,31]. This germline Wolbachia population is amplified over time by binary fission and likely to some extent by exogenously invading Wolbachia [31–36]. Wolbachia persist in the germline throughout oogenesis, and a subset of the bacteria concentrate at the oocyte posterior pole during mid- to late oogenesis [31,37,38]. This ensures incorporation of Wolbachia into germline progenitor cells that form at the embryonic posterior pole, perpetuating the maternal germline transmission cycle [39]. Thus, maintenance of a sufficient Wolbachia titer in germline cells is important for success of the germline-based transmission strategy.
Here we examined how host diet affects Wolbachia titer in Drosophila melanogaster. The data demonstrate that yeast-enriched diets suppress Wolbachia titer and lead to altered nucleoid morphology during oogenesis. Genetic and chemical disruptions indicate that the somatic insulin and TORC1 pathways (Fig. 1) are required for yeast-based suppression of oocyte Wolbachia titer. The data also indicate that sucrose-enriched diets increased oocyte Wolbachia titer, with little impact on nucleoid morphology. Evidence indicates that yeast-enriched diets substantially increase somatic Wolbachia titers, though this was not the case in the central nervous system (CNS). These studies demonstrate that Wolbachia, and likely other bacterial endosymbionts, exhibit distinct, tissue-specific responses to host nutrients that involve conserved signaling and metabolic pathways.
Nutrient availability strongly affects the life cycle of cultured bacteria, raising questions about how host nutrient conditions affect intracellular Wolbachia bacteria. As D. melanogaster in nature preferentially consume yeast [40–45], we tested the effect of dietary yeast on Wolbachia titer in vivo. Female flies were aged first for two days on standard food, then fed yeast paste for 3 days, and examined for Wolbachia titer in oogenesis. Ovarian tissues were stained with propidium iodide to label Wolbachia DNA, and the Wolbachia nucleoids imaged in oocytes of stage 10 egg chambers by confocal microscopy [38]. This analysis demonstrated that yeast paste-fed oocytes carried far less Wolbachia than control oocytes (Fig. 2A-B) (S1 Table). Wolbachia were further quantified within single oocyte focal planes to determine relative titer for each condition [32]. This revealed that Wolbachia titer in yeast paste-fed oocytes was at 27% of the control level. Oocytes treated with standard fly food exhibited an average of 229 +/- 21.1 Wolbachia puncta (n = 30), as compared to yeast paste-fed oocytes that carried 62.6 +/- 4.33 Wolbachia (n = 29) (p < 0.001) (Fig. 2C). This indicates that host exposure to yeast paste significantly reduces Wolbachia titer in oogenesis.
One possibility is that yeast paste diets reduce oocyte titer because other critical nutrients provided by standard fly food are unavailable. To address this issue, 2-day old Drosophila were fed with either standard food diluted 1/3 with water, thereafter referred to as “control food”, or fed with standard food diluted 1/3 with yeast paste, thereafter referred to as “yeast-enriched food” (S1 Table). After 3 days of exposure to these conditions, titer was assessed in oogenesis. The yeast-enriched condition exhibited 55% of the control titer level, with controls displaying 124 +/- 10.8 Wolbachia (n = 58), compared to yeast-enriched oocytes carrying 68.7 +/- 5.12 Wolbachia (n = 35) (p = 0.001) (Fig. 2D). To further assess whether this is due to differences in food hydration between control and yeast-enriched conditions, we also exposed flies to a 1/3 dilution of corn syrup into standard fly food (S1 Table). Although corn syrup-enriched food is less hydrated than control food, it resulted in similar oocyte titer measurements as the control, with an average of 128 +/- 12.9 Wolbachia visible per oocyte (n = 31) (Fig. 2D). These data together suggest that yeast-induced titer reduction is not due to depletion of specific nutrients or water available in standard food. Rather, the data indicate that dietary yeast is responsible for reducing Wolbachia titer carried by oocyte cells.
To determine whether dietary yeast can induce a similar oocyte titer response in wild insects as seen in laboratory fly stocks, Drosophila melanogaster and Drosophila simulans were collected from nature. These flies were exposed to yeast-enriched food and assessed for Wolbachia titer in oogenesis. We found that oocyte Wolbachia titer in the yeast-enriched condition was at 47% of the control level, with an average of 94.8 +/- 21.8 Wolbachia detected in control oocytes (n = 12), versus 44.6 +/- 6.52 Wolbachia detected in the yeast-enriched condition (n = 13) (p = 0.029) (S1 Fig). Thus, yeast-enriched diets suppress oocyte Wolbachia titer in wild-caught Drosophila analogous to laboratory D. melanogaster strains.
To further investigate the basis for yeast-associated Wolbachia depletion in oocytes, Wolbachia titer was examined in the germline-derived nurse cells associated with the oocyte. It is currently unclear in Drosophila when or how frequently Wolbachia travel through the ring canals between the nurse cells and oocyte. Thus, it is possible that Wolbachia depletion in oocytes could be due to preferential retention in the nurse cells. To investigate this, we imaged Wolbachia in equivalent focal planes of nurse cells and oocytes within single egg chambers and analyzed their Wolbachia titer [32]. Overlaid images showing a planar reconstruction of egg chambers indicated fewer Wolbachia throughout the germline cells of yeast-exposed organisms (Fig. 3A-B). Quantitation of the yeast-enriched condition indicated that nurse cells carried 27% of the control titer level (Fig. 3C). Specifically, 52.6 +/- 4.93 Wolbachia per nurse cell were detected in the control (n = 20), in contrast to 14.4 +/- 1.65 Wolbachia per nurse cell in the yeast-enriched condition (n = 20) (p < 0.001) (Fig. 3C). Furthermore, oocyte titer in the yeast-enriched condition was 14% of the control level, with 420 +/- 44.6 Wolbachia detected in control oocytes (n = 17), versus 59.0 +/- 11.1 Wolbachia in oocytes from the yeast-enriched condition (n = 20) (p < 0.001) (Fig. 3D). These data indicate that Wolbachia redistribution between germline cells is not responsible for the low oocyte titer observed in yeast-exposed organisms. Rather, yeast-enriched food induces similar Wolbachia depletion in nurse cells and oocytes.
Cells coordinate intracellular events in response to exogenous nutrients using multiple signaling pathways that converge upon the Target of Rapamycin kinase complex 1 (TORC1) (Fig. 1) [46]. TORC1 can be activated by an amino-acid dependent signaling mechanism, or by insulin signaling (Fig. 1) [46–48]. To test whether TORC1 activity affects oocyte Wolbachia titer, flies were exposed to standard food containing the TORC1 inhibitor, rapamycin [49–52]. This experiment indicated that rapamycin treatment drove a 1.7-fold increase in oocyte Wolbachia titer (Fig. 4A). The average titer from control oocytes, exposed to DMSO-containing standard food, was 207 +/- 22.1 Wolbachia (n = 28). By contrast, oocytes exposed to rapamycin-containing standard food had 357 +/- 31 Wolbachia (n = 30) (p < 0.01) (Fig. 4A). Since rapamycin exposure leads to higher oocyte Wolbachia titer, this suggests that a normal consequence of TORC1 activity is suppression of oocyte Wolbachia titer.
If TORC1 function normally leads to decreased oocyte Wolbachia titer, then hyper-activation of TORC1 would be expected to drive a further reduction of oocyte titer. Branched chain amino acids (BCAAs) taken up through the Slimfast transporter can induce up-regulation of TORC1 (Fig. 1) [53–58]. Therefore, we fed flies a slurry of BCAAs diluted 1/3 into standard food (S1 Table), and assessed Wolbachia titer in oogenesis. Wolbachia titer in the BCAA condition was reduced to 77% of the control (Fig. 4B). This was indicated by an average of 137 +/- 9.71 Wolbachia in control oocytes (n = 34) versus 105 +/- 8.48 Wolbachia in oocytes from the BCAA condition (n = 33) (p = 0.015) (Fig. 4B). The data suggest that TORC1 stimulation with BCAAs drives oocyte titer reduction, opposite the effects of the TORC1 inhibitor, Rapamycin.
To further investigate a possible role for TORC1, we genetically manipulated a key regulator of TORC1 activity. Tsc2, known as Gigas in Drosophila, is downstream of the insulin receptor (Fig. 1) [59–64]. If Tsc2 function is suppressed by any means, this allows TORC1 to become active (Fig. 1) [46,64–68]. Therefore, we tested the impact of Tsc2 on oocyte Wolbachia titer by expressing Tsc2 dsRNA under the control of germline- and soma-specific GAL4 drivers [69–72]. This investigation revealed different oocyte Wolbachia titer responses to tissue-specific Tsc2 RNAi knockdowns. Our efforts to manipulate Tsc2 dosage in germline cells had no impact on oocyte titer (Fig. 4C). An average of 182 +/- 13.5 Wolbachia were detected in control oocytes (n = 53), which was not significantly different from the 207 +/- 17.7 Wolbachia detected in response to germline Tsc2 RNAi (n = 56) (Fig. 4C). By contrast, Tsc2 RNAi knockdowns in the somatic cells reduced oocyte Wolbachia titer to approximately 50% of the control level (Fig. 4D). Control oocytes exhibited an average of 402 +/- 43.4 Wolbachia (n = 24). However, oocytes somatic Tsc2 knockdown flies exhibited an average of 181 +/- 19.8 oocyte Wolbachia (n = 21) (p < 0.001) (Fig. 4D). As such, these data implicate somatic Tsc2, and thus somatic TORC1 signaling, in regulation of oocyte Wolbachia titer.
A role for somatic TORC1 in regulating oocyte Wolbachia titer raised the question of whether dietary yeast stimulates TORC1. This could occur through either protein- or insulin-based mechanisms (Fig. 1). As yeast is major source of protein for D. melanogaster, perhaps its amino acid content stimulates TORC1 to ultimately suppress oocyte Wolbachia titer. To test this possibility, we exposed flies to food enriched in Bovine Serum Albumin, prepared specifically to match the protein content of yeast-enriched food (S1 Table). Oocyte Wolbachia titer was similar for control and BSA-enriched conditions, however, with the control exhibiting 1260 +/- 102 Wolbachia (n = 26), and the BSA-enriched condition exhibiting 1190 +/- 48.2 Wolbachia (n = 18) (S2 Fig). This suggests that amino acid availability in the host diet has little impact on oocyte Wolbachia titer.
An alternate possibility is that yeast-enriched diets affect oocyte Wolbachia through insulin stimulation of TORC1. It was previously shown that dietary yeast stimulates insulin-producing cells (IPCs) the brain to release the insulin-like-peptides (Dilps) into the hemolymph [73,74]. To test whether yeast acts through somatic Dilp secretion to oocyte Wolbachia titer, we ablated the IPCs in the brain of fully mature Drosophila females. This is achieved using a dilp2: Gene-Switch-GAL4, UAS: Reaper system that specifically kills off the brain IPCs in response to a 2-week mifepristone treatment [74].
We first investigated whether mifepristone on its own modulates the yeast effect in wild-type flies. After completing a two-week exposure to either DMSO or mifepristone, flies were exposed to either control or yeast-enriched food for 3 days, and their oocyte titer levels were assessed. DMSO-treated flies exhibited substantial oocyte titer depletion in response to yeast-enriched food, down to 30% of the titer in the control condition (Fig. 5A). This was indicated by 785 +/- 64.8 Wolbachia per oocyte in the DMSO-control food condition (n = 24), in contrast to 191 +/- 26.9 Wolbachia in the DMSO-yeast-enriched condition (n = 25) (p <. 001) (Fig. 5A). Mifepristone-treated flies showed a similar titer reduction after exposure to yeast, exhibiting 21% of the titer seen in the control food condition (Fig. 5B). This was indicated by 896 +/- 77.2 Wolbachia per oocyte in the mifepristone-control food condition (n = 23), versus 264 +/- 39.5 Wolbachia in the mifepristone-yeast-enriched condition (n = 25) (Fig. 5B) (p <. 001). Therefore, mifepristone alone has no effect on yeast-based suppression of oocyte Wolbachia titer.
Next, the exact same treatment regimens were performed on flies with the dilp2: Gene-Switch-GAL4, UAS: Reaper genotype. In this experiment, DMSO-treated flies, which retained functional IPCs, exhibited a severe oocyte Wolbachia depletion in response to yeast-enriched food, exhibiting only 7% of the oocyte titer seen on DMSO-control food (Fig. 5C). This was indicated by the presence of 999 +/- 116 Wolbachia per oocyte in the DMSO-control food condition (n = 17), versus 66.5 +/- 6.61 Wolbachia in the DMSO-yeast-enriched condition (n = 20) (p < 0.001) (Fig. 5C). In stark contrast, mifepristone-treated flies that had lost their IPCs exhibited no oocyte titer change after exposure to yeast (Fig. 5D). This was indicated by detection of 583 +/- 72.6 Wolbachia per oocyte in the mifepristone-control food condition (n = 20), versus 503 +/- 68.0 Wolbachia in the mifepristone-yeast-enriched condition (n = 20) (Fig. 5D). Since mifepristone in combination with the dilp2: Gene-Switch-GAL4, UAS: Reaper system specifically prevented yeast from affecting oocyte Wolbachia titer, this demonstrates that somatic IPCs mediate Wolbachia titer suppression by dietary yeast.
To further investigate the sensitivity of oocyte Wolbachia titer to somatic insulin signaling, we also examined the effect of a sucrose-rich, high sugar diet. High sugar diets have been shown to induce insulin resistance in Drosophila [75,76]. This is may be due in part to increased expression of NLaz [75], which in mammals is known to suppress Akt function within the insulin signaling pathway (Fig. 1) [77–79]. To test the impact of sucrose-enriched diets on oocyte Wolbachia titer, 2-day old D. melanogaster were fed standard food diluted 1/3 with saturated sucrose solution, hereafter referred to as “sucrose-enriched food” (S1 Table). After 3 days of exposure to this diet, Wolbachia titer was assessed in oogenesis. Oocytes from the sucrose-enriched condition exhibited a 2.4-fold increase in Wolbachia (Fig. 6A). Unlike oocytes raised on control food, which exhibited an average of 165 +/- 22.2 Wolbachia (n = 24), D. melanogaster oocytes exposed to sucrose-enriched food exhibited 392 +/- 25.3 Wolbachia (n = 26) (p < 0.001) (Fig. 6A). These data indicate that a high sugar diet significantly elevates oocyte Wolbachia titer, possibly via an insulin-related mechanism.
A sucrose-based impact on oocyte Wolbachia titer is surprising, as corn syrup-enriched food did not induce a similar effect (Fig. 2D). Notably, sucrose is a disaccharide, composed of glucose and fructose, whereas corn syrup consists mainly of glucose. To elucidate the basis for sucrose-induced titer effects in oogenesis, food enriched for glucose and fructose were also tested. However, none of the monosaccharide-enriched conditions significantly affected oocyte Wolbachia titer (Fig. 6B). Control food yielded an average oocyte titer of 478 +/- 27.6 Wolbachia per oocyte (n = 71). Similarly, oocytes in the glucose-enriched condition displayed 520 +/- 31.1 bacteria (n = 33), the fructose-enriched food condition resulted in 478 +/- 33.0 Wolbachia (n = 29), and a mixture of glucose + fructose yielded 499 +/- 28.0 Wolbachia (n = 32). By contrast, oocytes from the sucrose-enriched condition presented 883 +/- 95.4 Wolbachia (n = 22) (p <. 001) (Fig. 6B). This confirms that disaccharide sucrose molecule specifically elicits Wolbachia titer increases in oogenesis.
To further test the possibility that insulin signaling mediates sucrose impact on ovarian Wolbachia titer, we coupled genetic disruptions of the insulin pathway with sucrose-enriched food. Chico is a Drosophila homolog of the Insulin Receptor Substrate that relays signals from the Insulin Receptor to AKT kinase, and thus ultimately TORC1 (Fig. 1) [80,81]. Germline and soma-specific GAL4 drivers were used to drive expression of chico dsRNA [69–72], and oocyte Wolbachia titer was assayed in control and sucrose-enriched conditions. This test did not indicate any effect of germline chico RNAi on sucrose-induced oocyte titer elevation, with sucrose-enriched food corresponding to 2.4-fold higher oocyte titer than the control (Fig. 6C). Germline chico RNAi oocytes exhibited 125 +/- 10.6 Wolbachia when exposed to regular food (n = 26) as compared to 299 +/- 27.2 Wolbachia in response to sucrose-enriched food (n = 19) (p < 0.001) (Fig. 6C). By contrast, somatic chico RNAi eliminated sucrose-induced titer effects in oogenesis (Fig. 6D). Oocytes from somatic chico RNAi flies exhibited 180 +/- 12.9 Wolbachia in the control condition (n = 25), as compared to 169 +/- 12.5 Wolbachia per oocyte in the sucrose-enriched condition (n = 25) (Fig. 6D). Analysis of sibling controls further indicated that the genetic background for the somatic chico RNAi experiment was not responsible for differential oocyte titer responses to sucrose (Fig. 6E). In flies carrying the somatic da-GAL4 driver used for this experiment, the sucrose-enriched condition continued to exhibit 2-fold more Wolbachia than the control food condition. An average of 124 +/- 11.1 Wolbachia were detected in control oocytes (n = 27) as compared to 251 +/- 32.8 Wolbachia detected in oocytes from the sucrose-enriched condition (n = 20) (p <. 001) (Fig. 6E). Though the complete mechanistic implications of somatic chico disruption remain unclear, these data demonstrate that sucrose acts through somatic insulin signaling to elevate oocyte Wolbachia titer.
These data raise the fundamental question of why diet-modulated insulin signaling affects Wolbachia titer so strongly in germline cells. One possibility is that these titer responses are an indirect result of nutrient-induced changes in ovary size and productivity [76]. Yeast-rich diets and insulin signaling are known to drive formation of larger, more productive ovaries [60,76,80,82–91], while high-sucrose diets have the opposite effect [76–79]. To test the contribution of ovary size and productivity variables on oocyte Wolbachia titer, we manipulated ovary productivity by controlling female mating. Mating stimulates ovary development, resulting in a moderately sized, productive ovary. By contrast, virgin females exhibit very large ovaries, filled mainly by mature eggs [92–96]. Oocytes from mated versus virgin females revealed similar oocyte Wolbachia titers, however (S3 Fig). The mated condition displayed 449 +/- 27.5 Wolbachia per oocyte (n = 26), while the virgin female condition that carried 470 +/- 40.6 Wolbachia per oocyte (n = 24) (S3 Fig). These data suggest that ovary size and productivity do not serve as the primary determinants of oocyte Wolbachia titer.
To further investigate the effects of host diet on Wolbachia, we examined Wolbachia nucleoid morphology. Other studies indicate that nucleoid morphology can serve as a proxy indicator of replication-associated changes in cell shape, or stress-induced DNA compaction [97–99]. Multiple, zoomed-in images of Wolbachia stained with propidium iodide were projected as a single image, and nucleoid shape was measured. The images indicated that Wolbachia nucleoid shape differs between nutrient conditions (S4 Fig). To specifically analyze changes in nucleoid length, 120 nucleoids were selected at random from each treatment condition and their lengths were compared. This analysis indicated that 50% of nucleoids in the control condition exceeded 2 μm in length (S4 Fig). The sucrose-enriched condition was similar, with 53% of nucleoids exceeding 2 μm. In the yeast-enriched condition, however, only 37% of nucleoids exceeded this measure (p <. 05). Thus, yeast-enriched food significantly shortened Wolbachia nucleoids. We further determined an elongation index (EI), representing bacterial length divided by width, for the same 120 nucleoids per treatment condition as above. This analysis indicated that 50% of nucleoids measured in the control condition had an EI greater than 2. In the sucrose-enriched condition, only 33% of nucleoids showed an EI greater than 2 (p <. 05). In the yeast-enriched condition, even fewer nucleoids showed this degree of elongation, with only 22% of nucleoids exceeding this EI (p <. 001) (S4 Fig). These data indicate that dietary conditions, and especially exposure to yeast-enriched food, alter Wolbachia nucleoid morphology in oogenesis. This is consistent with a bacterial physiological response to host diet.
The striking impact of dietary nutrients on oocyte Wolbachia titer raises the question of whether Wolbachia titer in other tissues is responsive to nutrient conditions. Wolbachia are present in insect somatic cells, and the Drosophila brain is particularly amenable to assessment of somatic Wolbachia titer [100,101]. To take advantage of this, we imaged Wolbachia in the central brain of D. melanogaster exposed to different nutrient conditions. This analysis revealed that D. melanogaster on control food already carry very low Wolbachia titer in the central brain (Fig. 7A, A’, n = 3), and flies fed with either yeast-enriched or sucrose-enriched food were indistinguishable in appearance from the control (Fig. 7B, B’, n = 3) (Fig. 7C, C’, n = 3). Thus, Wolbachia titer in D. melanogaster brain does not appear to be affected by the dietary conditions used in this study. An alternative possibility, however, is that the overall low Wolbachia titer detected under these conditions hampered our ability to assay nutrient-induced changes in titer.
To pursue this further, the impact of nutrient-altered food was tested in the closely related D. simulans species, known for carrying high Wolbachia titer in its brain cells [101]. Flies exposed to control food exhibited a high titer of Wolbachia in the central brain overall (Fig. 7D, D’, n = 7). Similarly high Wolbachia titer was detected in the brain after exposure to yeast-and sucrose-enriched food (Fig. 7E, E’, n = 5) (Fig. 7F, F’, n = 4). Further quantification of Wolbachia infection frequency did not reveal any differences between nutrient conditions (Fig. 7G). In control food, yeast-enriched, and sucrose-enriched conditions, 55–56% of brain cells exhibited Wolbachia infection (n = 1171, 767, and 665 cells, respectively). No differences were seen in formation of large Wolbachia aggregates either (Fig. 7H). Brain samples reared on control food, yeast-enriched, and sucrose-enriched conditions all exhibited between 16–19 large bacterial clusters per hundred cells. This indicates that Wolbachia titer in the D. simulans brain is unresponsive to the nutrient-altered conditions used in this study.
To address the possibility that D. simulans tissues are generally unresponsive to nutrients, we also assessed D. simulans oocyte titer in response to nutrient-altered food. In contrast to the brain, D. simulans oocytes exhibited a clear nutrient-dependent Wolbachia titer response (S5 Fig). Control oocyte images carried 293 +/- 49.9 Wolbachia (n = 10). By contrast, oocyte titer from the yeast-enriched condition was at 40% of the control level, with an average of 116 +/- 20.1 bacteria detected per oocyte (n = 10) (p = 0.004). Furthermore, the sucrose-enriched condition exhibited 2.3-fold higher titer than the control, with 662 +/- 73.6 Wolbachia detected per oocyte (n = 10) (p = 0.001) (S5 Fig). Thus, D. simulans Wolbachia titers are capable of responding similarly to nutrient conditions as D. melanogaster.
To further probe the impact of host diet on somatic Wolbachia titer, we analyzed relative amounts of Wolbachia versus host DNA in ovarectomized female flies. In this analysis, females were exposed to nutrient-altered diets, dissected to remove ovarian tissues, and analyzed by qPCR. The results indicate the relative level of Wolbachia per host genome copy number. This analysis indicated that yeast-enriched dietary conditions led to higher levels of Wolbachia than the control food condition (Fig. 7I). Control samples exhibited a mean relative level of Wolbachia of 0.989 (n = 37), whereas the yeast-enriched condition displayed a mean relative level of Wolbachia of 1.28 (n = 35) (p < 0.05). Females exposed to sucrose-enriched diets were not significantly different from the control, however, exhibiting a mean Wolbachia relative level of 0.792 (n = 36) (Fig. 7I). This titer response profile differs from analyses of Wolbachia titer in the ovary as well as the brain. This suggests that host diet affects Wolbachia titers in a tissue-specific manner.
As host nutrition has a different impact on ovarian versus somatic Wolbachia titers, this raises the question of what would happen in organism lacking ovarian tissue altogether. To address this issue, qPCR analysis was performed on intact male flies. This indicated that bodywide Wolbachia titer also increases in response to yeast-enriched food, although not sucrose-enriched food (Fig. 7J). The control food condition carried a mean Wolbachia relative level of 1 (n = 16), in contrast to the yeast-enriched condition, which displayed a mean Wolbachia relative level of 1.545 (n = 15) (p < 0.05). Sucrose-enriched diets corresponded to a mean Wolbachia relative level of 1.027 (n = 16). This analysis confirms that the profile of bodywide titer responses in males is equivalent to ovarectomized females. This suggests that somatic Wolbachia titers overall respond to host dietary conditions in a consistent manner.
The major finding of this study is that dietary intake by Drosophila strongly influences Wolbachia titer in the host female germline: a high yeast diet decreases Wolbachia oocyte titer and a high sucrose diet increases Wolbachia oocyte titer. This finding adds to a small but growing literature on the impact of host diet on endosymbionts [1,15,16]. Prior studies of Wolbachia suggest that this endosymbiont relies heavily upon host provisioning of amino acids and carbohydrates [102–104]. A very recent study analyzing the Drosophila midgut and ovary surprisingly indicated that neither dietary yeast nor sucrose had any affect on the Wolbachia:host genomic ratio in those tissues [105]. The image-based analyses of this study demonstrate that yeast and sucrose affect germline Wolbachia titer at the cellular level, however. It is unclear why Wolbachia titer in the oogenesis should be particularly sensitive to diet and whether this is an adaptive response to changes in the host metabolic environment. The evolutionary success of Wolbachia depends on its ability to localize at the posterior pole of the oocyte, the site of germline formation. Significantly, we find that Wolbachia localize to the posterior pole regardless of whether the host is exposed to the low titer, yeast-enriched diet, or the high titer, sucrose-enriched diet. This suggests the previously described microtubule and motor protein based mechanisms driving posterior localization of Wolbachia [38] are robust, even in the face of dramatic titer changes caused by nutrient-altered diets.
Insight into the mechanism of yeast-induced titer suppression comes from our functional studies demonstrating that this response is mediated through TORC1. Genetic up-regulation of TORC1 suppresses oocyte Wolbachia titer, whereas drug-based inhibition of TORC1 increases titer. This finding creates the basis for a sensible functional connection between intracellular Wolbachia and host diet, as both amino acids and insulin signaling are known to drive TORC1 activity [46]. Our finding that BSA-enriched food had no effect on oocyte Wolbachia titer argues that yeast protein content is not the major determinant of germline titer suppression, and alternatively suggests a role for insulin signaling. Prior work has shown that yeast-rich diets trigger insulin signaling in Drosophila, and that Wolbachia interact with host insulin signaling processes [89,106]. Our finding, that loss of somatic IPCs eliminates yeast impact on oocyte Wolbachia titer, confirms that insulin signaling facilitates the titer-suppressing effects of yeast. Furthermore, disrupting the somatic insulin receptor substrate, Chico, suppressed the impact of dietary sucrose on oocyte Wolbachia titer. This suggests that both dietary yeast and sucrose affect germline Wolbachia titer via antagonistic impacts on somatic insulin signaling (Fig. 8).
In considering the mechanism of insulin-based impact on germline Wolbachia titer, one possibility is that changes in ovary productivity are responsible. Diet-modulated insulin signaling affects the relative rates of germline stem cell division, germline cell survival and egg chamber development [60,76,80,82–91]. If Wolbachia are unresponsive to nutrient-induced adjustments in germline cell growth and development, significant titer changes in oogenesis would be expected. However, oocyte Wolbachia titers were very similar in mated and virgin females, despite the different rates of germline stem division expected for each type of flies [76,83,86,88,90–96]. Another possibility is that yeast-induced insulin signaling affects Wolbachia physiology in oogenesis. The “rounded” Wolbachia nucleoids visible in the yeast-enriched condition could indicate substantially slowed bacterial growth or a bacterial stress response, for example [97–99]. Insulin signaling has been shown to induce changes in cytoskeleton organization, proteasome activity and chaperonin activity [107–111], any of which could affect Wolbachia physiology. It is also possible that dietary yeast in particular carries one or more bioreactive agents that are toxic to germline Wolbachia (Fig. 8).
The impact of somatic insulin signaling on germline Wolbachia titer also raises the question of whether somatic Wolbachia titers are similarly affected by host nutrient conditions. Our initial findings that Wolbachia titers in the Drosophila brain are non-responsive to host diet suggested that nutrient-associated titer changes are restricted to the ovary. Analysis of sucrose-fed, ovarectomized females is further consistent with that interpretation. However, analysis of ovarectomized females also indicated that dietary yeast triggers somatic titer changes opposite of oogenesis. It is possible that this occurs by physical relocation of Wolbachia within the body, with dietary yeast driving Wolbachia egress from ovarian cells, followed by invasion of somatic target tissues. Alternatively, host dietary conditions may drive tissue-specific differences in the Wolbachia life cycle. Perhaps yeast-enriched diets favor Wolbachia replication and survival in specific somatic tissues while disfavoring the same in oogenesis. Support for this hypothesis comes from our finding that yeast-enriched food induces the same bodywide titer changes in male flies as seen in ovarectomized females. This demonstrates that ovarian Wolbachia titer responses are distinct from that of other tissues.
The pathways downstream and upstream of TORC1 that mediate yeast-based suppression of Wolbachia germline titer are yet to be determined. An obvious possibility is the role of TORC1 in suppressing autophagy (Fig. 8). There are numerous examples in which autophagy either enhances or suppresses intracellular bacteria titer [112]. Since TORC1 disruptions increase Wolbachia titer in oogenesis, it is possible that Wolbachia interact positively with autophagy, consistent with other endosymbionts [113] [114]. As insulin signaling is expected to down-regulate autophagy (Fig. 1), the low Wolbachia titers seen in yeast-fed oocytes are further consistent with this possibility. However, the finding that dietary yeast also increases somatic Wolbachia titers implies that somatic autophagy is normally bactericidal in that context, consistent with another recent report [115]. These conflicting results may indicate that tissue-specific differences in autophagy regulation contribute to Wolbachia titer control, or that other mechanisms downstream or independent from autophagy are responsible (Fig. 8). Perhaps responses from one or more other TORC1 effectors further contribute to Wolbachia titer regulation (Fig. 1).
Wolbachia have been shown to suppress replication of RNA viruses in insects, including the human pathogens, Dengue Fever Virus and Chikungunya Virus [116–118]. This finding, together with the fact that Wolbachia-induced Cytoplasmic Incompatibility rapid spreads Wolbachia through insect populations [25,119], has led to a novel strategy of combating these diseases by releasing Wolbachia-infected insect carriers of these viruses into afflicted regions [120,121]. Although the mechanism of Wolbachia-induced viral suppression is unknown, several studies demonstrate that the higher the Wolbachia titer, the greater the viral suppression [122–126]. Our finding that host diet dramatically affects tissue-specific Wolbachia titers suggests that the natural diets of the released insects should be taken into account when evaluating the potential effectiveness of a Wolbachia-based viral suppression field study. Finally it will be of interest to determine whether diet has a similar effect on Wolbachia titer in disease-associated filiarial nematodes.
Natural D. melanogaster and D. simulans flies were harvested daily from collection buckets distributed in the Santa Cruz, CA area. As the female flies of these species are morphologically indistinguishable, but both species were well-represented in the area, this wild-caught population was presumed to represent both species. The laboratory strain of D. simulans used was a w- stock that carried the endogenous wRi Wolbachia strain. The D. melanogaster strain used for the initial nutrient feeds and for crossing wMel Wolbachia into the other fly strains was w; Sp/Cyo; Sb/TM6B. Other D. melanogaster fly strains used were the gigas VALIUM20 TRiP line: y, sc, v; P{TRiP.HMS01217}attP2/TM3, Sb; the chico VALIUM20 TRiP line: y, sc, v; P{TRiP.HMS01553}attP2/TM3, Sb; the somatic daughterless driver: w; P{w+, GMR12B08-GAL4}attP2; the germline triple driver: P{otu-GAL4::VP16.1}; P{GAL4-Nos.NGT}40; P{GAL4::VP16-Nos.UTR}MVD1; and the stocks used for IPC ablation: w; P{w+, dilp2::GS-GAL4}/Cyo, and w; P{w+, UAS::Reaper}. During this work, wMel was introduced into the somatic daughterless driver, the germline triple driver, and the dilp2::GS-GAL4 driver, and the infected versions of these stocks were crossed to the TRiP or UAS:Reaper responders. DrosDel isogenic flies carrying wMel were used for real-time quantitative PCR analyses [122].
The standard food recipe used was based upon that of the Bloomington Drosophila Stock Center [127]. The food was prepared in large batches that consisted of 20L water, 337g yeast, 190g soy flour, 1325g yellow corn meal, 96g agar, 1.5L Karo light corn syrup and 94mL propionic acid. To create yeast paste for this study, live bakers yeast was mixed together with water to create a smooth, thick paste. To create the “control food” used in this study, we mixed together 1.5mL ddH2O and 3.5mL of melted standard food in a narrow-mouthed vial, then let cool in an ice bucket to solidify the food suspension. The same procedure applied to creation of all other nutrient-altered foods used in this study. For “corn-syrup-enriched” food condition, 1.5mL Karo light corn syrup was used. For “yeast-enriched” food condition, 1.5mL of heat-killed yeast paste was used. The “BSA-enriched” food carried 0.4g BSA, 1.5mL water, and 3.5mL standard food. For the “sucrose-enriched”, “glucose-enriched” and “fructose-enriched” foods, fresh sugar solutions were prepared at a final concentration of 1g/mL, then 1.5mL of this concentrate was combined with 3.5mL standard food for each vial. The “glucose + fructose enriched” condition carried 0.75mL of 1g/mL glucose, 0.75mL 1g/mL fructose, and 3.5mL standard food. Alternate methods were used to prepare food for the other treatments. For the branched chain amino acid condition, the control condition contained 400μL water and 50μL DMSO mixed with 4.5mL standard food, whereas the experimental condition carried 200μL of 1mg/mL Arginine, 200uL of 1mg/mL Isoleucine and 50μL DMSO mixed with 4.5mL standard food. For the TORC1 testing, 50μL of either control DMSO or 30mM rapamycin/DMSO stock was mixed into 5mL standard food. For tests of IPC function, 50μL of either control DMSO or a 10mM mifepristone-DMSO stock was mixed into 5mL standard food.
Laboratory Drosophila stocks were maintained on standard food at 23–24°C. Identical population density was used in all vials, and control and experimental conditions run in parallel. Flies of the genotype w; Sp/Cyo; Sb/TM6B were used in all imaging experiments that assessed nutrition as the only variable. In the cases where crosses were needed to drive expression from TRIP line stocks or the dilp2:GAL4 stocks were used, we performed all crosses using identical population density and female age distribution in all vials, with control crosses always run in parallel. Virgin female flies were collected during the first 3 days of eclosion only, then subjected to nutrient conditions. The procedure was to collect a range of 0–24 hour old adults, age these young flies for 2 days on standard food, and expose to treatment conditions for 3 more days. The mixture of D. melanogaster and D. simulans flies collected from nature likely varied in age. These flies were also exposed to standard food for 2 days, and transferred to experimental food for 3 days. In the case of IPC ablation, the collected flies were allowed to mature 2 days, then transferred to mifepristone-containing food or DMSO control food. The flies were maintained on this food for 14 days, transferring the population to a fresh vial every 3 days of the treatment period. After this was completed, the flies were exposed to nutrient-altered food for 3 days.
Samples were prepared from a minimum of 10–15 flies per condition in each replicate. Ovary dissection, fixation, and propidium iodide staining were done as previously described in order to label germline Wolbachia nucleoids [38]. Ovarian tissues for all samples in each replicate were mounted on slides in parallel to ensure maximal consistency in sample compression between slide and coverslip. All samples were then imaged on a Leica SP2 confocal microscope at 63X magnification with 1.5X zoom. Experimental samples verified to exhibit the same degree of compression as the control sample were pursued further, while any experimental samples deviating from that were discarded. Z-series images were acquired from each egg chamber of interest at 1.5 μm intervals. Uniform intensity settings were applied to all egg chambers imaged within each replicate. A minimum of 7–10 oocytes were ultimately imaged from each condition, with all experimental oocytes matched for morphological consistency against control oocytes of the same replicate. Using this rigorous method, significant fold-differences in Wolbachia titer were consistently identified between control and experimental conditions, regardless of the baseline quantity of Wolbachia detected in each replicate.
To quantify Wolbachia titer in the confocal images, we used established methods to identify the deepest possible focal plane where Wolbachia are clearly visible in all samples tested for each replicate [32]. The images were processed in Photoshop to remove everything from the images except oocyte Wolbachia, which were then quantified using the Analyze Particles feature in Image J. This analysis ultimately quantifies the Wolbachia nucleoids carried per oocyte, or per nurse cell, within a single, representative focal plane of each egg chamber. Although the graphical data displayed in the figures present all experimental averages as normalized against the control averages, all statistical calculations were run by comparing each condition only against controls that were run in parallel. Significant differences were indicated by ANOVA. A minimum of 2–3 replicates were performed for most germline staining experiments described in this study. The only exception was the experiment in which Wolbachia titer responses were analyzed in both brain and ovary tissues. In that case, single replicates were done for each type of tissue stained, with all conditions run in parallel.
To analyze Wolbachia titer by real-time quantitative PCR, single flies were homogenized with a pestle in 250 μl of Tris HCl 0.1M, EDTA 0.1M and SDS 1% (pH 9) and incubated for 30 minutes at 70 ºC. After 35 μl of KAc were added the sample was incubated 30 minutes on ice, centrifuged for 15 minutes at 13.000 rpm at 4ºC and the supernatant stored. Samples were diluted 100x for qPCR. qPCr was performed as described previously [122], using the CFX384 Real-Time PCR Detection System and iQ SYBR Green Supermix (both BioRad). The relative amount of Wolbachia was calculated with the Pfaffl method [128], using the primers for the gene wsp to determine Wolbachia DNA levels and primers for host Rpl32 and Actin5C genes to normalize male and female samples, respectively [122]. Data from males were analyzed using a linear model on the log of the relative wsp levels (Im in R) [129]. Data from females were analyzed using a mixed linear model on the logs of relative wsp levels (lmer in R).
To analyze Wolbachia in the Drosophila central nervous system, brains were dissected and fixed as previously described [101]. Brains were incubated in anti-rabbit wsp antibody + PBST (0.1% Triton X-100) for 4 hours at room temperature or at least 12 hours at 4 degrees. For secondary antibody staining, goat anti-rabbit Alexa Fluor 546 (Invitrogen) was used at room temperature or at least 12 hours at four degrees. Actin labeling was done with phalloidin conjugated to Alexa 488, diluted 1:100 in PBST, for one hour at room temperature. Brain tissues were imaged on a Leica SP2 confocal microscope at 63X magnification. Brains were quantified with Leica LAF AS software. One representative focal plane per brain was scored. Cells containing one or more Wolbachia were scored as infected. Wolbachia aggregates larger than 10 microns2 were scored as a “cluster” [101].
To assess Wolbachia nucleoid shape, we acquired Z-series images of stage 10A oocytes at 63X magnification with 5X zoom. Then we created a projection of 4 images from each Z-series, located just beneath the follicle cell layer, and measured the length of individual nucleoids using the “line” tool located within the Profile function of Quantification Tools in the Leica SP2 software. Elongation index was calculated as a function of length divided by width. It is assumed that the bacteria are random in orientation, and thus detecting a range of nucleoid morphologies ranging from spherical to rod-shaped is possible. Chi square tests were used to compare Wolbachia length and elongation index exhibited by bacterial populations from each treatment condition.
|
10.1371/journal.pcbi.1004711 | Erosion of Conserved Binding Sites in Personal Genomes Points to Medical Histories | Although many human diseases have a genetic component involving many loci, the majority of studies are statistically underpowered to isolate the many contributing variants, raising the question of the existence of alternate processes to identify disease mutations. To address this question, we collect ancestral transcription factor binding sites disrupted by an individual’s variants and then look for their most significant congregation next to a group of functionally related genes. Strikingly, when the method is applied to five different full human genomes, the top enriched function for each is invariably reflective of their very different medical histories. For example, our method implicates “abnormal cardiac output” for a patient with a longstanding family history of heart disease, “decreased circulating sodium level” for an individual with hypertension, and other biologically appealing links for medical histories spanning narcolepsy to axonal neuropathy. Our results suggest that erosion of gene regulation by mutation load significantly contributes to observed heritable phenotypes that manifest in the medical history. The test we developed exposes a hitherto hidden layer of personal variants that promise to shed new light on human disease penetrance, expressivity and the sensitivity with which we can detect them.
| A central goal of personal genomics is to interpret an individual’s genome to identify variants that confer disease risk, an aim that has far-reaching implications for personalized, precision medicine. Here, we leverage next generation sequencing, health records, and functional genome annotations to develop statistical methods that predict disease risk from a single genome. Motivated by the fact that about 90% of genome-wide association study disease-associated variants lie in the non-coding genome, we identify personal variants that mutate conserved transcription factor binding sites. To identify if such non-coding personal variants collectively dysregulate a key biological process, we employ the enrichment analysis tool GREAT to identify if a person’s noncoding mutations are over-represented in the regulatory domains of genes involved in a common biological pathway. Notably, in five unrelated genomes we analyzed, the most statistically significant, seemingly dysregulated pathway is indicative of that person’s medical history, ranging from neuropathy to heart disease. Statistical analysis confirms that associations from our predicted pathway to an individual’s medical record are rigorous and significant in the context of the un-phenotyped, race-matched 1,000 Genomes cohort. As such, we present a novel method that leverages the contribution of multifactorial non-coding variation to predict disease risk in individual genomes.
| The advent of high-throughput genotyping spurred the rise of genome-wide association studies (GWAS) aimed at identifying the basis of genetic diseases. GWAS variants, over 90% of which have been found to localize outside of protein-coding sequences [1], and the growing body of non-coding genome annotations have helped improve our understanding of the genetic basis of diseases by shifting the focus from protein coding and copy number variations [2–4], to the non-coding genome. Though GWAS have been instrumental in suggesting a gene regulatory component to human disease susceptibility [5,6], they have been plagued by the “missing heritability problem”, which observes that loci detected by GWAS in general only explain a small fraction of the genetic variance responsible for phenotype [3,7].
Suggested models of genetic variance responsible for the “missing heritability problem” include “the infinitesimal model”–a large number of small effect common variants and “the rare allele model”–a large number of large-effect rare variants [7]. In the case of the infinitesimal model, the missing heritability can be explained due to additive or epistatic interactions between variants rather than independent polymorphisms [8]. But, selecting and evaluating all sets of variants results in a combinatorial explosion of sets that we are currently statistically underpowered to evaluate.
In this work, we will show how to not only successfully avoid the combinatorial explosion, but also simultaneously address the crucial role of additive and epistatic noncoding variation in human disease. Specifically, we develop a novel statistical framework to identify putatively deleterious noncoding variation in personal genomes that en masse, confers disease risk by dysregulating key genes involved in a common biological process.
A central role of the non-coding genome lies in cis-regulation of gene expression. GREAT (Genomic Regions Enrichment of Annotations Tool) is a tool commonly used to address the functional enrichment of a set of cis-regulatory genomic regions [9]. GREAT tests whether an arbitrary set of genomic regions, most of which are thought to regulate the expression of nearby genes, congregate next to genes of particular functions or pathways. GREAT assigns different genes variable length gene regulatory domains, accounts for distal regulatory elements and rewards observing multiple elements next to the same gene–reflecting observed properties of vertebrate gene regulation. GREAT has been shown to be superior to gene based tests (following the one probe—one gene paradigm of transcript analysis) in analyzing different types of ChIP-seq and related data [9].
Since we are interested in identifying disease-associated noncoding variation to interpret personal genomes, we asked whether disease-associated noncoding mutations would be functionally enriched for key biological pathways using GREAT. First, we subjected noncoding GWAS significant SNPs associated with several phenotypes ranging from Crohn’s disease to fasting glucose traits to GREAT (S1 Table). Not all GWAS tag-SNPs are themselves causal, but because they lie in proximity to the causal mutation, we can assume that GREAT will in most cases associate the tag-SNP with the same affected gene/s it would associate the underlying causal mutation. For example, if we subject the 40 non-exonic non-linked GWAS SNPs associated with cholesterol levels to GREAT analysis, the topmost enriched term (P = 3 x 10−5) in the entire GO ontology is genes involved in “abnormal circulating cholesterol level”. We show similar results across several different GWAS sets in S1 Table. In each case, we see that the non-linked tag-SNPs are most enriched next to genes of functional categories strikingly relevant to the assayed phenotype, providing in-silico assurances for the quality of the study and the validity of a GREAT analysis, but also suggesting that multiple of these mutations may accumulate in afflicted individuals. As such, we hypothesized that much more signal may hide in cis-regulatory variants beyond what GWAS may reveal.
The coherence of target gene enrichment for GWAS variants suggests additive and/or epistatic effects of variations to confer phenotype. Modeling such interactions is generally limited to heuristic search of pairs due to the high computational requirement and lack of statistical power [10]. The statistical power for identifying causal variants is further weakened in non-coding regions due to most variants resulting from neutral evolution of the genome [11]. Thus, to obtain a high quality set of variants on which GREAT can be applied we require a method of obtaining a set of functionally relevant noncoding variants without enumerating all possible sets.
To obtain a set of functionally relevant, putatively deleterious noncoding variants, we make use of transcription factor (TF) binding site prediction. Novel high throughput technologies, such as HT-SELEX and Protein Binding Microarrays, are revealing the precise DNA binding preferences of the majority of human transcription factors [12,13]. Using these preferences to predict TF binding in a single genome is notoriously hard. However, if one is willing to predict only a subset of binding sites, namely those conserved through evolution, one can then predict the existence of a binding site only if one sees the site in orthologous locations in a number of different mammals [14]. Such a scheme will naturally miss many evolutionarily newer binding sites, but, as we and others have shown, those conserved binding sites that we do predict are predicted with great precision and are useful for downstream analysis such as functional enrichment and protein complex prediction [14–16].
As shown in the GREAT paper, while a ChIP-seq experiment reveals that a TF binds non-specifically to many genomic locations, the strongest GREAT gene enrichment reflects the process or function the TF is regulating, highlighting the subset of binding sites involved in the regulatory process [9]. Previously, in our binding site prediction (PRISM) paper, we predicted the conserved subset of binding sites of a given TF motif and subjected this set, in place of ChIP-seq peaks, to GREAT analysis. In many cases, such as for transcription factors REST, GABPA, SRF, and STAT3, such analysis revealed multiple functional contexts in which the TF was involved without requiring a cell-type matched TF ChIP-seq experiment [14].
Additionally, in our previous binding site prediction work, we intersected our conserved binding site predictions with GWAS tag SNPs. To maximize the chance a GWAS tag SNP was indeed the functional, causal mutation, we set out to search for the following: A GWAS tag SNP overlapped by a conserved binding site prediction, such that: 1) the two observed alleles significantly differed in the predicted TF’s ability to bind to the motif, and 2) the TF we predict to bind has been previously implicated in the GWAS phenotype. In our paper, we highlighted only five such predictions (Table 1 in [14]). One striking example, in the context of prostate cancer, is our prediction that a GWAS risk allele at 6q22 modifies the conserved binding site of HOX13, thus modifying the expression of the downstream RFX6 gene. Our prediction was later beautifully experimentally validated by Taipale and colleagues, illustrating the utility of PRISM predictions in assessing the impact of noncoding variation on disease [17].
As exemplified above, the confluence of binding site prediction with PRISM and functional assessment of cis-regulatory regions with GREAT suggests a potent combination to understand the role of noncoding variation in disease. Accordingly, in this study we looked at personal genomes, tallied all the locations where the individual carries a SNP that disrupts an evolutionarily conserved binding site, and asked (using GREAT) which biological function or process these mutations aggregate next to most. Guided by our hypothesis that a pathway with the most unexpected mutational load may contribute to a person’s medical history, we then assessed our pathway predictions for relevance to the person’s health record.
Using a large library of unique high quality binding motifs for 657 different transcription factors, covering all major human DNA binding domain families and a multiple alignment of 33 primates and mammals, we first predict cross-species conserved binding sites present in the reference human genome (see Materials and Methods). We then examine the genetic variants of a human individual against the reference genome. We focus on the subset of variants (heterozygous or homozygous) that overlap conserved binding site predictions. From these, we pick only variants where the human reference base is identical to its chimpanzee orthologous base (and thus most likely ancestral), and the individual variant base differs from both. Finally, of these we keep only the binding sites where the individual (derived) variant is predicted to significantly decrease binding affinity compared to the ancestral base–we call these conserved binding site eroding loci, or CoBELs (see Fig 1 and Materials and Methods).
We downloaded from UCSC whole genome variant files for all four individuals for whom public medical history summaries are also available: Stephen Quake [18], and three individuals from the personal genome project (PGP10) [19]. An additional file was obtained for James Lupski [20]. We then compared each separately to the reference genome to obtain 6,321 CoBELs for Stephen Quake, 5,291 for George Church, 5,775 for Misha Angrist, 5,861 for Rosalynn Gill, and 6,447 for James Lupski (S4–S8 Tables).
Because CoBELs weaken conserved ancestral binding sites, we asked whether an individual’s set is found preferentially next to genes encoding any particular function, and if so, whether this function relates to the individual’s medical history (Fig 1C). GREAT, as described earlier, is an approach devised specifically to assess enriched functions within a set of genomic regions thought to regulate the adjacent genes [9] by associating with each gene in the genome a variable length regulatory domain, bracketed by its two neighboring genes. GREAT also holds a large body of knowledge about gene functions and phenotypes–here we use over 1.1 million such gene annotations (see Materials and Methods). For a given set of CoBELs, GREAT iterates over 16,000 different biological functions and phenotypes, asking whether CoBELs are particularly enriched in the regulatory domains of genes of any particular function. For example, 33 genes in the human genome are annotated for “abnormal cardiac output”. Their GREAT assigned regulatory domains cover 0.45% of the genome. Of the 6,321 Quake CoBELs, 28 (0.45%) are expected in the regulatory domains of these 33 genes by chance, but 57 CoBELs, over twice as many, are in fact observed. To determine statistical significance, GREAT computes two statistics for this enrichment, and corrects them for multiple hypothesis testing (see Materials and Methods).
Prominent in Stephen Quake’s medical records is a family history of arrhythmogenic right ventricular dysplasia/cardiomyopathy, including a possible case of sudden cardiac death [18]. Strikingly, when Quake’s set of CoBELs is analyzed using GREAT, the top phenotype enrichment (using default parameter settings, optimized for inference power in the original GREAT paper [9]) is “abnormal cardiac output” (57 CoBELs, false discovery rate Q = 1.69 x 10−4). This enrichment is suggestive of susceptibility to heart diseases responsible for reduced cardiac output [21]. Meaningful associations between CoBELs and personal medical records are in fact observed for all five genomes (Table 1 and S9–S13 Tables).
The top enrichment for George Church, who suffers from narcolepsy, is “preganglionic parasympathetic nervous system development” (23 CoBELs, Q = 1.18 x 10−4). The autonomic nervous system is strongly suspected to be involved in narcolepsy [22]. Misha Angrist, whose personal reporting indicates possible keratosis pilaris, a follicular condition manifested by the appearance of rough, slightly red, bumps on the skin, has “epithelial cell morphogenesis” as his top biological process enrichment [23] (60 CoBELs, Q = 1.38 x 10−5). For Rosalynn Gill, who suffers from hypertension, the top enriched phenotype is “decreased circulating sodium level” (32 CoBELs, Q = 4.94 x 10−6). Sodium intake is strongly associated with hypertension [24]. Intriguingly, the top biological process enrichment we obtain for James Lupski, whose family has a history of axonal neuropathies in the peripheral nervous system (PNS) [20], is “regulation of oligodendrocyte differentiation” (59 CoBELs, Q = 2.93 x 10−5). Oligodendrocytes are the neuroglia that create the myelin sheath around axons in the central nervous system (CNS) and maintain long-term axonal integrity [25,26].
While a statistically significant functional enrichment from GREAT rejects the null hypothesis of uniformly random distribution of the CoBELs in the regulatory domains of the function-associated genes, it does not check whether there is an inherit bias in the distribution of conserved binding sites (eroded or not) in the regulatory domains of genes involved in the enriched functions. Thus to further assess the significance of our results we replaced every CoBEL with a random binding site prediction for the same transcription factor of same affinity and similar cross-species conservation. Using 10,000 random control sets, the likelihood of obtaining the functions reported in Table 1 as top prediction due to bias in the distribution of binding sites in the genome is low (Quake P = 3 x 10−4, Church P = 5.7 x 10−3, Angrist P = 4.8 x 10−3, Gill P = 1 x 10−4, Lupski P = 1.9 x 10−3, and combined P = 1.6 x 10−15). Significance remains high when we relax the requirement to recover each exact same term with matching any one of a broader group of 12–60 related functions as a top prediction (Quake P = 1.1 x 10−3, Church P = 1.3 x 10−2, Angrist P = 7.7 x 10−3, Gill P = 7.4 x 10−3, Lupski P = 6.5 x 10−3, and combined P = 5.2 x 10−12; see Materials and Methods).
While phenotypic data is not available for the 1,000 genomes project subjects [27], the availability of whole genome sequences for 1,094 individuals allows us to ask how unique are our top predictions for the five phenotyped individuals against a large background of controls. We asked whether the phenotype predictions were unique to a given personal genome by testing whether they rarely appeared in control individuals from the 1,000 genomes project, thereby testing the specificity of our screen. This control analysis was performed due to the inclusion of both common and rare variants in our analysis. We wanted to verify that enrichments observed in our five genomes were not dominated by common CoBELs shared with many other individuals.
Thus, we computed the frequency of the observed enrichments in all control, un-phenotyped 1,094 genomes sequenced by the 1,000 genomes project [27]. We verified the CoBEL set size of the 1,094 genomes were comparable to those of the five analyzed genomes (min 6,121; European median 6,385), submitted the CoBELs to GREAT and noted top enrichments. Each one of our observed top enrichments for the five individuals had an occurrence rate less than 0.05 (S2A Table) and the enrichment’s p-value and fold statistics placed them as significantly removed from the 1,000 genomes cohort (Fig 2). Next, we performed PCA to verify that the five genomes analyzed in our study are both predominantly of the expected (European) ancestry, and not an outlier compared to the 1,000 genomes project data (Fig 3A). We then recomputed the occurrence rate for the enrichments using only the 381 European genomes and only the 181 admixed genomes to correct for any population specific enrichments. Again, all the enriched terms had an occurrence rate less than 0.05 (S2A Table). Since ontology terms in GREAT are related in a directed acyclic graph (DAG) structure, terms such as “abnormal cardiac output” (the Quake genome prediction) share similar gene sets to their umbrella term “abnormal cardiovascular output”, which a control patient from the 1,000 Genomes project may exhibit. To account for the case when two such related terms are predicted, we calculated the false discovery rate for a term by counting its broader group of related functions as well. Still, the occurrence rate for the findings remained less than 0.05 (S2B Table) when we repeated both the full 1,094 genomes, 381 European genomes and 181 admixed genomes calculations for the broader group of related functions, except for slightly higher p-values (up to 0.088) for the more common heart and hypertension disorders. Indeed, 8% of the un-phenotyped 1,000 genome subjects (who may themselves suffer or be predisposed to various complex diseases, especially the more common ones) had a top enrichment in the broader set of terms associated with hypertension, and 5% were similarly most enriched for a heart term.
Finally, we assessed the specificity of associating the CoBEL enrichments of five individuals with their medical histories (Fig 1C and S14 Table). This test was performed to verify that the predicted top enrichments were not so broad that they would match different medical histories and likewise that the individuals selected did not have such a broad range of disease phenotypes as to match different possible top enrichments. We defined an association matrix linking enrichment and medical history, with the phenotypes observed in the five individuals as rows, and top enriched terms in all as columns. A cell in the matrix would be marked “true” only where the enriched term (of any individual) is thought to be related to the etiology of the phenotype (of any individual; see Materials and Methods). One instance of this matrix was filled by a medical doctor based on their medical knowledge and training (S15 Table) and another instance was independently filled using a literature survey (S16 Table). The objective was to compute the chance of associating a set of five individuals with random medical histories with the observed enrichments using one of the two association matrices as the “gold” association. We generated 1,000 sets of five individuals with random medical histories composed of similar disease profiles and assessed the likelihood of being able to associate them with enrichments (see Materials and Methods). Successfully linking five random individuals with enrichments was highly significant using the association matrix generated by the medical doctor (P = 3.0 x 10−3) and by the matrix generated by literature survey (P = 3.0 x 10−2) suggesting our links between enrichment and medical histories are not just a function of the listed histories. The literature survey derived association matrix potentially offers a stricter null model since it includes associations that are currently research topics hinting at associations that may or may not become clinically relevant in the future.
Our CoBEL predictions are distinct from known GWAS associations. The 238 variant alleles that underlie all Table 1 predictions overlap a single, phenotype irrelevant, GWAS SNP, suggesting our method as a complementary method to discovering disease loci. While GWAS aims to find loci most likely to individually distinguish disease cohorts from matched controls, our method tries to identify the sum of both common and rare loci that can contribute to disease. GWAS is underpowered to find such combinatorial interactions. Similarly, none of the CoBELs responsible for the 238 variants intersect with a HGMD [28] disease variant (a large set of very rare, highly penetrant variants thought to individually trigger the underlying disease). When the overlap analysis is extend to include GWAS SNPs in possible linkage disequilibrium (LD), only two possible phenotype matches arise: “cardiac hypertrophy” associated [29] SNP rs3729931 for Quake, and “multiple sclerosis” (another demyelination disease [26]) associated [30] SNP rs882300 for Lupski. Indeed, nearly half the total number of CoBEL variant alleles we predict (7,115, 49%) are unique to only one of our five individuals. Similarly, for each of the five top function predictions in Table 1, of sixteen possible subsets (CoBELs shared or not with each of the other four individuals), the biggest contribution (17–34%) always comes from private sites (S1 Fig).
When the CoBEL frequencies are examined at the population level, Quake and Gill’s enriched CoBELs show higher population frequencies (Fig 3B and 3E) for their presumably more common enriched phenotypes of heart disease and hypertension. Conversely, Church, Lupski and even Angrist to a lesser extent, show more enriched CoBEL with low population frequencies (Fig 3C,3D and 3F). To examine the population frequency dependence of the CoBEL analysis, we restricted ourselves to rare CoBELs, defined as those with frequency less than or equal to 0.01 in the 1,000 genomes. None of our functional enrichments are significant for the rare CoBELs. Even when we increase the 1,000 genome frequency 10-fold to 0.1, only Angrist’s “epithelial cell morphogenesis” enrichment is rescued, albeit with diminished enrichment statistics (16 CoBELs, Q = 1.85 x 10−2) compared to the full set (60 CoBELs, Q = 1.38 x 10−5). This further corroborates that our enrichments are a combination of both common and rare variants.
The screen we perform is underpowered: we do not have the binding affinities of all human transcription factors or all functional (ancestral or not) binding sites; variant mapping may miss more complex gene regulatory mutations; and in particular our knowledge of phenotype to gene associations is far from complete. Additionally, we focus only on the top enrichment obtained rather than all enrichments to maintain the ability to test for statistical rigor of the associations. All these limitations, however, only reduce our power to detect true associations, but do not elevate the likelihood of false predictions. In contrast, by focusing on deeply conserved binding sites, we greatly increase the likelihood that their disruption carries a fitness cost. Indeed, considering that GREAT tests over 16,000 different biological processes or phenotypes (from “abdominal aorta aneurysm” to “zymogen granule exocytosis”), the links we obtain between genomic prediction and medical phenotype seem highly significant.
Our CoBEL predictions compliment known disease alleles. For example, a particular human leukocyte antigen (HLA) allele is found in a vast majority of narcolepsy patients who suffer from cataplexy, and is also common in narcolepsy patients who do not [31]. The affected Church genome is homozygous for a different HLA allele (see Supplementary Methods). Four GWAS SNPs, all with modest effect size (OR = 1.29–1.79) are currently associated with narcolepsy. Church carries two of these, but the other four unaffected genomes we analyze each carry 2–3 narcolepsy risk alleles as well, due to their common prevalence (see S3 Table).
The Quake genome was previously analyzed for coding and GWAS variants [18]. While no single strong mutation emerged, the sum of collected mutations was enough to assess heart disease as a relatively large risk. The evaluation process of the many personal variants however was biased towards genic variants and previously determined risk loci with a focus on explaining the family history of heart disease. The enrichment we obtain for cardiac output not only comes from novel, non-genic loci, it is also obtained in a completely agnostic fashion.
Our analysis is complementary to state of the art analyses that focus on searching for the primary disease causing variant by intersecting with known (predominantly coding) variant databases, exploring rare or novel coding or splicing variants in known disease associated genes and prioritizing coding candidate variants using computational tools such as SIFT [32], PolyPhen2 [33] and VAAST [34]. Few such tools exist for the non-coding genome, none of which to the best of our knowledge focuses explicitly on binding site disruption. Methods such as CADD [35] score the pathogenicity of non-coding variants, but train their model on positive sets only weakly enriched for deleterious non-coding mutations. Because the non-coding portion of the genome is so large (97%), and because most such tools do not aggregate mutations on functional or any other categories, most usage is restricted to splice variants or non-coding RNA. This is exemplified by the genome analyses performed by Lupski et al. [20] and Ashley et al. [18], for Lupski and Quake, respectively. Both works focused primarily on coding variants in known disease associated genes. They identified non-synonymous SNPs and searched for matches in known pathogenic variant databases such as HGMD [28] and OMIM [36]. When known disease variants were not identified, the search was expanded to include rare and novel variants in genes relevant to their patient (neuropathies in the case of Lupski et al. [20] and cardiovascular disease in the case of Ashley et al. [18]). Neither study pursued any potential gene regulatory mutations.
In addition to the enrichment obtained by our analysis, the accumulation of binding sites in our top enrichments is also revealing: First, each target gene in Table 1 is affected, on average, by more than three CoBELs, chipping away at the gene’s presumed regulatory robustness [37]. Second, Table 1 also shows that in all five cases, CoBELs affect a majority (58–89%) of all human genes annotated for said function/phenotype.
Together, our observations suggest the gradual erosion of gene regulation over both (human generation) time and (gene regulation) space, ultimately manifesting as medical history. These observations corroborate a long held notion that lineage accumulation of small deleterious mutations, even when combined with different lifestyles and environments, ultimately increase the likelihood of familial disease phenotypes [38]. Depending on the selection coefficient of these deleterious mutations and their genetic background, these mutations may eventually be swept out of the population, but are currently visible due to nonrandom human mating patterns and the relatively short timescales since erosion.
Our screen provides an exciting glimpse of the latent genetic load of human gene regulation contribution to personal medical histories. As our ability to characterize individual genetic load improves, so will our understanding of genome–environment interactions, and the thresholds that are crossed to trigger onset of human disease.
Our transcription factor binding motif library, represented as a position weight matrix (PWM), contains 917 unique high quality monomer and dimer motifs for 657 transcription factors from the UniPROBE [39], JASPAR [40], and TransFac [41] databases, secondary UniPROBE motifs, motifs from published ChIP-seq datasets and from other primary literature [16]. We included both monomeric and dimeric (where the TF complexes either with itself or with another TF) motifs to improve our sensitivity since previous work has found that complexes tend to have modified binding affinities [16].
We downloaded variant calls mapped to the human reference assembly hg19 (GRCh37) from the UCSC genome browser [42]. The tables were pgQuake for Stephen Quake, pgChurch for George Church, pgAngrist for Misha Angrist and pgGill for Rosalynn Gill. The variants for James Lupski were downloaded from dbSNP [43] and processed to remove non-single nucleotide polymorphism and those that had ambiguous mapping to the reference genome. The medical history summaries for Stephen Quake and James Lupski were obtained from Ashley et al. [18] and Lupski et al. [20], respectively. Medical history summaries for the remaining individuals were obtained from their public profiles on the Personal Genome Project [19] website.
We identified conserved binding sites using the UCSC human reference assembly hg19 (GRCh37) based multiple alignment of 33 primates and mammals [42]. Binding site prediction was done by identifying conserved binding site matches using PRISM [14]. We chose the default PRISM thresholds of a minimum of five species preserving each site prediction, with the total phylogenetic (neutral) branch length [44] of the preserving species amounting to two substitutions per site or more. Additionally, we kept only the top 0–5,000 binding site predictions that had a conservation p-value less than or equal to 0.05. The conservation p-value was computed by comparing the binding conservation for (CpG preserving) shuffled versions of the motif in similarly conserved regions of the genome. All parameter settings we used have been previously optimized in the PRISM paper for predictive power [14], including against multiple ENCODE [6] datasets.
Next, we identified all the heterozygous or homozygous variants in an individual genome where the human reference (hg19) base is identical to the orthologous chimp (panTro2) base, and thus most likely human ancestral. We then found all human reference genome conserved binding sites affected by our individual specific variants. Of these we kept only sites where replacing the reference human (ancestral) base(s) with the individual derived variant(s) lowers binding affinity by five per cent or more. Binding affinity was computed using the MATCH scoring scheme [45]. Overlapping binding sites were combined to obtain our final set of conserved binding site eroding loci (CoBELs).
Define set of human conserved binding sites, TFBS ← PRISM(motif library)
For each individual with genomic variants Vi:
Intersect TFBS with Vi (using overlapSelect from UCSC genome browser)
For each TFBS tfbs in intersection:
Compute tfbs MATCH score difference between ancestral and variant D
If D decreases the MATCH score more than 5% add binding site to set of CoBELs
Run GREAT(set of CoBELs) -> Output Top Enrichment
The majority of the motifs used in our screen are public, and can be obtained, along with their predictions, from the PRISM website at PRISM.stanford.edu. A small fraction of motifs comes from the proprietary Transfac database. A list of these will be provided upon request. We also include all five CoBEL sets in S4–S8 Tables, which can be processed using GREAT at GREAT.stanford.edu to reproduce the results of Table 1 and S9–S13 Tables.
Each set of CoBELs was submitted to GREAT (for Genomic Regions Enrichment of Annotations Tool) v2.0.2 [9]. As explained in the main text, GREAT searches for statistically significant genomic regions (in this case CoBELs) accumulation in the regulatory domains of genes that share the same annotation. For this study, we used GREAT’s default regulatory domain definition: a constitutive 5,000 bases upstream and 1,000 bases downstream of a gene’s canonical transcription start site (TSS), extended up to the constitutive regulatory domain of the adjacent genes on either side, or up to one million bases. Significance was also defined using the default GREAT thresholds: 0.05 FDR threshold for both binomial and hypergeometric test and binomial fold greater than 2. These parameter settings have all been optimized for inference power in the original GREAT paper [9]. We queried the GO Biological Processes [46] and MGI Phenotype [47] ontologies allowing GREAT to test for possible enrichment of any of 16,054 different functions, using 1,140,682 gene to function mappings.
The CoBEL methodology was applied to each of the 1,094 genomes and the top enrichment satisfying the default GREAT filters in the GO Biological Processes and MGI Phenotype ontologies was tracked. For each of the enrichments highlighted for the five genomes analyzed in this report, the frequency of the enrichment in the full 1,094 genomes was computed. Additionally, the frequency of the enrichments in the 381 European (EUR) subset and 181 admixed (AMR) subset was measured since principal component analysis revealed that the five genomes analyzed in this report are closest to these two population subgroups (Fig 3A).
All SNPs from the NHGRI GWAS catalog [48] were downloaded from a build containing 8,967 records in hg19 (GRCh37) co-ordinates, and intersected with the set of enriched CoBEL variant alleles from Table 1. Quake, Angrist, Gill and Lupski had no overlaps. Church had a single, phenotype irrelevant, overlap with rs10808265 which is GWAS associated with pulmonary function decline [49].
To assess linkage disequilibrium (LD) between the enriched CoBEL variants and GWAS SNPs we used HapMap [50] rel27 LD data for the CEU (Utah residents with Northern and Western European ancestry) population. CoBEL variant alleles from Table 1 were mapped to HapMap by taking the HapMap provided hg18 (NCBI Build 36.1) coordinates, lifting them to hg19 using the UCSC browser liftOver utility [45] and intersecting with the CoBEL variants. Nearly half (49%, 112/227) the enriched variants sites could be mapped to HapMap probes. NHGRI GWAS SNPs were mapped to HapMap SNPs using rsIDs. A GWAS SNP and a CoBEL variant were called in LD, using a maximalist approach, if either D’ > 0.99 or r2 ≥ 0.8 or LOD (log odds) ≥ 2 between their matching HapMap probes.
All enriched CoBELs from the five individuals were overlapped with the HGMD PRO 2015.2 set containing 130,218 disease mutations using overlapSelect from UCSC genome browser.
Over 90% of narcolepsy patients with cataplexy, and around 40% of narcolepsy patients without cataplexy carry human leukocyte antigen (HLA) type DQB1*06:02 [31]. The crystal structure of HLA-DQB1*06:02 (PDB ID: 1UVQ) [51] identified the representative amino acid haplotype of DQB1*0602 as F9G13L26Y30Y37A38D57 (subscript represents amino acid number in exon 2 of HLA-DQB1). Based on the variant call file, the haplotype present is George Church is different: Y9G13L26H30Y37A38D57. When we used BLAST to search the Church version of exon 2 against the IMGT/HLA Database [52], the allele closest to the observed haplotype was DQB1*06:03, not found associated with narcolepsy patients [53].
|
10.1371/journal.pgen.1007791 | Ancestral origin of ApoE ε4 Alzheimer disease risk in Puerto Rican and African American populations | The ApoE ε4 allele is the most significant genetic risk factor for late-onset Alzheimer disease. The risk conferred by ε4, however, differs across populations, with populations of African ancestry showing lower ε4 risk compared to those of European or Asian ancestry. The cause of this heterogeneity in risk effect is currently unknown; it may be due to environmental or cultural factors correlated with ancestry, or it may be due to genetic variation local to the ApoE region that differs among populations. Exploring these hypotheses may lead to novel, population-specific therapeutics and risk predictions. To test these hypotheses, we analyzed ApoE genotypes and genome-wide array data in individuals from African American and Puerto Rican populations. A total of 1,766 African American and 220 Puerto Rican individuals with late-onset Alzheimer disease, and 3,730 African American and 169 Puerto Rican cognitively healthy individuals (> 65 years) participated in the study. We first assessed average ancestry across the genome (“global” ancestry) and then tested it for interaction with ApoE genotypes. Next, we assessed the ancestral background of ApoE alleles (“local” ancestry) and tested if ancestry local to ApoE influenced Alzheimer disease risk while controlling for global ancestry. Measures of global ancestry showed no interaction with ApoE risk (Puerto Rican: p-value = 0.49; African American: p-value = 0.65). Conversely, ancestry local to the ApoE region showed an interaction with the ApoE ε4 allele in both populations (Puerto Rican: p-value = 0.019; African American: p-value = 0.005). ApoE ε4 alleles on an African background conferred a lower risk than those with a European ancestral background, regardless of population (Puerto Rican: OR = 1.26 on African background, OR = 4.49 on European; African American: OR = 2.34 on African background, OR = 3.05 on European background). Factors contributing to the lower risk effect in the ApoE gene ε4 allele are likely due to ancestry-specific genetic factors near ApoE rather than non-genetic ethnic, cultural, and environmental factors.
| The strongest risk gene identified for late-onset Alzheimer disease is ApoE. However, the risk for Alzheimer disease due to ApoE is not consistent across populations. For example, individuals with African ancestry experience less risk from ApoE ε4 than individuals of European or Asian ancestry. The cause of the difference in risk effect is currently unknown. This has led us to ask: What is/are the factor(s) contributing to the risk effect variation of ApoE across the populations? We hypothesized two possibilities for the variability of ApoE risk: 1) ethnic-related environmental factors that vary across populations, such as diet and lifestyle activities, or 2) a population-specific genetic difference in the ApoE gene, or in its surrounding region. We tested our hypothesis using populations with more than one genetic ancestral background, specifically African Americans and Puerto Ricans. Our study showed that the risk of Alzheimer disease is lower for individuals who inherited the genomic region surrounding the ApoE gene from an African ancestor than it is for risk allele carriers who inherited the region from a European ancestor. These findings suggest that protective genetic variant(s) most likely lie(s) within the genetic region surrounding the ApoE gene on the African ancestral background.
| Late-onset Alzheimer disease (LOAD) is a progressive neurodegenerative disorder characterized by loss of memory and other cognitive functions. It is the most common form of dementia worldwide [1], with prevalence increasing with age (e.g., ~30–40% by 85–89 years) [2]. The etiology of AD is multifactorial with genetic, and environmental factors all influencing risk.
The most significant genetic risk factor for LOAD is the ApoE gene [3,4]. Three common ApoE alleles have been identified (ε2, ε3, and ε4). The ε3 allele is the most frequent and is typically considered “neutral” regarding AD risk. The ApoE ε4 allele both increases the risk and decreases the age-of-onset of developing AD [4]. Conversely, the ε2 allele is protective against AD [4,5]. Although ApoE is an AD risk factor in nearly all populations, the risk of AD for ε4 carriers differs among racial/ethnic groups [6]. The strongest reported risk for ε4 allele is in East-Asian populations (ε3/ε4 odds ratio OR: 3.1–5.6; ε4/ε4 OR: 11.8–33.1) [6,7] followed by non-Hispanic Whites (NHW) (ε3/ε4 odds ratio [OR]: 3.2; ε4/ε4 OR: 14.9) [6,8–10] with a considerably lower risk to develop AD for an ε4 carrier in African-Ancestry populations, such as African Americans (AA) and Caribbean Hispanics (CHI). Studies in African-ancestry cohorts consistently reported significant association between ApoE ε4 homozygosity and AD, but showed inconsistent results for ε4 heterozygote allele individuals (ε3/ε4 OR:1.1–2.2; ε4/ε4 OR: 2.2–5.7) [6,8–13]. The reason for this heterogeneous risk effect of ApoE is currently unknown. This disparity in risk may be due to ethnic-related environmental factors that vary across populations, such as diet and lifestyle activities, or the difference may be due to population-specific genetic factors. Exceptions include studies among the Wadi Ara and American Indian populations, but these studies may suffer from low power due to small sample sizes [14–16].
Ancestral methods examining both global (GA) and local (LA) ancestry can be used to explore these different hypotheses. GA refers to an individual’s average ancestry across his/her entire genome while LA refers to the ancestral background of a particular (i.e., “local”) chromosomal region within an individual genome (Fig 1). GA is predominantly correlated with ethnic, cultural, and environmental factors that are related to broader definitions of race and ethnicity [17–20]. Conversely, LA is often correlated with ancestry-specific genetic factors that are located in or near the genomic region in question [21,22]. As such, an understanding of LA around the ApoE region may help inform how we interpret the race/ethnicity differences observed in ε4 risk. Specifically, if cultural and environmental effects play a major role in ApoE heterogeneity, we would expect GA to interact with ε4 to influence AD risk. There will also be GA and allele interaction if there is epistasis with alleles on other choromsomes that have different frequencies between ancestral populations. However, if genetic modifiers or protective factors local to the ApoE region (e.g., cis-acting enhancers, eQTL, etc.) play a major role in ApoE ε4 heterogeneity, we would expect LA to interact with ε4 to influence AD risk.
Admixed populations, due to their ancestral heterogeneity, often show complex patterns of GA and LA, enabling us to test these hypotheses. As such, we utilized two admixed populations (CHI from Puerto Rico (PR), and AA) to assess the relationship between ApoE ε4 risk and patterns of GA and LA. PR individuals commonly have European (EU), African (AF) and Amerindian (AI) ancestors, while AA individuals often have both EU and AF ancestors. To test the hypothesis that the population-specific risk is due to ethnic-related environmental factors that vary across populations, we compared those ApoE ε4 carriers who inherited most of their chromosomes from AF ancestors to those who inherited most of their chromosomes from their EU ancestors by using GA. If there are additional genomic loci outside of the ApoE gene contributing to the population risk difference, then individuals with the highest GA load of EU (or AF) ancestry would match the EU (or AF) population risk. Alternatively, to test the hypothesis that the disparity in risk may be due to genetic modifiers or protective factors local to ApoE, we compared the LAs in the admixed populations with those of the corresponding ancestral population (e.g., if one inherited his/her ApoE LA from the EU ancestors, his/her risk for AD would be similar to the EU population risk).
Our results strongly suggest that an ancestry-specific region surrounding the ApoE gene is contributing to the lower risk of AD in AA and PR ε4 carriers, supporting the hypothesis that the “protective” effect is due to the ancestry-specific genetic factors around the ApoE genomic region.
First, we performed two genotype-based regression tests to assess global ancestry and local ancestry interaction with ApoE genotype (see Methods for details). Results showed that the LA by ApoE interaction term (dose of AF ancestry by dose of ε4 allele; LAxApoE) was significantly different from 0 in both PR and AA populations (PR: likelihood ratio test (LRT), p-value = 0.019; AA: LRT, p-value = 0.005). The effect size of the interaction term was negatively correlated with AD (PR: OR = 0.2 (CI: 0.05–0.76); AA: OR = 0.75 (CI: 0.61–0.91)). This was in contrast to the GA by ApoE interaction term (GAxApoE), which was not significant in either PR or AA (PR: LRT, p = 0.49; AA: LRT, p-value = 0.65).
Since we identified a significant interaction, we performed a haplotype-based regression test to assess the effect size of ancestry-specific alleles (see Methods for details). We found that the effect size of the ε4 risk allele was significant across the ancestral haplotypes, even while accounting for correlations with GA (Table 1). In the PR dataset, the ε4 alleles on an EU ancestral background were significantly associated with AD (p-value = 3.7e-05; OR = 4.49) compared to ε3 alleles from an EU ancestral background. However, ε4 vs ε3 showed no significant effect on the AF LA background (p-value = 0.67; OR = 1.26). Similarly, in the AA dataset, the ε4 haplotypes of EU ancestry showed a stronger risk effect (OR = 3.05; p-value = 4.9e-17) than those in the AA dataset of AF ancestry (OR = 2.34; p-value = 9.2e-45). We tested the difference between the effect sizes of ancestral backgrounds by using t-test for means. Test results showed that effect sizes between the ancestral backgrounds are different with nominal significance in both populations (PR: p-value = 0.059; AA: p-value = 0.068). It is of note that these models all include GA as covariates, indicating that the effects seen are independent of the GA.
In the subgroup of individuals with homozygote ε4 and ε3 alleles, results showed that ε4 haplotypes of EU ancestry have a stronger risk effect (OR = 18.44 (CI: 9.6–35.6); p-value = 3.5e-18) than those with AF ancestry (OR = 6.48 (CI: 3.4–12.5); p-value = 4.3e-63). The t-test of means showed that effect sizes of EU and AF backgrounds are significantly different (p-value = 0.003).
Since we observed that AF ancestral background surrounding the ApoE gene is contributing to the lower risk of AD, we examined the genetic region surrounding ApoE by using 1000 Genome sequence data from three populations of the Utah Residents with Northern and Western European Ancestry (CEU), Japanese in Tokyo (JPT), and Yoruba in Ibadan (YRI). We identified 43 variants using Pearson’s chi-square test between the CEU vs. YRI and JPT vs. YRI populations, which were significant following the Bonferroni correction for multiple comparisons. Table 2 shows the list of 15 most significant variants with the Bonferroni corrected p-values less than 1 ×10−5. The whole list of variants is shown in the S2 Table. Fig 2 demonstrates Bonferroni corrected p-values for the pairwise comparisons between CEU and YRI, and JPT and YRI populations. The primary CEU and JPT peaks align, and lie within the strongest Topologically Associated Domain (TAD) containing the ApoE gene. None of the significantly different variants were in the protein-coding DNA in the defined region around the ApoE gene. It is noteworthy that just 6 variants in sequence data comparison showed significant difference (with the lowest p-value = 0.0052) between the CEU and JPT and all of them were found out of the TAD region containing the ApoE.
These findings strongly support our hypothesis that genetic modifiers local to the ApoE region influence the risk of the ε4 allele, showing a weaker risk effect on the AF ancestral background and stronger effect on the EU ancestral background. There was no evidence that overall ancestry (GA) has an effect on the heterogeneity of ApoE ε4 risk within the populations, which we used as a surrogate for non-genetic cultural/ethnic differences. Additionally, we observed a stronger risk effect on the EU ε4 haplotypes (or conversely, a protective effect on AF ε4 haplotypes). This effect was especially pronounced in an analysis of ε4 homozygotes against ε3 homozygotes, a result consistent with previous reports on ApoE risk across populations [6,8–13].
The overlapping of the subTAD (~50kb) region and the peaks of the allele frequency differences between the CEU, JPT and YRI support the hypothesis that the variant(s) modifying ε4 risk are most likely to lie in this region. The significant differences found in non-protein-coding DNA, suggests the protective effect is due to a regulatory difference between the local ancestries. This would also suggest that possible a modifier(s) would affect ApoE expression itself and supports the hypothesis that the genomic region surrounding ApoE with AF background reduces the risk for ε4 carriers and is evidence that genetic factors may be underlying the discrepancy in ε4 allele risk effect across populations.
It should be noted that this study was not well-powered to test AI background influence on ε4 risk allele. Further research is needed to study populations with higher AI ancestral background, such as Peruvian, Mexican, and Central American populations, to understand the correlation between the AI ancestry and ApoE. Similarly, limitations in sample size prevented us from assessing effects in ε2 carriers.
Our findings suggest that the ApoE region from AF populations may contain protective factors that help mitigate the effect of the ε4 allele. In particular, comprehensive analysis of the ApoE region and testing for protective loci may reveal previously unappreciated biological pathways and provide translational opportunities. Research that focuses on locating protective variants represents a complementary approach to accelerating the identification of more effective targets for drug development. This, in turn, will lead to better treatments, and help reduce health disparities.
All AA cases and controls selected for genotyping were obtained from the John P. Hussman Institute for Human Genomics (HIHG) at the University of Miami Miller School of Medicine (Miami, FL), North Carolina A&T State University (Greensboro, NC), Case Western Reserve University (Cleveland, OH), and the Alzheimer’s Disease Genetic Consortium (ADGC). Samples were collected as described previously [23,24]. The AA dataset contained 1,766 AD cases (69.8% female, mean age at onset (AAO) 77.6 years [SD 8.2]) and 3,730 cognitively healthy controls (72.0% female, mean age-of-examination (AOE) 76.5 years [SD (8.3)]).
PR individuals were ascertained as a part of the Puerto Rico Alzheimer Disease and Related Disorders Initiative study. Ascertainment was focused in metropolitan areas of New York, North Carolina, Miami, and Puerto Rico. Participants were recruited and enrolled after they (or a proxy) provided written informed consent and with approval by the relevant institutional review boards. For the PR cohort, 220 unrelated cases (69.6% female, mean AAO 75.1 years [SD 9.7]) and 169 unrelated cognitively intact controls (66.4% female, mean AOE 73.6 years [SD 7.1]) were ascertained.
For both AA and PR datasets, cases were defined as individuals with AD with AAO>65 years of age; controls were defined as individuals with no evidence of cognitive problems and AOE>65 years of age. All participants were evaluated to determine case or control status based on the National Institute of Neurological and Communicative Disorders and Stroke—Alzheimer’s Disease and Related Disorders Association, criteria [25,26]. Individuals with known or suspected dementia were evaluated using the LOAD study reference [27]. Individuals who were deemed to be cognitively normal were screened with the Mini-Mental State Examination [28] or the Modified Mini-Mental State [29]. The participants were classified as AA and PR based on self-report, and the GWAS analysis confirmed these data.
Genome-wide single-nucleotide polymorphism (SNP) genotyping was processed on three different platforms: Expanded Multi-Ethnic Genotyping Array, Illumina 1Mduo (v3) and the Global Screening Array (Illumina, San Diego, CA, USA). ApoE genotyping was performed as in Saunders et al. [30]. Quality control analyses were performed using the PLINK software, v.2. [31]. The samples with a call rate less than 90% and with excess or insufficient heterozygosity (+/- 3 standard deviations) were excluded. Sex concordance was checked using X chromosome data. To eliminate duplicate and related samples, relatedness among the samples was estimated by using identity by descent (IBD). SNPs with minor allele frequencies less than 0.01 and SNPs available in samples with the call rate less than 97%, or those not in Hardy-Weinberg equilibrium (p<1.e-5), were eliminated from further analysis [32]. Further details of the QC analysis can be found in the Supplement (S1 Table).
To explore the reasons for the differences in ε4 allele risk between the populations we first assessed the genetic ancestry (LA and GA), and then tested the effect of LA and GA on the ε4 allele by building three logistic regression models.
To assess the LA, we phased our datasets independently applying the SHAPEIT tool ver. 2 [33] using 1000 Genomes Phase 3 reference panel [34] with default settings. We defined a region around the ApoE that was broad enough (chr19: 44,000,000–46,000,000) to include potential enhancers, topological associated domains, etc. while narrow enough to ensure contiguous LA blocks for most individuals in the study. After selecting the ApoE region, we used RFMix [35], discriminative modeling approach, to infer LA at loci across the genome. We ran RFMix with the TrioPhased option and a minimum node size of 5. We used Human Genome Diversity Project (HGDP) data as the reference panel; two for AA (EU, and AF), and three for PR (EU, AF, and AI). Then, we eliminated samples with ancestral break points across the 2Mb window (N = 892) and labeled each admixture block using the RFMix estimates. As a result, we obtained haplotype data with three LA states (AF, EU, AI) in PRs and two (AF, EU) in AAs. Afterwards, we defined haplotypes according to LA states and ApoE variants. S1 Fig illustrates the defining of LA at the ApoE gene and S2 Table shows the number of e3 and e4 alleles along AF and EU local ancestry in each population for cases and controls.
Next, we assessed GA by performing principal components analysis (PCA) using the Eigenstrat program [36]. The AA and PR datasets were combined with reference panels (using HGDP reference panels) representing diverse ancestries: EU and AF for AA, and EU, AF and AI for PR.
To assess the effects of GA and LA on ε4 risk we used three logistic regression-based models. The first model utilized a genotype-based test to assess GA interaction with ApoE genotype. This model evaluated the role of GA and factors strongly correlated with GA (e.g., ethnic-related environmental factors) on ApoE risk variation among populations. The second model utilized a genotype-based approach to assess LA interaction with ApoE genotype. In this model, we examined the role of genetic modifiers or protective factors local to ApoE in risk variation. The third model utilized a haplotype-based approach to assess the effect sizes of ancestry-specific alleles (e.g., ε4 and ε3 alleles on the AF background) while accounting for correlations with GA. Statistical analyses were performed using the “GLM2” [37] and “GEE” [38] packages available in R computing environment.
To define the potential genetic factors modifying the ApoE effect size we assessed the sequence differences between the ancestral backgrounds among the ε4 haplotypes. First, using the 1000 genomes database, we obtained genomic DNA sequence data from three populations of the CEU, JPT, and YRI. Secondly, we extracted the ε4 haplotypes across the defined LA block of 2 mB. In addition to EU, we tested Japanese haplotypes because ε4 allele in East Asian populations has a high-risk effect as well [6,7]. Then, we performed Pearson’s chi-square test using allele frequencies at the region of interest among the populations (CEU vs. YRI and JPT vs. YRI) to identify the list of significantly different variants that likely contain the protective variant(s). We assessed the allele frequency difference on ε3 and ε4 haplotypes separately. To make a list of ε4 haplotype-specific alleles with the significantly different frequencies we removed those that showed significant difference also among the ε3 haplotypes. Finally, we performed the Bonferroni correction [39] for the multiple comparisons.
|
10.1371/journal.ppat.1004658 | Identification of Effective Subdominant Anti-HIV-1 CD8+ T Cells Within Entire Post-infection and Post-vaccination Immune Responses | Defining the components of an HIV immunogen that could induce effective CD8+ T cell responses is critical to vaccine development. We addressed this question by investigating the viral targets of CD8+ T cells that potently inhibit HIV replication in vitro, as this is highly predictive of virus control in vivo. We observed broad and potent ex vivo CD8+ T cell-mediated viral inhibitory activity against a panel of HIV isolates among viremic controllers (VC, viral loads <5000 copies/ml), in contrast to unselected HIV-infected HIV Vaccine trials Network (HVTN) participants. Viral inhibition of clade-matched HIV isolates was strongly correlated with the frequency of CD8+ T cells targeting vulnerable regions within Gag, Pol, Nef and Vif that had been identified in an independent study of nearly 1000 chronically infected individuals. These vulnerable and so-called “beneficial” regions were of low entropy overall, yet several were not predicted by stringent conservation algorithms. Consistent with this, stronger inhibition of clade-matched than mismatched viruses was observed in the majority of subjects, indicating better targeting of clade-specific than conserved epitopes. The magnitude of CD8+ T cell responses to beneficial regions, together with viral entropy and HLA class I genotype, explained up to 59% of the variation in viral inhibitory activity, with magnitude of the T cell response making the strongest unique contribution. However, beneficial regions were infrequently targeted by CD8+ T cells elicited by vaccines encoding full-length HIV proteins, when the latter were administered to healthy volunteers and HIV-positive ART-treated subjects, suggesting that immunodominance hierarchies undermine effective anti-HIV CD8+ T cell responses. Taken together, our data support HIV immunogen design that is based on systematic selection of empirically defined vulnerable regions within the viral proteome, with exclusion of immunodominant decoy epitopes that are irrelevant for HIV control.
| Attempts to develop an HIV vaccine that elicits potent cell-mediated immunity have so far been unsuccessful. This is due in part to the use of immunogens that appear to recapitulate responses induced naturally by HIV that are, at best, partially effective. We previously showed that the capacity of CD8+ T cells from patients to block HIV replication in culture is strongly correlated with HIV control in vivo, therefore, we investigated the virological determinants of potent CD8+ T cell inhibitory activity. We observed that CD8+ T cells from patients with naturally low plasma viral loads (viremic controllers) were better able to inhibit the replication of diverse HIV strains in vitro than CD8+ T cells from HIV-noncontroller patients. Importantly, we also found that the potency of the antiviral activity in the latter group was strongly correlated with recognition of selected regions across the viral proteome that are critical to viral fitness. Vaccines that encode full-length viral proteins rarely elicited responses to these vulnerable regions. Taken together, our results provide insight into the characteristics of effective cell-mediated immune responses against HIV and how these may inform the design of better immunogens.
| Only two HIV vaccines designed to elicit protective T cell responses have reached clinical efficacy testing, both with disappointing results [1][2][3]. The reasons for this are not completely understood, despite much accumulated knowledge regarding the characteristics of cell-mediated immune responses associated with HIV and SIV control. The limited magnitude and breadth of vaccine-induced T cell responses, particularly when compared with responses to similar vaccines in non-human primate models, the modest cytotoxic capacity of CD8+ T cells, waning of responses over time, bias towards targeting of more variable regions of the viral proteome and the modest immunogenicity of the vaccine vector regimens are all likely contributing factors [2][4][5][6][7][8]. A critical first step towards addressing this is to determine whether the antiviral efficacy of CD8+ T cells is a function of their specificity.
The HVTN 502 (Step) and 503 (Phambili) trials were a test-of-concept for induction of protective T cell responses that collectively evaluated Merck’s trivalent adenovirus type 5 HIV-1 Gag/Pol/Nef vaccine in ∼3800 subjects at high risk of HIV acquisition [1][9]. Post-hoc analyses of HVTN 502 have shown that individuals in whom vaccine-induced responses targeted ≥3 epitopes in Gag achieved a lower viral load after HIV infection than subjects without Gag responses; it is striking, however, that these subjects were a small minority among the vaccinees (<7%) [6]. While this confirms several observational studies that showed an association between HIV control and preferential recognition of Gag epitopes [10][11], the question remains as to why vaccines that express full-length Gag proteins have so far failed to induce responses that can impact on HIV replication after infection. The answer may be two-fold: first, immunodominance hierarchies of the T cell responses elicited by these vaccines often mimic those of natural infection, with ‘hotspots’ in variable and least vulnerable regions of the viral proteome [12]; second, even within Gag and other conserved proteins, not all epitopes are equal in terms of vulnerability to immune pressure, or ‘fragility’, which is defined by the capacity to maintain function in the face of genetic mutations [13]. Thus, the efficacy of cell-mediated immune responses may depend on the specific epitopes targeted, both within and outside Gag. This was demonstrated in an observational study of 950 clade B- and C-infected individuals, in whom responses to overlapping peptides (OLP) spanning the entire viral proteome were systematically analysed [14]. A ‘protective ratio’ (PR) was calculated for each OLP from the ratio of the median viral load in subjects who failed to respond to the OLP to responders. OLP with a protective ratio >1 were defined as ‘beneficial’. Of note, Gag proteins contained the majority of the beneficial regions, though not all of them, and also contained regions that were not targeted by protective responses. Together, these data support the ‘decoy’ hypothesis, which proposes that certain epitopes within the viral proteome elicit dominant yet irrelevant responses that serve to undermine effective targeting of regions of vulnerability [15]. This question will only be adequately addressed by clinical testing of rationally designed immunogens based on ‘beneficial’ regions, as proposed by Rolland et al. and Mothe et al. [15][14].
Aside from identifying specific beneficial targets, the precise mechanisms and effector functions of antiviral T cell responses that underlie heterogeneity in HIV control among infected individuals need to be defined. We showed in a prospective study that CD8+ T cell viral inhibitory activity in vitro strongly correlated with HIV control in vivo, reflected in both viral load set-point and CD4+ cell decline over time [16]. This indicates that CD8+ T cell viral inhibitory activity is expressed on a continuum and is not a discrete function that is unique to HIV controllers with protective HLA alleles, providing scope for induction of effective CD8+ T cell responses by vaccination of subjects who do not have a favourable genotype. Viral inhibition assays that use polyclonal T cell populations provide a composite measure of lytic and non-lytic activity of all circulating HIV-specific CD8+ T cells, which may be heterogeneous in their functional capacity [17][18][19][20][21][22]. This activity is detectable in acute infection in a minority but rapidly wanes, likely as a result of viral escape and / or functional impairment [23][24][21][25]. Low level activity has also been detected in HIV-naïve recipients of DNA and adenovirus type 5-vectored vaccines encoding full-length HIV proteins even though such vaccines are capable of eliciting substantial numbers of Gag- and Pol-specific cytokine-secreting T cells [23][26][3]. These observations underscore the need for better understanding of the factors that determine the potency of CD8+ T cell viral inhibitory activity.
We also showed previously that CD8+ T cell viral inhibition in chronically infected individuals did not correlate with the total magnitude of IFN-γ-positive T cell response to any single HIV protein, including Gag [16]. This was surprising, given the known associations between Gag responses and HIV control, and led us to propose the hypothesis that potent viral inhibition depends on preferential targeting of selected regions that are not limited to Gag nor predicted by conservation score alone. We hypothesised that responses to such critical regions are generally subdominant and that this may explain the lack of efficacy of T cell-inducing vaccines. To this end, we investigated CD8+ T cell-mediated inhibitory activity in a subset of HIV-positive HVTN 502 and 503 vaccine trial participants. This comprised recipients of both the vaccine and placebo who were sampled at the same time during early HIV infection (1 year). They were naïve to antiretroviral therapy (ART), with CD4 cell counts >350 cells/μl, and were not selected for low virus loads or protective HLA class I alleles. In parallel, we studied ART-naïve subjects who showed spontaneous long-term control of HIV, with plasma viral loads consistently <5000 copies/ml (viremic controllers, VC). They were sampled later in infection (median 4.5 years) and were included as a reference cohort, as potent CD8+ T cell antiviral activity has been reported in such individuals [23][26][16].
CD8+ T cell antiviral activity was measured in 34 HIV-positive HVTN 502 & 503 trial participants, who were infected with clade B and C viruses respectively. They were aligned for duration of infection, early post-infection viral load and CD4+ cell counts. Only a minority had either a protective HLA class I allele (n = 7, 20%) or evidence of spontaneous viremia control, indicated by plasma viral loads consistently below 5000 copies/ml (n = 5, 15%) (Table 1). We included both vaccinees and placebos in order to maximise the number of subjects with samples available for analysis. Fourteen VC with viral loads <5000 copies/ml were studied in parallel as a reference cohort. The estimated duration of HIV infection in latter ranged from 1–11 years. Six (43%) had a protective HLA class I allele and all were presumed clade B-infected (Table 2). The inclusion of clade B and C cohorts enabled us to ascertain whether the association between CD8+ T cell inhibitory activity and HIV control was clade-independent, as suggested by our previous results [16]. However, a major goal of this study was to explore the extent of cross-clade inhibition (breadth) using a panel of laboratory-adapted and primary HIV isolates representing clades A, B and C strains, as this had not been systematically examined in HIV-positive individuals before. CD8+ T cells from HIV-positive HVTN 502 & 503 participants were tested according to PBMC availability, using at least one clade B and one clade C virus, while all VC were tested against five viral isolates (S1 Table). Among the HIV-positive trial participants in whom viral inhibitory activity against a clade-matched virus was analysed at a CD8+/CD4+ cell ratio of 2:1, it was not significantly different between vaccinees (n = 20) and placebo (n = 8) recipients (ranges 0–87% vs. 0–93%, p = 0.32; Fig. 1A). Because no difference was observed, analyses presented in the main were performed by combining data from both the vaccinee and placebo groups. However, the vaccinees were also analysed independently as they accounted for two-thirds of the HVTN cohorts. The data are shown in Supplementary Results, S1 Text. Whether data were combined or independent, the results were similar. Inhibition of a clade-matched virus was significantly higher among VC at CD8+/CD4+ T cell ratios of both 2:1 (medians 85% and 37% respectively, p <0.0001) (Fig. 1B) and 1:10 (medians 61% and 0% respectively, p <0.0001 (Fig. 1C). VC also showed more potent cross-clade inhibition than HVTN 502 participants when tested using a clade C virus (CD8+/CD4+ T cell ratio of 2:1—medians 60% vs. 14%, p = 0.002) (Fig. 1D). These differences remained significant when the placebos were excluded from the analyses (Supplementary Results, S1 Text). Cross-clade activity was analysed further using at least 3 viruses in 14 HVTN 502 & 503 participants and 14 VC. Differences in the potency and breadth of CD8+ T cell-mediated inhibitory responses in these groups are highlighted in the heatmap (Fig. 1E).
We have previously reported a significant inverse relationship between CD8+ T cell antiviral activity measured 6 months post-infection in a primary HIV infection cohort and viral load set-point, a known predictor of the rate of progression to AIDS [16]. In the present study, CD8+ T cell inhibitory activity was measured later. Nevertheless, there was still a significant inverse correlation between CD8+ T cell inhibition of a clade-matched virus and viral load set-point (which was attained within 100 days of infection in the HVTN trial participants) or current viral load in the VC (r = -0.49, p = 0.0009, S1 Fig.).
The finding that HIV-positive trial participants showed less potent inhibition of a clade-matched virus isolate than VC was consistent with results from previous studies of early infected individuals [16][24]. Here, we extended these observations to clade-mismatched viruses. The broader CD8+ T cell inhibitory responses in VC suggested that they preferentially recognised conserved viral epitopes. However, when examining responses within the groups, we observed more potent inhibition of clade-matched than mismatched viruses in VC and HIV-positive trial participants alike. This indicated that CD8+ T cells targeting clade-specific viral epitopes must contribute to the overall potency of the response. To investigate this further, we used ex vivo IFN-γ Elispot assays to measure the magnitude of responses to two sets of overlapping 15-mer peptides. The first corresponded to the beneficial regions that were defined by Mothe et al. in clade B and clade C-infected populations (S2 and S3 Tables) and the second to a set of ‘conserved elements’ (CE) peptides that were originally defined by Rolland et al. and consisted of 7 regions in Gag p24 (S4 Table) [14][27][28]. The peptides representing beneficial regions were constituted in pools according to their previously defined protective ratio, with the first pool of each protein containing the peptides with the highest protective ratio (higher number indicating lower viral load in responders compared with non-responders) [14]. CE peptides were divided into pools A & B, also in accordance with previously observed associations with low virus loads [29] (S4 Table). To match their infecting clade, VC and HVTN 502 participants were tested with a peptide set representing beneficial regions in Clade B and HVTN 503 subjects were tested with the Clade C beneficial peptide set. All three groups were tested with the same CE peptide set. For all Elispot assays, CD8+ T cells were obtained from the same sample as that used in the viral inhibition assay (except for 2 VC in whom it was necessary to use an additional sample obtained within 1 year of the original bleed). Summed frequencies of IFN-γ-producing CD8+ T cells targeting the beneficial and CE peptides are shown in Fig. 2. The median response to beneficial peptides was 190 and 262 SFU/million CD8+ T cells for HVTN 502 and 503 groups respectively and 210 SFU/million CD8+ T cells for the VC (Fig. 2A). The median response to the CE peptides was 60 SFU/million CD8+ T cells for the combined HVTN groups and 35 SFU/million CD8+ T cells for the VC (Fig. 2B). These differences were not statistically significant, nor were there significant differences between vaccinees and placebos in terms of the magnitude of response to either beneficial (medians 198 vs. 415 SFU/million CD8+ T cells, p = 0.99) or CE peptides (medians 55 vs 60 SFU/million CD8+ T cells, p = 0.6).
This group of VC did not show significantly higher responses to beneficial or CE peptides than the HVTN subjects. This was unexpected in the light of previous reports but likely reflected the longer duration of infection (median 4.5 years vs. 1 year), which may be associated with loss of responses to epitopes within the regions studied, due to mutational escape [14][29][30][31][32]. For example, the two VC who were HLA-B*5701-positive did not make detectable responses to the beneficial or CE peptide pools that contained immunodominant Gag epitopes restricted by this allele (TW10 and KF11).
We next explored the relationship between virus inhibition and the magnitude of CD8+ T cell responses to the beneficial and CE regions in the HVTN subjects. We observed a strong correlation between the magnitude of T cell responses to beneficial regions and CD8+ T cell-mediated inhibition of a clade-matched virus (r = 0.69, p = 0.0001 for a CD8+/CD4+ cell ratio of 2:1) (Fig. 2C). This relationship was also confirmed using a lower CD8+/CD4+ cell ratio of 1:1 (r = 0.5, p = 0.01) and importantly, was maintained after removal of subjects with protective HLA class I alleles (HLA-B*27, B51, B*5701/03, B*5801, B*81) (r = 0.71, p = 0.0005) (Fig. 2D). Furthermore, these correlations remained statistically significant after exclusion of placebos (Supplementary Results, S1 Text). Taken together, these analyses suggested that CD8+ T cell viral inhibition of >85% (i.e. the median response in VC) was associated with a beneficial peptide response threshold of ∼1300 SFU/million CD8+ T cells. Additional support for the relationship between CD8+ T cell viral inhibition and magnitude of T cell responses to beneficial regions was obtained in a subset of subjects (n = 15) in which individual peptides were tested in cultured Elispot assays. The highest viral inhibition also correlated with the higher magnitude T cell responses to individual beneficial peptides (r = 0.61, p = 0.02, Fig. 2E). Unexpectedly, there was a weaker association between the magnitude of T cell responses to the conserved elements pools and CD8+ T cell viral inhibition (r = 0.41, p = 0.04) (Fig. 2F). This positive relationship was also maintained after exclusion of placebos (Supplementary Results, S1 Text) and was largely driven by responses to the conserved elements pool B, containing peptides spanning CE 4, 5 and 6.
We also analysed the frequency of T cell responses to the total HIV proteome as these had been measured previously by intracellular staining for IFN-γ (at a median of 5 weeks after HIV infection) after stimulation of PBMC with clade B consensus potential T cell epitope (PTE) peptide sets [2][30]. These were selected to optimise the detection of CD8+ T cell responses to circulating viruses and thus ensure accurate measurement of the maximum response [33]. The total proteome response (median) was 1.81% CD8+ T cells (Fig. 3A), with no significant difference between HVTN 502/503 vaccinees and placebos (median 2.1% and 1.5% of CD8+ T cells respectively, p = 0.23), which is similar to data obtained from chronic infection cohorts [31]. There was no correlation between CD8+ T cell antiviral activity and responses to the whole proteome, either for HVTN subjects as a whole (r = 0.14, p = 0.5) (Fig. 3B) or for the vaccinees only (Supplementary Results, S1 Text).
In view of the strong correlation between CD8+ T cell antiviral activity and recognition of beneficial peptides, we explored this relationship further using a series of univariate and multivariable regression models, with CD8+ T cell antiviral activity as the dependent variable. We investigated associations with the following independent variables: 1) the Shannon entropy score for each beneficial region as a measure of its variability at the population level; 2) the magnitude of responses to beneficial regions (‘total beneficial’ response); 3) the magnitude of the Gag component of the beneficial regions (‘beneficial Gag’ response), in order to ascertain how much this contributed to the total beneficial response; 4) the magnitude of responses to CE peptides; 5) the ratio of magnitude of responses to beneficial regions to the total proteome response (relative magnitude or immunodominance) and 6) the presence of protective (‘good’) or non-protective (‘bad’) HLA class I alleles.
For our first set of models, entropy was used as the primary independent (or predictor) variable of interest since both the beneficial regions and CE regions were largely derived from conserved, i.e. low entropy regions in the viral proteome [14][28]. Thus, our first regression model included entropy as the only independent variable. Total beneficial responses, beneficial Gag responses or CE responses were then each added separately to this baseline model to ascertain whether they improved the fit of the model (as captured by a change in the model r2) and, thus, whether they were independently associated with CD8+ T cell activity. Entropy alone explained 13.5% of the variance in inhibition. Addition of the total beneficial response or the beneficial Gag response each improved the fit of the model (by 46% and 24% respectively) and the contribution of each of these was statistically significant (Table 3). By contrast, addition of the CE response had no effect (increase in model r2 of 0.1%).
We also constructed three multivariable regression models that included various combinations of the following factors: magnitude of total beneficial responses, relative magnitude or entropy of beneficial regions and good or bad HLA class I alleles. The combinations of covariates for these models were chosen to allow us to investigate several potential pathways for any associations, based on hypothesised interactions between absolute and relative magnitude of responses and between certain HLA alleles and entropy of epitopes restricted by these alleles. These models explained 39–49% of the variance in CD8+ T cell inhibition and all were significant as a whole. However, in each case the magnitude of the response to beneficial regions made the strongest unique contribution whereas the contribution of the other variables was not statistically significant (Table 4).
Given that responses to beneficial regions were subdominant in HIV-infected individuals, we next investigated whether this was also the case for responses that are primed in HIV-naïve individuals by vaccines encoding full-length HIV proteins. Data on responses that developed post-vaccination and prior to HIV acquisition were available for 13/20 of the HVTN 502 trial participants in this analysis (sampled 4 weeks after the second vaccination) [6]. We compared the magnitudes of vaccine-induced responses to peptides spanning the entire Gag/Pol/Nef immunogen with beneficial and CE regions. Vaccination induced responses to beneficial regions in 5/13 patients and to CE regions 3/13 patients, while no response to any of these regions was detected in 5 subjects. Overall, vaccine-induced responses to beneficial regions accounted for a median (range) of 0% (0–43%) of the response to the entire immunogen in these subjects, despite representing 36% of the immunogen sequence (Fig. 4A).
Finally, we investigated whether natural immunodominance hierarchies were maintained or altered following the administration of a Gag immunogen as a therapeutic vaccine in chronic HIV infection. We mapped T cell responses to beneficial and non-beneficial regions before and after vaccination with an immunogen, ‘HIVA’ comprising full-length Gag p24/p17 sequences fused to a multiepitope string, delivered as a modified vaccinia virus Ankara-vectored vaccine to chronic ART-treated HIV-positive subjects with suppressed viremia [34][35]. Epitope mapping was performed in 9 subjects using overlapping 15-mers spanning p24 and p17, together with optimal 8–10-mer peptides for epitopes that had been defined previously (Table 5) [36]. We confined our analysis to responses to the Gag component of the immunogen, since the epitope string was, by definition, designed to focus responses on selected regions of the proteome. Prior to vaccination, the magnitude of summed responses to beneficial regions was lower than for non-beneficial Gag regions, although the difference was not statistically significant (median 205 and 615 SFU/million PBMC respectively, p = 0.27). MVA.HIVA vaccination significantly boosted T cell responses to the beneficial Gag regions (median change +150 SFU/million PBMC, p = 0.03). However, responses to non-beneficial Gag regions were preferentially expanded (median change +845 SFU/million PBMC, p = 0.004) (Fig. 4B, Table 5). Taken together, these data suggest that vaccines encoding full- or near full-length HIV proteins mimic natural HIV infection by eliciting responses that are biased towards non-beneficial targets, regardless of whether they are administered to HIV-naïve or primed individuals.
The lack of a reliable correlate of protective immunity against HIV is a significant obstacle to systematic evaluation of vaccine candidates. Consequently, efforts to develop a T cell-based vaccine have focused broadly on recapitulating the immunological phenotype of HIV controllers, using immunogens incorporating near-complete gene sequences for many proteins. Recently, there has been greater emphasis on rationally designed immunogens, in particular, those that aim to maximise coverage of variable viral epitopes (mosaics) or avoid them altogether (conserved regions) [15][37][38][39][8]. CD8+ T cell-mediated viral inhibition was found to correlate with the frequency of T cells targeting conserved epitopes in HIV-uninfected vaccinees [8][40]. However, no vaccine candidate has yet been shown to elicit viral inhibitory activity of similar potency to that observed in HIV controllers. Here, we report that the total viral inhibitory capacity of anti-HIV CD8+ T cells is highly dependent on their specificity and we provide a mechanism to explain why conventional HIV immunogens elicit largely ineffective CD8+ T cell responses.
We reported previously that ex vivo CD8+ T cell-mediated viral inhibitory activity is inversely correlated with viral load set-point; we confirmed this finding here in genetically unrelated cohorts infected with different viruses [16]. While this is consistent with well-established associations between primary CD8+ T cell responses to HIV-1 and control of acute viraemia [41][42][32][43], the time interval between attainment of viral load set-point and sampling for the viral inhibition assay was longer in the present study, thus we cannot rule out the possibility that early control of viraemia was the cause rather than the consequence of the level of antiviral activity. It is also conceivable that a viral inhibition ‘set-point’ is attained soon after infection; this could explain the findings of Lecuroux et al., who reported that most HIV-infected individuals showed modest CD8+ T cell inhibitory activity throughout acute and early infection [24]. Nevertheless, our data give insight into the level of inhibitory activity that might be used as a benchmark to assess vaccine candidates: for example, inhibition of a clade-matched virus by ≥ 85% (observed in 50% of VC subjects but only 7% of HVTN trial participants) was associated with a median viral load of ∼ 2000 copies/ml. This suggests that the bar must be set very high if such assays are to be used to identify vaccine strategies that could clear HIV infection or reduce viral loads to undetectable levels [44].
We report for the first time, to our knowledge, that the breadth of inhibitory activity, indicated by inhibition of clade-mismatched viruses, was significantly greater in VC than subjects with uncontrolled viraemia. This suggested two non-mutually exclusive explanations: enrichment of the HIV-specific repertoire in VC for T cells recognising conserved epitopes and / or high frequencies of circulating cross-reactive CD8+ T cells that can tolerate epitope variation. However, potent clade-specific viral inhibitory activity, together with differential inhibition of diverse viruses was evident in both study groups. This led us to hypothesise that factors other than epitope conservation must play a role in the control of viral replication. We found that CD8+ T cell antiviral activity in HVTN subjects was highly correlated with the frequency of CD8+ T cells targeting selected peptides that had been shown in an independent study of two large cohorts to associate with control of viraemia [14]. This correlation was independent of protective HLA class I alleles, which suggests that effective CD8+ T cell responses may be restricted by a broader range of HLA class I alleles than previously suspected, as was also proposed by Mothe et al [14]. While the viral regions that were defined as beneficial were predominantly of low entropy, our regression analysis indicated that the magnitude of these responses accounted for a significantly greater proportion of the variation in viral inhibition than entropy alone. The Gag component of these regions explained nearly two-thirds of the effect. Interestingly, T cell responses to conserved elements peptides were weakly correlated with viral inhibition and this effect was driven by only three of the seven conserved regions tested. This is consistent with other studies showing that high population-level conservation per se does not necessarily predict viral fitness and may reflect the presence of invariant regions that are immunologically inert [27][45]. Collectively, these observations are not only reconcilable with previously described associations between broad Gag-specific T cell responses and reduced viral loads at the population level but also point to a mechanism that could explain them with greater precision [10][14][6]. The greater the breadth of responses to Gag, the higher the probability of targeting the most vulnerable epitopes, even though there is also the possibility of targeting the non-beneficial regions. The lack of responses to beneficial regions in some of the VC studied is quite likely explained by the small sample size studied and / or the extended time of untreated HIV infection which may have led to elimination of some of these T cell responses, or possibly that these VC made responses to other critical epitopes that were not represented in our peptide sets [32][46][47]. However, this does raise questions as to how long the effect of responses to beneficial regions lasts, in the face of ongoing viral escape. The rate of escape from CD8+ T cell responses is determined by the net effect on viral fitness of all escape mutations and is significantly slower in chronic than acute infection [48]. The association between the prevalence of T cell responses to beneficial regions and population-level viral load was made in chronically infected cohorts and suggests, therefore, that even though these beneficial responses may drive viral escape, the net effect is an overall impairment of viral fitness. This is consistent with observations made by Boutwell et al. who showed that CD8+ T cell escape mutations in HIV-1 Gag frequently impair viral fitness; many of the susceptible epitopes in their study were located in the beneficial regions [49].
It is possible that we have overlooked functional characteristics of Gag-specific CD8+ T cells such as the capacity to produce multiple cytokines simultaneously, as these have also been associated with control of viraemia [50][51]. However, viral inhibition assays arguably provide the most direct and complete measure of antiviral function, whereas the cytokines that are typically detected in assays of T cell polyfunctionality provide an indirect assessment. Our analysis indicated that individuals with potent viral inhibitory responses are rare, as was reported by others [24], and furthermore highlighted that responses to beneficial regions within the HIV proteome are both infrequent and subdominant. This is consistent with a previous study that showed infrequent targeting of epitopes in these regions in acute infection [32]. As spontaneous control of viraemia is itself a rare event, this provides further evidence that viral inhibitory activity in vitro accurately reflects immune control in vivo. It also raises questions as to whether long-term control or even clearance of infection can be achieved by vaccines that mimic priming by HIV. Responses elicited by the Ad5-HIV vaccine in HVTN 502 trial participants were shown previously to be limited in breadth, with a bias towards variable regions [2][7]. Our retrospective analysis of a subset of HVTN 502 vaccinees indicated preferential targeting of non-beneficial regions, which was concerning given that the Gag/Pol/Nef immunogen contained the majority of the previously described beneficial regions [14]. We observed a similar skewing of responses in HIV-positive subjects who received a therapeutic MVA vaccine encoding the immunogen, HIVA, which included 9 of the identified beneficial regions within Gag. Newer vaccine candidates such as Ad35-GRIN and Ad35-ENV, which comprise Gag, Reverse transcriptase, Integrase and Nef and Env sequences, induced responses to a median of one Gag epitope in HIV-uninfected healthy volunteers [40]. The common factor among these immunogens is the inclusion of full or near-full-length Gag sequences. A non-human primate study showed that full-length HIV immunogens induced responses to conserved regions that were of similar breadth to those elicited by non-native conserved region immunogens [52]; by contrast, Kulkarni et al. compared vaccination with p55 Gag and a conserved elements-only immunogen and showed better recognition of conserved elements epitopes with the latter approach [28,53]. Taken together, these observations highlight the need for vaccines to overcome natural immunodominance hierarchies in humans through the development of immunogens that focus responses on specific critical regions of the viral proteome. Additional refinements, such as inclusion of sequences that pre-empt predictable escape mutations, should also be considered [54]. Vaccine-mediated clearance of an AIDS virus infection in the non-human primate model was recently demonstrated for the first time with a persistent rhesus CMV SIV vaccine [55,56][57]. It is noteworthy that the responses elicited were unique in terms of their unprecedented breadth, absence of immunodominance and specificity for non-canonical viral epitopes, although the immunogen comprised entire proteins. While this may reflect unusual properties of the CMV vector and the specific mechanisms that contributed to virus eradication have yet to be resolved, such studies may provide vital lessons for human vaccine development.
In summary, these data provide several new insights that should inform HIV vaccine design. First, they suggest that induction of effective anti-HIV CD8+ T cell responses could be achieved with an immunogen comprising only a few selected regions of the viral proteome. In addition to the regions defined by Mothe et al., which were identified in chronically infected individuals, comprehensive analyses of responses that arise during acute / early HIV infection may yield viral targets that are critical to early and sustained control [32][58]. Secondly, we have identified a possible threshold for the magnitude of responses to these critical regions that should be attained in order to have a meaningful impact on viral replication. Our analysis of responses to vaccination with Ad5 Gag/Pol/Nef in a small subset of HVTN 502 subjects prior to HIV infection, together with other post-hoc studies, suggests that this is extremely unlikely to be achieved using immunogens that comprise full-length proteins. Exclusion of irrelevant decoy regions that when present, often induce immunodominant T cell responses, may be essential to prevent the development of such non-protective responses. Finally, our previous experience with potent heterologous viral vector combinations has shown that it is feasible to induce HIV-specific T cell responses in human subjects of the order of magnitude that we have proposed here [8]; rationally designed immunogens that exploit these vectors should be prioritised for clinical development.
Approval was obtained from the Oxford Tropical Research Ethics Committee for analysis of anonymised PBMC samples that were made available to University of Oxford, UK by Fred Hutchinson Cancer Research Center via a Material Transfer Agreement (‘HVTN 502/Merck 023—HVTN 503 Ancillary Study’) following approval of the study by HVTN Protocol Committee. The PBMC samples were gathered and obtained from a collection held by HVTN. Viremic controllers (VC) were recruited at Duke University Medical Center with IRB approval and after obtaining written informed consent.
The HVTN 502 and 503 studies have been described previously [1][9]. PBMC sampled from 36 HIV-positive HVTN 502 and 503 participants who were still naïve to ART 12 months after HIV acquisition, with CD4+ cell counts >350 cells/μl, were provided through the HVTN 502 Oversight Committee. Plasma viral load data were provided by SCHARP and set-point was determined using the method described by Fellay et al. [59]. Participants’ characteristics are given in Table 1. Criteria for enrolment of VC were plasma viremia consistently <5000 copies/ml for at least one year and a CD4+ cell count >400 cells/μl in the absence of ART. However, one subject was included despite a CD4+ cell count <400 cells μl because of viral loads consistently <2280 copies/ml for 5 years prior to enrolment; this individual maintained viral loads <448 copies/ml during the study. Two subjects had transient viraemia >5000 copies/ml which was subsequently spontaneously controlled. Patients’ characteristics are given in Table 2. All VC had presumed clade B infection, due to the geographical location. Therapeutic vaccine trial participants were patients with chronic HIV infection, receiving effective ART for at least 12 months, with CD4+ cell counts >350 cells/μl, who received two intramuscular immunisations of MVA.HIVA 5x107 pfu 4 weeks apart [34][60]. HLA typing was performed as described previously [6].
Virus subtyping was performed by near full-length genome sequencing, as described previously [61] or by bulk sequencing of p17 Gag and analysis using REGA HIV-1 & 2 Automated Subtyping Tool (Version 2.0) [62][63].
HIV-1 isolates were obtained from the Programme EVA Centre for AIDS Reagents, National Institute for Biological Standards and Control (NIBSC), a centre of the Health Protection Agency, UK. The virus panel comprised two laboratory-adapted clade B isolates, BaL (CCR5-tropic) and IIIB (CXCR4-tropic) and three primary isolates, ES X-1936 (clade C, CCR5-tropic), 92UG029 (clade A, CCR5 / CXCR4 dual-tropic) and RW93024 (clade A, CXCR-tropic). All virus propagation was performed using primary CD4+ cells and 50% tissue culture infectious doses (TCID50) for each virus was calculated as described previously [64].
Clades B and C consensus peptides spanning the entire HIV proteome (15-mers overlapping by 11 amino acids) were obtained from the NIH Aids Reagent Programme. 10mg/ml stocks were stored at -80°C until required, then were diluted to generate working stocks. One or more 15-mer peptides that matched most closely the beneficial OLP described by Mothe et al. and the CE peptides described by Kulkarni et al. were selected for use in Elispot assays [14][28] (Tables 3–5).
The viral inhibition assay has been described in detail elsewhere [16,65]. Briefly, CD8+ T cells were isolated from cryopreserved PBMC by magnetic bead selection (Miltenyi Biotec) and retained for use in IFN-γ Elispot assays. CD8-depleted cells (hereafter referred to as CD4+ T cells) were stimulated with PHA (5 μg/mL) in RPMI 1640 medium supplemented with 10% fetal calf serum (R10) for 3 days, washed, and infected with HIV-1 isolates at pre-determined optimal MOI (National Institute for Biological Standards and Control, United Kingdom). To assess viral inhibition, HIV-superinfected CD4+ T cells (5 × 104) were cultured in triplicate in R10 with interleukin 2 (20 IU/mL) in 96-well round-bottomed plates, alone or together with unstimulated ex vivo CD8+ T cells, obtained by positive bead selection of PBMCs from a second freshly thawed vial on day 3. CD8+ T cells were confirmed as >98% pure by staining for CD3, CD8, and CD56. CD8+ and CD4+ T cells were co-cultured for 6 days for all virus isolates except clade A2, for which the peak of virus replication is attained after 3 days [65]. CD8+/CD4+ ratios of 2:1, 1:1 and 1:10 were tested, according to cell availability. On the day of harvest, cells were stained first with Aqua Live/Dead Fixable stain (Invitrogen), fixed with 1% paraformaldehyde/20 μg/mL lysolecithin at RT, permeabilized with cold 50% methanol followed by 0.1% Nonidet P-40, and finally stained with p24 antibody (KC-57-FITC; Beckman Coulter) and antibodies to CD3, CD4, and CD8 (conjugated to APC-Cy7, PerCP, and APC, respectively; BD Biosciences). Samples were acquired on a CyAn flow cytometer. Data were analyzed using FlowJo software. Antiviral suppressive activity was expressed as percentage inhibition and determined as follows: [(fraction of p24 + cells in CD4 + T cells cultured alone)–(fraction of p24 + in CD4+ T cells cultured with CD8+ cells)]/(fraction of p24 + cells in CD4 + T cells cultured alone) × 100.
Purified CD8+ T cells from the PBMC sample that was used to isolate CD4+ T cells for the viral inhibition assay were tested in IFN-γ Elispot assays with pools of beneficial or CE peptides (final concentration 2μg/ml) as described previously [16]. Mapping of responses to epitopes in the Gag component of the HIVA immunogen was performed using PBMC sampled pre- and 2 or 4 weeks post-vaccination, with overlapping 15-mer peptides (final concentration 4μg/ml) spanning the entire immunogen sequence, with confirmation using optimal 8–10-mer peptides where available [60]. Elispot assays with CD8-depleted PBMC were performed to confirm that these responses were CD8+ T cell-mediated. In selected assays, CD8+ T cells were recovered from the Elispot plate after overnight incubation with peptides, washed and cultured (2x106/ml) in R10 medium (RPMI with 10% fetal calf serum) plus IL-7 (25ng/ml). Cultures were supplemented with IL-2 (1.8 x103 units/ml) on day 3 and R10/IL-7/IL-2 medium was replaced on day 7. Cells were starved of IL-2 for 30 hours on day 10 and then used in cultured IFN-γ Elispot assays with individual peptides (2μg/ml).
Intracellular cytokine staining was performed as described previously, typically at the second visit after HIV infection had been confirmed [66][2].
Group comparisons were performed using the Mann Whitney test and correlations were investigated by determination of Spearman’s rank coefficient, using Graphpad Prism software, version 6. Models to explore predictors of inter-subject variation in viral inhibition by CD8+ T cells were tested using univariate and multivariable linear regression. Analyses were performed using SPSS version 22.
|
10.1371/journal.pgen.1004834 | Inactivation of PNKP by Mutant ATXN3 Triggers Apoptosis by Activating the DNA Damage-Response Pathway in SCA3 | Spinocerebellar ataxia type 3 (SCA3), also known as Machado-Joseph disease (MJD), is an untreatable autosomal dominant neurodegenerative disease, and the most common such inherited ataxia worldwide. The mutation in SCA3 is the expansion of a polymorphic CAG tri-nucleotide repeat sequence in the C-terminal coding region of the ATXN3 gene at chromosomal locus 14q32.1. The mutant ATXN3 protein encoding expanded glutamine (polyQ) sequences interacts with multiple proteins in vivo, and is deposited as aggregates in the SCA3 brain. A large body of literature suggests that the loss of function of the native ATNX3-interacting proteins that are deposited in the polyQ aggregates contributes to cellular toxicity, systemic neurodegeneration and the pathogenic mechanism in SCA3. Nonetheless, a significant understanding of the disease etiology of SCA3, the molecular mechanism by which the polyQ expansions in the mutant ATXN3 induce neurodegeneration in SCA3 has remained elusive. In the present study, we show that the essential DNA strand break repair enzyme PNKP (polynucleotide kinase 3’-phosphatase) interacts with, and is inactivated by, the mutant ATXN3, resulting in inefficient DNA repair, persistent accumulation of DNA damage/strand breaks, and subsequent chronic activation of the DNA damage-response ataxia telangiectasia-mutated (ATM) signaling pathway in SCA3. We report that persistent accumulation of DNA damage/strand breaks and chronic activation of the serine/threonine kinase ATM and the downstream p53 and protein kinase C-δ pro-apoptotic pathways trigger neuronal dysfunction and eventually neuronal death in SCA3. Either PNKP overexpression or pharmacological inhibition of ATM dramatically blocked mutant ATXN3-mediated cell death. Discovery of the mechanism by which mutant ATXN3 induces DNA damage and amplifies the pro-death signaling pathways provides a molecular basis for neurodegeneration due to PNKP inactivation in SCA3, and for the first time offers a possible approach to treatment.
| Spinocerebellar ataxia type 3 (SCA3) is an untreatable neurodegenerative disease, and the most common dominantly inherited ataxia worldwide. SCA3 is caused by expansion of a CAG tri-nucleotide repeat sequence in the ATXN3 gene’s coding region. The expanded CAG sequences encode a run of the amino acid glutamine; the mutant ATXN3 interacts with multiple proteins in vivo to create insoluble aggregates in SCA3 brains. It is thought that the loss of function of the aggregated proteins contributes to cellular toxicity and neurodegeneration in SCA3. Despite significant progress in understanding SCA3’s etiology, the molecular mechanism by which the mutant protein triggers the death of neurons in SCA3 brains remains unknown. We now report that the mutant ATXN3 protein interacts with and inactivates PNKP (polynucleotide kinase 3’-phosphatase), an essential DNA strand break repair enzyme. This inactivation results in persistent accumulation of DNA damage, and chronic activation of the DNA damage-response ATM signaling pathway in SCA3. Our work suggests that persistent DNA damage/strand breaks and chronic activation of ATM trigger neuronal death in SCA3. Discovery of the mechanism by which mutant ATXN3 induces DNA damage and amplifies the pro-death pathways provides a molecular basis for neurodegeneration in SCA3, and perhaps ultimately for its treatment.
| Spinocerebellar ataxia type 3 (SCA3), also known as Machado-Joseph disease (MJD), is an autosomal dominant neurodegenerative disease caused by CAG repeat expansion in the C-terminal coding region of the ATXN3 gene [1–3]. SCA3 is the most common dominantly inherited ataxia world-wide, and a late-onset disease that manifests with cerebellar ataxia, peripheral nerve palsy, and pyramidal and extrapyramidal signs [1–4]. SCA3 neurodegeneration is primarily observed in the brainstem, cerebellum, basal ganglia and spinal cord [5–8]. Ataxia symptoms appear between the ages of 20 and 50 years, and manifest with cerebellar ataxia, opthalmoplegia, dysarthria, dysphagia, dystonia, rigidity and distal muscle atrophies [1–3, 8, 9]. The wild-type ATXN3 gene encodes 12 to 41 CAG repeats in its 10th exon at the human chromosomal locus 14q32.1 [3]. ATXN3 is a deubiquitinating enzyme that edits specific poly-ubiquitin linkages [10, 11]. It has also been linked to transcriptional regulation [9, 12]. However, ATXN3 does not seem to be essential for brain development and function, as mice lacking ATXN3 do not develop overt neurological phenotypes [13]. Therefore, the exact function of ATXN3 remains unknown, limiting efforts to establish the possible role of mutant ATXN3 in eliciting neuronal death in SCA3. In SCA3, the polymorphic CAG repeats are expanded to 62 to 84 glutamines and the mutant ATXN3 forms aggregates that are deposited in SCA3 neurons [2, 3]. A large body of literature supports the hypothesis that multiple proteins aberrantly interact with the mutant ATXN3 and that the loss of function of the mutant ATXN3-interacting proteins contributes to neurodegeneration and SCA3 pathology [2, 8–9]. Recent studies have reported that depletion of the mutant ATXN3 allele in a SCA3 transgenic mouse brains rescues the molecular phenotypes of SCA3 supporting the hypothesis that mutant ATXN3 elicits toxicity and neuronal dysfunction in SCA3 [14]. Recent studies have also shown that the mutant ATXN3 causes p53-mediated neuronal death in vitro and in vivo by activating the transcription of the p53-inducibe pro-apoptotic genes such as BAX (Bcl2-associated X protein) and PMAIP1 (PUMA, p53 upregulated modulator of apoptosis), triggering mitochondrial apoptotic pathways [15, 16]. However, the mechanism by which mutant ATXN3 increases p53 phosphorylation and activates the p53-dependent pro-apoptotic signaling pathways to facilitate neuronal death and dysfunction remains unknown.
In the present study we show that PNKP (Polynucleotide kinase 3’-phosphatase), a dual- function DNA strand break repair enzyme [17, 18], is a native ATXN3-interacting protein, and is inactivated by its interaction with the mutant ATXN3 in SCA3. Our data also show that PNKP is also present, in part in the polyQ aggregates in SCA3 brain. Diminished PNKP activity results in persistent accumulation of DNA strand breaks, leading to chronic activation of the DNA damage-response ataxia telangiectasia mutated (ATM) protein kinase and the downstream pro-apoptotic p53-dependent signaling pathways in SCA3. Additionally, activated ATM stimulates phosphorylation of c-Abl tyrosine kinase, which phosphorylates and facilitates nuclear inclusion of protein kinase C delta (PKCδ), further amplifying pro-apoptotic output in SCA3. Either overexpression of PNKP or pharmacological inhibition of ATM in mutant ATXN3-expressing cells blocked aberrant activation of the pro-death pathways and reduced cell death, suggesting that mutant ATXN3-mediated chronic activation of the DNA damage-response ATM signaling pathway plays a pivotal role in neuronal dysfunction and neurodegeneration in SCA3. Therefore, our current study not only provides an insight into the mechanism of neurodegeneration in SCA3, but also delineates potential drug targets for developing mechanism-based efficacious therapeutic modalities to combat systemic degeneration of neuronal cells in SCA3.
Our studies described in the accompanying manuscript by Chatterjee et al suggest that PNKP is a native ATXN3-interacting protein, and that ATXN3 modulates PNKP activity and DNA repair (Chatterjee et al, Figs. 1–3). Immunoprecipitation of PNKP from the nuclear extract from human neuroblastoma SH-SY5Y cells and subsequent mass spectrometric analysis showed the presence of ATXN3 in the immunoprecipitated (IP) pellet; conversely, immunoprecipitation of ATXN3 and Western blot analysis revealed the presence of PNKP in the ATXN3 IP (Chatterjee et al, Figs. 1, S1, 2A and 2B). Further, GST pull-down from the nuclear extract, followed by Western blot analysis, indicated that both wild-type and mutant ATXN3 directly interact with PNKP in vitro, (Chatterjee et al; Fig. 2D). The wild-type ATXN3 protein stimulated, and in contrast, the mutant ATXN3 dramatically diminished, the 3’ phosphatase activity of PNKP in vitro (Chatterjee et al; Figs. 3A and 3B). The interaction between these two proteins was further validated in SH-SY5Y cells co-transfected with the plasmids pCherry-PNKP and pGFPC-ATXN3–28, expressing cherry-tagged PNKP and GFP-tagged ATXN3-Q28, respectively, and imaged by confocal microscopy. Analysis of the transfected cells showed significant co-localization of the red fluorescence of PNKP with the green fluorescence of ATXN3-Q28 (Fig. 1A). Similarly, cells co-transfected with pCherry-PNKP and pGFP-ATXN3-Q84 (a plasmid expressing mutant ATXN3-Q84 encoding 84 glutamines) showed marked co-localization of PNKP and ATXN3-Q84 (Fig. 1B). However, co-transfection of plasmid pCherry-PNKP and pAcGFPC1 (an empty control vector expressing GFP) did not show any detectable reconstitution of yellow/orange fluorescence (S1 Fig.), suggesting specificity of these interactions. Together, these data support our previous interpretation that both wild-type and mutant ATXN3 interact with PNKP in the cell (Chatterjee et al).
To further confirm the interaction of PNKP and ATXN3 in cell, we performed bi-molecular fluorescence complementation (Bi-FC) assays, a versatile method to assess in cell protein-protein interactions [19]. We cloned PNKP cDNA with the N-terminal amino acids of modified GFP into plasmid pBiFC-VN173, and ATXN3 cDNA (encoding 28 and 84 glutamines) with the C-terminal amino acids of modified GFP into plasmid pBiFC-VC155 (a description of these Bi-FC plasmids is provided in the Methods section). Transfection of pVN173-PNKP, pVC155-ATXN3-Q28 or pVC155-ATXN3-Q84 into SH-SY5Y cells individually did not reconstitute green/yellow fluorescence (Fig. 1C, panels 1–3). In contrast, co-transfection of plasmids pVN173-PNKP and pVC155-ATXN3-Q28 effectively reconstituted green/yellow fluorescence (Fig. 1C, panel 4). Importantly, co-transfection of plasmids pVN173-PNKP and pVC155-ATXN3-Q84 also resulted in robust reconstitution of green/yellow fluorescence (Fig. 1C, panel 5). These data substantiate our interpretation that both wild-type and mutant ATXN3 interact with PNKP in the cell. Furthermore, we analyzed these protein-protein interactions in SCA3 patients’ brain sections by proximity ligation assays (PLA), a widely used technique to assess in vivo protein-protein interactions [20]. The PLA analysis clearly shows a robust reconstitution of red fluorescence in both SCA3 and normal control brains, suggesting an in vivo interaction between ATXN3 and PNKP (n = 3; Fig. 1D). Importantly, about 70% of the ATXN3-PNKP complexes were detected in the nuclei in the control brain sections (Fig. 1D; panel 2 and 3). By contrast, PLA analysis of the SCA3 patients’ brain sections shows that the ATXN3-PNKP complexes are predominantly present in periphery or outside the nuclei (n = 3, Fig. 1D). Since PNKP is present in the mitochondria [18], the extra-nuclear signals detected in the control brain sections presumably are from the PNKP-ATXN3 complexes present in the mitochondria. To further verify the specificity of the in vivo interaction of ATXN3 and PNKP, we performed PLA analysis to check the interaction of ATXN3 with DNA ligase 3α (DNA LIG 3α), another critical DNA strand break repair enzyme present in the PNKP complex (Chatterjee et al; Figs. 2A, 2B and S2). The PLA analysis showed no significant interaction of ATXN3 with DNA LIG 3α in the brain sections from SCA3 patients, or in control brains under identical experimental conditions (Fig. 1D; panels 1 and 4), suggesting specificity of the interactions between ATXN3 and PNKP in vivo. Consistent with these data, PLA analysis also suggested specific and pronounced interactions of ATXN3 and PNKP in SH-SY5Y cells (accompanying manuscript, Chatterjee et al, Fig. 2C).
The in vivo interaction of PNKP and ATXN3 in SCA3 and control brain sections was further confirmed by immunostaining brain sections from SCA3 patients and control subjects with specific antibodies, followed by confocal microscopy. For immunostaining the brain sections we used an anti-PNKP antibody that shows high specificity for PNKP as evidenced by Western blot and immunohistochemical analyses (S2 Fig.). Image analysis revealed a distinct co-localization of PNKP with ATXN3 in the cerebellum of normal control brain and SCA3 (expressing mutant ATXN3 with 72 glutamines, ATXN3-Q72) brain sections (Figs. 2A and 2B). Likewise, analysis of SCA3 brain expressing either ATXN3-Q79 or ATXN3-Q84 also showed discrete co-localization of ATXN3 with PNKP (S3 Fig.). Importantly, consistent with the PLA data as described in Fig. 1D, majority (70 to 80%) of the ATXN3-PNKP complexes were detected in the periphery and/or outside of nuclei in the SCA3 brain sections (Figs. 2B and S3); in contrast, the ATXN3-PNKP complexes were predominantly detected inside the nuclei in the control brain sections (n = 3; Figs. 2A and S3). We next assessed in vivo interactions of ATXN3 and PNKP in transgenic mouse brains expressing mutant ATXN3, as well as in wild-type control mouse brains. Previous studies have shown that this novel SCA3 mouse model (CMVMJD135) develops SCA3-like motor incoordination and exhibits neurodegeneration in spite of the absence of aberrant cleavage of the mutant ATXN3 and accumulation of polyQ aggregates in brain [21]. We observed a distinct co-localization of PNKP with ATXN3 in the deep cerebellar nuclei area of the SCA3 mouse brain, as well as in age-matched control brain sections (S4 Fig.). To assess whether PNKP is present in the polyQ aggregates in SCA3 patients’ brains, brain sections from SCA3 and age-matched normal subjects were co-immunostained with an anti-polyQ antibody (5TF1–1C2; green) and anti-PNKP antibody (red). Confocal image analysis showed distinct co-localization of PNKP- and polyQ aggregates in brain sections from SCA3 patients, but not in control brain sections (Figs. 2C and 2D). The much less intense fluorescence signals observed in the brain sections from control subjects (Fig. 2C) presumably are from the shorter polyQ sequences present in the wild-type proteins, and do not show detectable co-localization with PNKP. Collectively, these data suggest that PNKP is a native wild-type ATXN3-interacting protein that also interacts with the mutant ATXN3, and at least in part is recruited into the polyQ aggregates in the SCA3 brain.
Recent studies have shown elevated levels of DNA damage/strand breaks in peripheral blood lymphocytes in SCA3 patients [22], indicating that mutant ATXN3 can induce DNA damage. Our data in the accompanying manuscript by Chatterjee et al suggest that mutant ATXN3 binds PNKP and inhibits its 3’-phosphatase activity (Chatterjee et al, Figs. 2D and 3B). We thus examined SH-SY5Y cells expressing ATXN3-Q84, brain sections from SCA3 patients, as well as SCA3 transgenic mouse brains, for the presence of DNA damage. Oxidative DNA damage or double-strand breaks rapidly induce the phosphorylation of several damage-response kinases and thereby facilitate protein-protein interactions that collectively regulate a signaling cascade to repair DNA lesions. ATM is one of the major DNA damage-response kinases, and is rapidly activated by autophosphorylation at S1981 after genomic damage [23]. Genomic DNA damage or strand breaks also result in rapid phosphorylation of histone H2AX and p53-binding protein 1 (53BP1), and the phosphorylated H2AX (γH2AX) and 53BP1 (p-53BP1) are quickly recruited into damage sites and visualized as nuclear foci in damaged cells [23]. Since mutant ATXN3 interacts with and inactivates PNKP, we next investigated whether ectopic expression of mutant ATXN3 induces DNA damage. To this end we developed SH-SY5Y cells expressing ATXN3-Q28 and ATXN3-Q84; Western blotting and confocal image analysis confirmed effective expression of the wild-type ATXN3-Q28 as well as mutant ATXN3-Q84 in these cells (Figs. 3A and 3B). To assess the accumulation of genomic DNA damage in cells expressing mutant ATXN3, we used anti-phospho-53BP1 (p-53BP1-S1778) antibody to perform immunohistochemical (IHC) analysis on the SH-SY5Y cells expressing wild-type vs. mutant ATXN3. We observed about a 5-fold increase in 53BP1 focus formation in the SH-SY5Y cells expressing ATXN3-Q84 over the cells expressing ATXN3-Q28 (Figs. 3C and 3D; the foci are shown by arrows). IHC analysis also revealed significantly more (~10-fold) γH2AX foci in the mutant ATXN3-expressing cells compared to control cells (Figs. 3E and 3F). Consistent with these results, immunohistological analysis of SCA3 brain sections with phospho-53BP1 (p-53BP1-S1778) antibody also showed widespread 53BP1 nuclear foci, unlike controls (n = 3) (S5 Fig.; the foci are shown by arrows). Further, comet analysis of neuronal cells from the deep cerebellar nuclei from a SCA3 transgenic mouse brain revealed the presence of DNA damage (S6A and S6B Figs.). Compared to control cells, DNA damage in the mutant cells was significantly increased when exposed to hydrogen peroxide (S6C, S6D and S6E Figs.). To further test our hypothesis that sequestration of PNKP can compromise the cellular DNA damage repair ability, resulting in increased accumulation of DNA lesions, we depleted PNKP in SH-SY5Y cells with siRNA. Western blotting confirmed about 70 to 80% depletion of PNKP in the PNKP-siRNA-treated cells (S7A and S7B Figs.). These cells showed about a 5-fold increase in the formation of 53BP1 foci compared to the control siRNA-treated cells (S7C and S7D Figs.). Likewise, the PNKP-depleted cells also showed about 10-fold more γH2AX focus formation vs. cells transfected with control-siRNA (S7E and S7F Figs.). These data clearly suggest that the perturbation of PNKP activity by the mutant ATXN3 impairs DNA repair efficacy and facilitates the accumulation of DNA strand breaks in SCA3.
Activated ATM coordinates cell cycle progression with the damage-response checkpoints and DNA repair to preserve genomic integrity, via a well-orchestrated signaling network [23]. To investigate whether mutant ATXN3 activates ATM signaling in SCA3, we expressed ATXN3-Q84 in differentiated SH-SY5Y cells and assessed activation of the ATM pathway. Expression of ATXN3-Q84 strongly activated the ATM pathway, inducing the phosphorylation of ATM and H2AX and ATM’s downstream substrates Chk2 and p53 (Figs. 4A and 4B). By contrast, expression of wild-type ATXN3-Q28 did not activate the ATM pathway (Figs. 4C and 4D), suggesting that mutant ATXN3 strongly activates the DNA damage-response pathway and the polyQ sequence length is important for ATM pathway activation. Likewise, expression of the mutant ATXN3 carrying 72 and 80 poly-glutamines (ATXN3-Q72 and ATXN3-Q80) in SH-SY5Y cells also strongly activated the DNA damage-response ATM pathway (S8 Fig.). Furthermore, to test whether mutant ATXN3 activates p53 and Chk2 via activating ATM, we pre-treated the cells with ATM inhibitor Ku55933 and expressed ATXN3-Q84 and assessed the activation of DNA damage response pathway. Consistent with our hypothesis, ATXN3-Q84 expression failed to stimulate phosphorylation of Chk2 and p53 in the presence of the ATM inhibitor Ku55933 (S9 Fig.), substantiating our interpretation that mutant ATXN3 stimulates the DNA damage response p53 pathway via activating ATM. The dramatic increase in ATM, H2AX, Chk2 and p53 phosphorylation (Figs. 4A and S8) and formation of 53BP1 and γH2AX foci (Fig. 3) in response to mutant ATXN3 expression suggest that mutant ATXN3-induced genomic DNA strand breaks/damage is sufficient to activate the DNA damage- response pathway. Further, analysis of the tissue from the deep cerebellar nuclei (DCN) from SCA3 transgenic mice (CMVMJD135 mice) constitutively expressing mutant ATXN3 showed robust activation of the ATM pathway (increased phosphorylation of ATM, H2AX and p53) (Figs. 4E and 4F), suggesting that mutant ATXN3 strongly activates the DNA damage-response pathway in vivo. To further test whether inactivation of PNKP by mutant ATXN3 stimulates the ATM pathway, we examined PNKP-siRNA-treated differentiated SH-SY5Y cells for ATM pathway activation. Our data showed robust activation of the ATM and p53 pathways in cells transfected with PNKP-siRNA, but not in cells transfected with control-siRNA (S10 Fig.). To rule out the possibility that DNA damage and subsequent activation of the DNA damage response might be due in part to non-specific off-target toxic effects of the PNKP-siRNA, we used micro-RNA-adapted RNA interference (shRNAmir) to achieve more specific knockdown of PNKP in cells, and assessed activation of the DNA damage-response pathway in these cells. Similar to our previous observation described in S7 Fig., depletion of PNKP in SH-SY5Y cells with PNKP-shRNAmir constructs also resulted in increased genomic DNA damage (53BP1 and γH2AX foci formation; shown by arrows; S11 Fig.), and marked activation of the DNA damage-response ATM pathway (S12 Fig.). Moreover, recent studies have indicated that a mutant ATXN3-mediated increase in oxidative stress might be responsible for inducing DNA damage and SCA3 pathology [22]. Since oxidative stress alone can activate the ATM pathway [24], we sought to determine whether mutant ATXN3 activates ATM via an oxidation-dependent mechanism. To test this possibility, we induced ATXN3-Q84 expression in differentiated SH-SY5Y cells pre-treated with the antioxidant N-acetyl cysteine (NAC). However, pre-treating cells with NAC did not block mutant ATXN3-mediated activation of the DNA damage-response pathway (S13A and S13B Figs.). Likewise, expression of ATXN3-Q84 strongly activated the ATM pathway in cells overexpressing the antioxidant enzyme catalase (S13C and S13D Figs.), suggesting that the mutant ATXN3-induced DNA damage-response ATM pathway activation is oxidation-independent.
The transcription factor p53 is the primary target molecule in the ATM pathway, and many of the functions of ATM are p53-dependent. Activated p53 regulates a variety of cellular processes, such as transcription, cell cycle regulation, DNA damage-response repair and cell death [25–27]. The mutant polyQ proteins have been shown to induce p53-dependent apoptosis in SCA3, Huntington’s disease and spinocerebellar ataxia type 7 (SCA7) [15, 28, 29]. Ectopic expression of ATXN3-Q79 in cultured cerebellar neurons results in p53 activation, increased expression of BAX, increased release of cytochrome c from mitochondria, and apoptotic cell death [16]. However, the mechanism by which the polyQ proteins stimulate p53 activation and apoptosis remains unknown. We transfected the SH-SY5Y cells with either PNKP-siRNA or control-siRNA to test our hypothesis that the loss of PNKP activity triggers the pro-apoptotic signaling pathways in SCA3. Western blotting showed markedly decreased PNKP levels in cells treated with PNKP-siRNA, but not with control-siRNA (Fig. 5A). The TUNEL staining (Fig. 5B), and increased (~2-fold) caspase-3 activities (Fig. 5C) of the PNKP-depleted cells suggest robust activation of the pro-death pathways when PNKP was depleted. We thus developed SH-SY5Y cells overexpressing exogenous PNKP (Fig. 5D) and expressed the mutant ATXN3-Q84 in these cells to test whether PNKP overexpression blocks ATXN3-Q84-mediated cell death. The Western blot analysis clearly shows that ATXN3-Q84 failed to activate the ATM pathway in cells overexpressing PNKP (Fig. 5E). Moreover, overexpression of PNKP in these cells blocked ATXN3-Q84-mediated caspase-3 activation (Fig. 5F), suggesting that the loss of PNKP function plays an important role in mutant ATXN3-mediated cell death. Furthermore, quantitative RT-PCR (qRT-PCR) analysis of total RNA from the PNKP-depleted SH-SY5Y cells showed stimulated transcription of the p53-dependent pro-apoptotic genes such as BAX, BBC3 (encoding PUMA), Bcl2L11 (encoding BIM) and PMAIP1 (encoding NOXA) (Fig. 5G). Moreover, we found that pre-treating the cells with either the ATM inhibitor Ku55933 or the p53 inhibitor Pifithrin-α could block PNKP-siRNA-induced caspase-3 activation (Figs. 5H and 5I). Likewise, expression of ATXN3-Q84 stimulated caspase-3 activity, whereas pre-treating the cells with Ku55933 or Pifithrin-α ameliorated the ATXN3-Q84-induced caspase-3 activation (S14A and S14B Figs.). These data substantiate our previous interpretation that mutant ATXN3 activates the p53-dependent pro-death pathway by activating ATM, and that chronic activation of the ATM→p53 pathway plays a pivotal role in mediating neuronal death in SCA3.
In response to DNA damage, the tyrosine kinase c-Abl (encoded by the mammalian homolog of the v-Abl oncogene from the Abelson murine leukemia virus) is phosphorylated by ATM and DNA-dependent protein kinase, DNA-PK [30–32]. Activated c-Abl constitutively associates with PKCδ, resulting in the latter’s phosphorylation and nuclear translocation [33, 34]. Cytosolic retention of PKCδ is required to maintain cell survival, whereas its phosphorylation and nuclear translocation activates an apoptotic pathway [33–36]. Since ATXN3-Q84-induced DNA damage strongly activated the ATM pathway, we explored the possibility that ATXN3-Q84 could induce the phosphorylation of the ATM target proteins c-Abl and PKCδ. Expression of ATXN3-Q84 indeed increased the phosphorylation of c-Abl (T735) and PKCδ (T311), and caspase-3 cleavage (Figs. 6A and S15A). However, pre-treating the cells with Ku55933 blocked these events (Figs. 6B and S15B). Furthermore, consistent with our data in S10 Fig. showing marked activation of ATM pathway upon PNKP depletion, we found increased phosphorylation of c-Abl and PKCδ and higher caspase-3 activity in cells treated with PNKP-siRNA (Figs. 6C and S15C), but not with control-siRNA (Figs. 6D and S15D). However, depleting PNKP in cells pre-treated with Ku55933 stimulated the DNA damage response (assessed by increased H2AX phosphorylation), but did not increase the phosphorylation of c-Abl and PKCδ or caspase-3 cleavage (S16 Fig.). To test whether ATXN3-Q84 activates PKCδ by activating c-Abl, we pre-treated the cells with the c-Abl kinase inhibitor STI-571; ATXN3-Q84 expression failed to enhance PKCδ phosphorylation in cells pre-treated with STI-571 (Figs. 6E and S15E). Moreover, PNKP depletion did not enhance PKCδ phosphorylation when cells were pre-treated with STI-571 (Figs. 6F and S15F), suggesting that mutant ATXN3 increases PKCδ phosphorylation by activating the ATM→c-Abl signaling pathway.
Since phosphorylated PKCδ is translocated to nuclei [33–34] and we observed that ATXN3-Q84 stimulates PKCδ phosphorylation (Figs. 6 and S15), we assessed the relative abundance of PKCδ in the sub-cellular compartments of SH-SY5Y cells expressing ATXN3-Q84 and ATXN3-Q28. Cells expressing ATXN3-Q84 showed higher nuclear levels of PKCδ (Fig. 7A; upper panel); by contrast, cells expressing ATXN3-Q28 showed predominantly cytosolic PKCδ (Fig. 7A, lower panel). Western blotting of the nuclear and cytosolic protein fractions showed higher nuclear levels of PKCδ in ATXN3-Q84-expressing cells than in control cells (Figs. 7B and 7C). Depletion of PNKP also resulted in marked nuclear accumulation of PKCδ (S17 Fig.). Furthermore, immunohistological analysis revealed that about 60 to 70% of the nuclei had significant nuclear accumulation of PKCδ in the SCA3 transgenic mouse brain sections (S18 Fig. lower panels; arrows); in contrast, control mouse brains showed predominantly cytoplasmic PKCδ (S18 Fig. upper panels; arrowheads).
We next assessed whether blocking PKCδ phosphorylation by inhibiting c-Abl kinase activity ameliorates mutant ATXN3-mediated caspase-3 activation. Pre-treating cells with STI-571 significantly inhibited mutant ATXN3-Q84-mediated caspase-3 activation (S19A Fig.). Likewise, pre-treating cells with STI-571 also ameliorated PNKP-siRNA-mediated caspase-3 activation (S19B Fig.). Collectively, these data, together with the data presented in the accompanying manuscript by Chatterjee et al, suggest that the wild-type ATXN3 interacts with and stimulates PNKP’s 3’-phosphatase activity, and this interaction possibly modulates the efficacy of DNA repair, and helps maintain genomic integrity and neuronal survival. In contrast, mutant ATXN3 interacts with PNKP and abrogates its 3’-phosphatase activity, resulting in increased accumulation of DNA damage that chronically activates ATM→p53 and ATM→c-Abl→PKCδ pro-apoptotic signaling to trigger neuronal dysfunction and apoptosis in SCA3 (illustrated in Fig. 8).
The DNA damage-response pathway is rapidly activated in response to DNA damage to repair the damaged sites by activating a well-orchestrated signaling network. However, if the DNA damage/lesions are irreparable, the damage-response pathway activates pro-death signaling cascades to ensure apoptotic demise of the damaged cells to maintain cell and tissue homeostasis. Due to their high metabolic activity, the post-mitotic neurons generate higher amounts of reactive oxygen species, and also have a higher risk of accumulating strand breaks due to their high transcriptional activity [37–41]. Therefore, post-mitotic neurons have an elaborate mechanism to repair DNA strand breaks/lesions to ensure longevity and functionality when subjected to insults [37, 38, 42, 43]. An emerging picture suggests that mutation or loss of function of DNA repair genes in post-mitotic neurons results in the accumulation of DNA damage/strand breaks, neuronal dysfunction and systemic neurodegeneration [44–49]. For example, mutations in the DNA repair enzymes APTX and TDP1 have been shown to contribute to neurodegeneration and the development of ataxia in autosomal recessive disorders such as AOA1 (ataxia with oculomotor apraxia type 1) [44, 45] and SCAN1 (spinocerebellar ataxia with axonal neuropathy) respectively [46]. Another recent discovery suggests that mutations in FUS (fused in sarcoma), a DNA repair protein that associates with the HDAC1-SIRT1 repair complex, result in the accumulation of DNA strand breaks, neurodegeneration and neurological defects in amyotrophic lateral sclerosis (ALS) [47]. Moreover, using genome-wide linkage analysis in consanguineous families of an autosomal recessive disease, multiple point mutations were identified in PNKP that result in neurological phenotypes characterized by microcephaly, intractable seizures, and developmental delay [48]. The presence of a homozygous frame-shift mutation in PNKP was recently identified in an early-onset neurodegenerative disorder that manifests with polyneuropathy, cerebellar atrophy, microcephaly, epilepsy and intellectual disability [49]. Disease-causing point mutations ablate either the kinase or phosphatase activities of PNKP in vitro [50]. These findings explain how the loss of function of essential DNA repair enzymes and subsequent defective DNA repair and accumulation of DNA damage results in neurological abnormalities.
Recent studies also indicate that persistent accumulation of DNA damage and inappropriate activation of the DDR pathway may contribute to the pathogenic mechanism of fragile X mental retardation syndrome (Fragile X syndrome), a common form of inherited mental retardation caused by the loss of the fragile X mental retardation protein, FMRP [51, 52]. Accumulating evidence suggests that in addition to the translational regulation in the cytoplasm, FMRP is also present in the nuclei, where it strongly associates with the chromatin and plays an important role in the regulation of DDR pathway, maintenance of genomic DNA sequence integrity and neuronal survival [52, 53]. A FMRP mutant Drosophila has been shown to be hypersensitive to genotoxic stress, and fails to survive to adulthood when exposed to stress [51]. The FMRP mutant fly also shows the presence of DNA damage and activation of the p53-dependent apoptotic pathways after irradiation [51]. These findings suggest that FMRP plays an important role in the maintenance of genomic DNA sequence integrity and neuronal survival. It is likely that the loss of FMRP may impair the efficacy of DNA repair, resulting in the persistent accumulation of neuronal DNA damage and inappropriate activation of the DNA damage response p53-dependent pro-apoptotic pathways to elicit neuronal death. However, further molecular and interventional studies are required to identify the interacting protein partners of FMRP in the nuclei to establish the exact role of FMRP in DNA repair, regulation of DDR pathway and maintenance of genomic DNA sequence integrity and to determine whether the loss of FMRP function contributes to aberrant activation of the DDR pathway and neurodegeneration in fragile X syndrome.
In the present study, we show that PNKP is inactivated via its interaction with the mutant ATXN3, and also in part due to its recruitment into the insoluble polyQ aggregates in SCA3 brain. Our data, together with the data presented in the accompanying manuscript by Chatterjee et al., strongly support our interpretation that interaction of PNKP with mutant ATXN3 and/or trapping of PNKP in the polyQ aggregates markedly abrogate PNKP’s enzymatic activity and DNA repair efficacy, resulting in persistent accumulation of strand breaks in SCA3. Decreased PNKP activity and the presence of high levels of strand breaks in the SH-SY5Y cells expressing ATXN3-Q84, in SCA3 patients’ brain and SCA3 transgenic mouse brains expressing mutant ATXN3, strongly support this idea. Furthermore, we demonstrate that mutant ATNX3 potently activates the DNA damage-response ATM signaling pathway in SCA3, while increased phosphorylation of ATM, H2AX, Chk2 and p53, and the formation of 53BP1 and γH2AX foci in the ATXN3-Q84-expressing cells, SCA3 patients’ brains and SCA3 transgenic mouse brains suggest that mutant ATXN3 induces genomic DNA damage and chronically activates the ATM pathway in SCA3. Moreover, amelioration of ATXN3-Q84-mediated phosphorylation of p53, c-Abl and PKCδ by pharmacological inhibition of ATM suggests that mutant ATXN3 activates the pro-death signaling cascades via activating ATM in SCA3. The occurrence of higher levels of genomic DNA damage has been reported in several neurodegenerative diseases such as Huntington’s disease, Parkinson’s disease, Alzheimer’s disease, and ALS [54–58]. However, it remains to be determined whether higher genomic DNA damage and aberrant activation of the DNA damage-response pathway contributes to neurodegeneration and the etiology of these diseases.
Our data described in the present manuscript establish a mechanistic link between mutant ATXN3 expression and aberrant p53 pathway activation in SCA3, as previously reported [16, 22]. The transcription factor p53 regulates the cell cycle, DNA damage-response repair and cell death [25, 26, 59]. Activated p53 initiates neuronal death by activating the expression of BAX, BBC3 and PMAIP1 [26, 59–64], and these factors stimulate apoptosis by enhancing mitochondrial membrane permeabilization that facilitates the release of cytochrome c and Smac/DIABLO [65–67]. Mutant polyQ-containing proteins have been shown to activate p53-dependent apoptosis in SCA3, SCA7 and Huntington’s disease [15, 16, 28, 29]. Our data show elevated mRNA levels of BAX, BBC3 and PMAIP1 in ATXN3-Q84-expressing and PNKP-depleted cells. The amelioration of apoptosis by pharmacological inhibition of ATM and p53, suggest that mutant ATXN3-mediated aberrant activation of the DNA damage-response pathway facilitates the apoptotic demise of neuronal cells in SCA3. Moreover, in response to DNA damage, the protein tyrosine kinase c-Abl is phosphorylated by ATM and DNA-dependent protein kinase (DNA-PK) [30–32]. Activated c-Abl in turn phosphorylates and facilitates the nuclear translocation of PKCδ [33, 34], whose cytosolic retention is required to maintain cell survival, whereas phosphorylation and nuclear translocation of PKCδ activates the apoptotic pathway [34–36]. Phosphorylation of PKCδ allows its association with importin-α, resulting in nuclear translocation of PKCδ and activation of apoptosis [33–36]. Consistent with these reports, our data show that mutant ATXN3-mediated activation of ATM→c-Abl pathway enhanced the phosphorylation and facilitated the nuclear translocation of PKCδ in cells, and in SCA3 transgenic mouse brains. It has been suggested that nuclear PKCδ phosphorylates p73 and DNA-PK to facilitate apoptosis [68]; however, it is not yet clear how PKCδ triggers apoptosis. We are currently investigating whether mutant ATXN3 enhances the phosphorylation of p73 and DNA-PK to induce apoptosis in SCA3.
In conclusion, our current study provides compelling evidence of how mutant ATXN3 impedes the enzymatic activity of PNKP, and induces DNA damage that manifests with activation of the pro-apoptotic signaling pathways in SCA3. Our data suggest that in addition to activating the DNA damage-induced p53 pathway as described earlier in SCA3 [15, 16], mutant ATXN3 also activates the ATM-dependent cAbl→PKCδ pro-apoptotic pathway in parallel to cause neuronal dysfunction and eventually facilitating systemic neuronal degeneration in SCA3 brain (Fig. 8). Understanding the aberrant interaction between PNKP and mutant ATXN3 that results in the sustained activation of the pro-death signaling pathways may provide important insights to develop novel, mechanism-based therapeutic strategies for SCA3. Specific modulation of mutant ATXN3-mediated atypical activation of the DNA damage-response p53 and PKCδ pathways, or enhancing the efficacy of in vivo DNA damage repair may be effective strategies to combat the pathways leading to systemic neurodegeneration in SCA3. However, further investigation and interventional studies are required to device strategies to block this deleterious protein-protein interaction to rescue neuronal dysfunction and demise of neuronal cells in SCA3.
Finally, CAG repeat instabilities and expansions are causal factors for several other poly-Q expansion-related neurodegenerative diseases e.g., SCA1, SCA2, SCA6, SCA7, SCA17, DRPLA, SBMA and Huntington’s disease [2, 69]. Is has been challenging to understand why these repeats show a pronounced region-specific instability pattern in brain. However, a recent study has clearly shown significantly elevated expression levels of various DNA repair proteins in the cerebellum compared to the striatum, and consequent higher repeat instability in the striatum of HD mouse model [70]. This study suggests that lower activities of various DNA repair enzymes might be responsible for the higher instability of the CAG repeats observed in the striatum in HD brain compared to other regions of the brain. These data strongly suggest that optimal activities of various DNA repair proteins act as a safeguard against repeat instability and thus reduce somatic instability of the repeats in the specific brain region, dictating the severity of disease pathologies [70]. It will be interesting to determine whether the expression levels and/or activity of PNKP vary in various brain regions, which may provide important insight to understand the molecular basis of region-specific repeat instability and differential vulnerability of specific brain regions in different poly-glutamine expansion-associated neurological diseases.
Plasmids pEGFP-Ataxin3-Q28, pEGFP-Ataxin3-Q84 and pFLAG-Ataxin3-Q80 were kindly provided by Dr. Henry L. Paulson (Addgene plasmids 22122, 22123 and 22129). Expression plasmid ATXN3-Q72 was kindly provided by Dr. Randall Pittman (University of Pennsylvania). The wild-type and mutant ATXN3 cDNAs were sub-cloned in pAcGFPC1 (Clontech, USA) to construct plasmids pGFP-ATXN3-Q28, pGFP-ATXN3-Q84 and pGFP-ATXN3-Q80 respectively. The plasmids pGFP-ATXN3-Q28, pGFP-ATXN3-Q80 and pGFP-ATXN3-Q84 were digested with AgeI and MluI, and the GFP-ATXN3-Q28, GFP-ATXN3-Q80 and GFP-ATXN3-Q84 fragments were sub-cloned into the Tet-inducible plasmid pTRE-3G (Clontech, USA) using appropriate linkers. The trans-activator plasmid pTet-ON-3G (Clontech, USA) and responder plasmids pTRE-GFP-ATXN3-Q84, GFP-ATXN3-Q80 or pTRE-GFP-ATXN3-Q28 was co-transfected into SH-SY5Y cells, and positive clones were selected with G418. Stable SH-SY5Y clones inducibly expressing GFP-ATXN3-Q84, GFP-ATXN3-Q80 and GFP-ATXN3-Q28 were incubated with medium containing doxycycline (500 ng/ml), and expression of the transgene was assessed by Western blotting using anti-ATXN3 antibody. The PNKP cDNA was PCR-amplified from plasmid pWZL-Neo-Myr-Flag-PNKP (kindly provided by Drs. William Hahn and Jean Zhao; Addgene plasmid 20594) and sub-cloned into pcDNA3.1-Hygro (Invitrogen, USA) to construct plasmid pRP-PNKP. The catalase cDNA was PCR-amplified from a cDNA clone with appropriate primers and was sub-cloned into plasmid pcDNA3.1-Hygro to construct plasmid pRP-Catalase. Plasmid pRP-PNKP or pRP-Catalase was transfected into SH-SY5Y stable cell lines encoding inducible ATXN3-Q28 and ATXN3-Q84, and the clones were selected against hygromycin resistance. The effective expression of the exogenous PNKP and catalase in these cells were assessed by Western blot analyses with appropriate antibodies. Anti-PNKP mouse monoclonal antibody was a kind gift from Prof. Michael Weinfeld (University of Alberta, Canada) and anti-catalase antibody (Cat # SC-50508) was purchased from Santa Cruz Biotechnology, USA. The PNKP cDNA was PCR-amplified from plasmid pWZL-Neo-Myr-Flag-PNKP and the PCR product was sub-cloned into XhoI- and BamHI-digested pmCherryC1 (Clontech, USA) to construct pCherry-PNKP expressing Cherry-tagged PNKP.
SH-SY5Y cells were purchased from ATCC and cultured in DMEM medium containing 15% FBS, and 1% B-27 supplement, and differentiated in DMEM medium containing 10% FBS, 1% B-27 supplement (Invitrogen, USA) and 20 μM retinoic acid. The SH-SY5Y stable cells encoding inducible GFPC1-ATXN3-Q28, GFPC1-ATXN3-Q72, GFPC1-ATXN3-Q80 and GFPC1-ATXN3-Q84 were differentiated for 7 days and transgene expression in the differentiated cells was induced by adding doxycycline to the medium to a final concentration of 500ng/ml. The siRNA duplexes for PNKP and control scrambled siRNA duplexes with same base composition were purchased from Sigma, USA, and transfection of the PNKP-siRNA duplexes into SH-SY5Y cells or differentiated SH-SY5Y cells were performed using Lipofectamine RNAi-MAX reagent (Invitrogen, USA). Plasmids encoding micro-RNA-based PNKP-shRNAmir and control-shRNAmir were purchased from Thermo Scientific, USA, and transfected into SH-SY5Y cells using Lipofectamine 2000 reagent (Invitrogen, USA).
Human autopsy specimens were obtained from SCA3 patients and control subjects in accordance with local legislation and ethical rules. Control brain samples were collected from age-matched individuals who were deemed free of neurodegenerative disorders. The SCA3 brain tissue samples used for this study were obtained from SCA3 patients who were clinically characterized by cerebellar ataxia, opthalmoplegia, dysarthria and dysphagia. The molecular diagnosis of SCA3 was established by analyzing genomic DNA extracted from peripheral blood using a combination of PCR and Southern blotting. The exact lengths of the expanded CAG repeat sequences in ATXN3 gene were established by sequencing the CAG repeat expansion loci of the mutant allele. All brain autopsies were frozen in liquid nitrogen immediately after surgery, and stored in a −80°C freezer until further analysis.
The MJD transgenic mice used in this study were previously described [21]. CMVMJD135 mice express the human ATXN3 carrying 135 glutamines and develop a progressive neurological phenotype with onset at 6 weeks, which includes loss of strength, impairment of motor coordination, loss of balance and altered reflexes. At late stages they show an overall reduction in brain weight, with reduced volume and/or total cell number in pontine nuclei and deep cerebellar nuclei, and intranuclear ATXN3 inclusions in several disease-relevant regions of the brain and spinal cord. Transgenic mice and control non-transgenic littermate mice (n = 4–5 pools of two animals per genotype) with a mean age of 20 weeks were sacrificed by decapitation, and brain slices were obtained for the macrodissection of the deep cerebellar nuclei using a stereomicroscope (Model SZX7, Olympus America Inc., USA) and frozen at −80ºC. For immunofluorescence assays, transgenic and wild-type littermate mice (mean age: 24 weeks) were deeply anesthetized and transcardially perfused with sterile PBS followed by 4% paraformaldehyde (PFA) in PBS. Brains were post-fixed overnight in fixative solution and embedded in paraffin. Slides with 4-μm-thick paraffin sections were processed for immunostaining with anti-PKCδ, -ATXN3 and—PNKP antibodies.
The SCA3 transgenic mice (CMVMJD135 transgenic mice) were sacrificed and the brain tissues were collected according to the standard approved procedure and national and international guidelines and animal protocols were strictly followed.
Alkaline comet assays were performed using a Comet Assay Kit (Trevigen, USA). The brain tissue was homogenized in PBS at 4°C and sifted through a 300 μm sieve. Cells were suspended in 85 μL of ice-cold PBS and gently mixed with an equal volume of 1% low-melting agarose. The cell suspension was dropped onto an agarose layer, and incubated in lysis buffer for 1 hour. After lysis, slides were incubated in buffer containing 0.3 M NaOH, 1 mM EDTA (pH 13) for 40 min, and electrophoresed for 1 hour. After neutralization, slides were stained and analyzed with a fluorescent microscope. To assess genomic DNA damage after treatment with genotoxic agents, cells were treated with 10 μM hydrogen peroxide (H2O2) for 20 minutes in serum-free medium, washed twice with ice-cold PBS and subjected to comet analysis as described above.
The antibodies for p53 (Cat # 9282), P-p53-S15 (Cat # 9286), P-p53-S20 (Cat # 9287), P-p53-S46 (Cat # 2521), Chk2 (Cat # 2662), P-Chk2-T68 (Cat # 2661), c-Abl (Cat # 2862), P-cAbl-T735 (Cat # 2864), PKCδ (Cat # 2058), P-PKCδ-T311 (Cat # 2055), p-53BP1-S1778 (CAT # 2675) and caspase-3 (Cat # 9668) were from Cell Signaling, USA; anti-H2AX (Cat # ab11175) and P-γH2AX-S139 (Cat # ab11174) were from Abcam, USA; anti-ATM (Cat # 1549–1) and P-ATM-S1981 (Cat # 2152–1) were from Epitomics, USA; anti-ataxin-3 rabbit polyclonal antibody (cat # 13505–1-AP) was from Protein tech, USA; anti-ataxin-3 mouse monoclonal antibody (Cat # Mab5360) was from Millipore, USA; and anti-polyQ diseases marker antibody, 5TF1–1C2 (cat # Mab1574) was from Millipore, USA. Cell pellets or mouse brain tissues were homogenized and total protein was isolated using a total protein extraction kit (Millipore, USA). The cytosolic and nuclear fractions were isolated from SH-SY5Y cells expressing ATXN3-Q84 and ATXN3-Q28 using a NE-PER nuclear protein extraction kit (Thermo Scientific, USA). The Western blot analyses were performed according to the standard procedure and each experiment was performed a minimum of 3 times to ensure reproducible and statistically significant results.
Bi-molecular fluorescence complementation assays were performed as described previously by Shyu et al [19]. Plasmids pBiFC-VN173 (encoding 1 to 172 N-terminal amino acids of modified GFP) and pBIFC-VC155 (encoding 155 to 238 C-terminal amino acids of modified GFP) were kindly provided by Dr. Chang-Deng Hu, Purdue University (Addgene plasmids 22011 and 22010). The ATXN3 cDNA (encoding 28 and 84 glutamines) were cloned in-frame with the C-terminal amino acids of modified GFP in plasmid pBiFC-VC155 and the PNKP cDNA was cloned in-frame with the N-terminal amino acids of modified GFP in plasmid pBIFC-VN173. SH-SY5Y cells (2×104 cells) were grown on chamber slides and transfected after 24 hours of plating. Plasmids pVN173-PNKP and pVC155-ATXN3-Q28 or pVN173-PNKP and pVC155-ATXN3-Q84 were co-transfected into SH-SY5Y cells and reconstitution of green/yellow fluorescence was monitored by fluorescence microscopy to assess bimolecular protein-protein interactions. Transfections of pVC155-ATXN3-Q28, pVC155-ATXN3-Q84 and pVN173-PNKP as single plasmids into SH-SY5Y cells were used as negative controls.
Expression of ATXN3-Q28 and ATXN3-Q84 in SH-SY5Y cells was induced by incubating the cells with doxycycline (500 ng/ml), and after 48 hours of incubation, the cells were fixed with 4% paraformaldehyde in PBS for 30 minutes. For the PNKP depletion experiment, SH-SY5Y cells were grown on cover slips and transfected with PNKP-siRNA; 48 hours after transfection, the cells were fixed with 4% paraformaldehyde for 30 minutes. The fixed cells were immunostained with anti-p-53BP1, anti-γH2AX or anti-PKCδ antibodies. Frozen brain sections from SCA3 patients and control subjects were fixed in 4% paraformaldehyde, washed with PBS, and immunostained with anti-p-53BP1 and γH2AX antibodies. Paraffin-embedded transgenic and control mouse brain sections were deparaffinized and rehydrated, fixed in 4% paraformaldehyde for 30 minutes, washed, and immunostained with anti-p-53BP1, anti-γH2AX, and PKCδ antibodies. Slides were washed according to our standard protocol and the nuclei stained with DAPI (Molecular Probe, USA) and photographed under a confocal microscope. TUNEL staining of the SH-SY5Y cells transfected with PNKP-siRNA and control-siRNA was performed using an in situ Apoptosis Detection Kit per the manufacturer’s protocol (Calbiochem, USA).
SH-SY5Y cells or differentiated SH-SY5Y cells were incubated with 5mM of N-acetyl cysteine (NAC), the ATM inhibitor Ku55933, p53 inhibitor Pifithrin-α or c-Abl kinase inhibitor STI-571 for 3 hours before the expression of ATXN3-Q28 and ATXN3-Q84. The ATM inhibitor Ku55933 was purchased from EMD Biosciences, USA and Pifithrin-α and STI-571 were purchased from Santa Cruz Biotechnology, USA. Drugs were added to the cell culture medium to a final concentration of 1 µM, 2 µM or 5 μM and incubated for 3 hours before transgene induction; fresh medium with drugs was replaced after 12 hours. Cells were harvested at various time points for Western blot analyses and for isolating total RNA and qRT-PCR analyses.
Images were collected using a Zeiss LSM-510 META confocal microscope with 40X or 60X 1.20 numerical aperture water immersion objective. The images were obtained using two different lines of excitation (488 and 543 nm) by sequential acquisition. After excitation with 488 nm laser line emission was measured with a 505–530 nm filter and after excitation with 543 laser line emission was measured with a 560–615 nm filter. All images were collected using 4-frame-Kallman-averaging with a pixel time of 1.26 μs, a pixel size of 110 nm and an optical slice of 1.0 μm. Z-stack acquisition was done at z-steps of 0.8 μm. All orthogonal views were done with LSM 510 software at the Optical Microscopy Core Laboratory of UTMB.
Caspase-3 activities were measured using a Caspase-3 activity assay kit (BD Biosciences, USA). The Caspase 3 assay kit is based on the hydrolysis of the substrate, acetyl-Asp-Glu-Val-Asp p-nitroanilide (Ac-DEVD-pNA) by caspase 3, resulting in the release of the p-nitroaniline (pNA) moiety from the substrate, and released p-Nitroaniline (pNA) is detected at 405 nm. Comparison of the absorbance of pNA from the sample with the control allowed determination of the fold increase in caspase-3 activity and the relative caspase-3 activities are expressed in arbitrary units. SH-SY5Y cells encoding ATXN3-Q28 and ATXN3-Q84 were cultured in DMEM (Invitrogen, USA) containing 15% FBS and were cultured on 96 well dishes in DMEM (Invitrogen, USA) containing 15% FBS. Expression of ATXN3-Q28 and ATXN3-Q84 were induced in presence or absence of the drugs Ku55933 (2 μM), Pifithrin-α (10 μM) or STI-571 (5 μM) with doxycycline (500 ng/ml). The cells were harvested 3 days after induction and caspase-3 activities were measured. For measuring caspase-3 activities in the PNKP and control-siRNA-treated cells, the SH-SY5Y cells were cultured in 96-well dishes and transfected with PNKP-siRNA or control-siRNA in the presence or absence of Ku55933 (2 μM), Pifithrin-α (10 μM) or STI-571 (5 μM). The cells were harvested 48 hours after transfection, and caspase-3 activities were measured.
Total RNA was extracted from SH-SY5Y cells expressing ATXN3-Q28, ATXN3-Q84, PNKP-siRNA and control-siRNA using an RNA extraction kit (Qiagen, USA) and purified using the TURBO DNA-free DNAse Kit (Ambion, USA). Brain tissue (deep cerebellar nuclei) from SCA3 transgenic mouse expressing ATXN3-Q135 was homogenized in trizol reagent (Invitrogen, USA) and RNA was isolated as above. 1μg of total RNA was reverse-transcribed using an RT-PCR kit (Clontech, USA). A cDNA aliquot from each reaction was quantified and 500ng of cDNA from each reaction was used for real-time qRT-PCR. The qRT-PCR reactions were repeated three times; primers used for the analyses (BAX: PPH00078B; BBC3: PPH02204C; PMAIP1: PPH02090F; BCL2L11: PPH00893) were purchased from Qiagen, USA, and tested for accuracy, specificity, efficiency and sensitivity by the manufacturer.
PLA assays were performed by the method we described previously [18]. In brief, SCA3 and normal brain sections were fixed with 4% paraformaldehyde, permeabilized with 0.2% Tween-20 and washed with 1X PBS. Brain sections were incubated with primary antibodies for PNKP (mouse monoclonal) and ATXN3 (anti-ATXN3; rabbit polyclonal) or PNKP (mouse monoclonal) and DNA ligase 3 (rabbit polyclonal). These samples were subjected to PLAs using the Duolink PLA kit from O-Link Biosciences (Uppsala, Sweden). The nuclei were counterstained with DAPI, and the PLA signals were visualized in a fluorescence microscope (Nikon) at 20× magnification.
All tabulated data are expressed as mean ± SD, except where otherwise indicated. Differences between mean values of two groups were analyzed by Student’s t tests after checking for variance distribution via Levene’s test. We tested all data for normal distribution using the Kolmogorov-Smirnov test, followed by two-way ANOVA test to evaluate overall group differences. This was followed by Tukey’s post-hoc test to determine pair-wise significance if the ANOVA test indicated that a significant difference was present in the data set. In all cases, probability values of 0.05 or less were considered to be statistically significant.
|
10.1371/journal.pcbi.1003772 | Modeling Viral Evolutionary Dynamics after Telaprevir-Based Treatment | For patients infected with hepatitis C virus (HCV), the combination of the direct-acting antiviral agent telaprevir, pegylated-interferon alfa (Peg-IFN), and ribavirin (RBV) significantly increases the chances of sustained virologic response (SVR) over treatment with Peg-IFN and RBV alone. If patients do not achieve SVR with telaprevir-based treatment, their viral population is often significantly enriched with telaprevir-resistant variants at the end of treatment. We sought to quantify the evolutionary dynamics of these post-treatment resistant variant populations. Previous estimates of these dynamics were limited by analyzing only population sequence data (20% sensitivity, qualitative resistance information) from 388 patients enrolled in Phase 3 clinical studies. Here we add clonal sequence analysis (5% sensitivity, quantitative) for a subset of these patients. We developed a computational model which integrates both the qualitative and quantitative sequence data, and which forms a framework for future analyses of drug resistance. The model was qualified by showing that deep-sequence data (1% sensitivity) from a subset of these patients are consistent with model predictions. When determining the median time for viral populations to revert to 20% resistance in these patients, the model predicts 8.3 (95% CI: 7.6, 8.4) months versus 10.7 (9.9, 12.8) months estimated using solely population sequence data for genotype 1a, and 1.0 (0.0, 1.4) months versus 0.9 (0.0, 2.7) months for genotype 1b. For each individual patient, the time to revert to 20% resistance predicted by the model was typically comparable to or faster than that estimated using solely population sequence data. Furthermore, the model predicts a median of 11.0 and 2.1 months after treatment failure for viral populations to revert to 99% wild-type in patients with HCV genotypes 1a or 1b, respectively. Our modeling approach provides a framework for projecting accurate, quantitative assessment of HCV resistance dynamics from a data set consisting of largely qualitative information.
| Hepatitis C virus (HCV) chronically infects approximately 170 million people worldwide. The goal of HCV treatment is viral eradication (sustained virologic response; SVR). Telaprevir directly inhibits viral replication by inhibiting the HCV protease, leading to high SVR rates when combined with pegylated-interferon alfa and ribavirin. Telaprevir-resistant variants may be detected in the subset of patients who do not achieve SVR with telaprevir. While the clinical impact of viral resistance is unknown, typically the telaprevir-sensitive virus re-emerges after the end of treatment due to competition between the telaprevir-sensitive and resistant variants. Previous estimates of these competition dynamics were obtained from population sequence data, which are qualitative and have a limited sensitivity of ∼20%. We sought to improve these estimates by combining these data with clonal sequence data, which are quantitative and have a sensitivity of ∼5%, and using quantitative modeling. The resulting model, which was verified with an independent data set, predicted that the median time for telaprevir-resistant variants to decline to less than 1% of the viral population was ≤1 year. Our modeling approach provides a framework for accurately projecting HCV resistance dynamics from a dataset consisting of largely qualitative information.
| Hepatitis C is an inflammatory infection of the liver caused by the hepatitis C virus (HCV). HCV chronically infects approximately 170 million people worldwide [1]. HCV infection is a major risk factor for cirrhosis and hepatocellular carcinoma, and has become one of the leading causes of both liver transplant and cancer-related death in the United States [2], [3]. In contrast to other chronic viral diseases such as HIV and HBV, the goal of HCV treatment is eradication of the virus as determined by achievement of a sustained virologic response (SVR).
Telaprevir is a direct-acting antiviral that inhibits viral replication by binding to the active site of the HCV NS3-4a protease, an enzyme essential for viral replication [4]–[6]. In combination with pegylated-interferon alfa (Peg-IFN) and ribavirin (RBV), telaprevir increased SVR rates over Peg-IFN/RBV alone [7], [8]. Telaprevir exerts a strong directional selective pressure on the viral population, which leads to enrichment of variants with decreased sensitivity to the inhibitor. These telaprevir-selected variants have been well characterized and occur at or near the catalytic site of the protease, resulting in decreased sensitivity to telaprevir and other HCV protease inhibitors [9]. Given that other protease inhibitors besides telaprevir may be included as components of future drug regimens for patients that fail a telaprevir-based regimen, presence of telaprevir resistant variants may limit future treatment options. It is therefore essential to understand viral evolutionary processes and the rates at which telaprevir-resistant variants are outcompeted by wild-type virus to optimally inform patient's future treatment options.
In clinical studies, telaprevir-resistant variants were identified in the majority of patients who did not achieve an SVR with telaprevir treatment [10]–[13]. Monitoring of telaprevir-resistant variants after treatment failure revealed that these variants tend to be replaced over time by telaprevir sensitive, wild-type (WT) virus [14], presumably due to the lower intrinsic fitness of the resistant variants [10], [11]. However, monitoring was performed only by direct sequencing of RT-PCR products amplified from HCV RNA extracted from patient plasma (i.e., ‘population sequencing’). Population sequencing infers genetic variation within a population from polymorphic peaks within sequence chromatograms, and therefore provides only qualitative information about the frequency of the variants. Industry-standard interpretation of these data assumes a limit of sensitivity of minority variant detection of ∼20% (see, for example, [12], [15]–[18]). Thus, although these clinical studies analyzed a large number of samples [14], the interpretation of the results is limited by the sensitivity of the assay used.
The analysis reported here builds upon the work of Sullivan et al. [14] in two key ways. First, the data set was augmented with clonal sequence data from a subset of the patients previously analyzed. The previous work used only population sequence analysis. Compared to population sequence analysis, clonal sequence analysis has a LOD of ∼5% and provides a quantitative estimate of the frequency of resistant variants in a patient sample. Second, computational modeling was used to integrate the qualitative information provided by population sequence data with the quantitative information provided by clonal sequence data. Overall, the combination of these two approaches resulted in a more quantitative and patient-level understanding of resistant variant evolution which allows extrapolation of viral resistance quantification both beyond the sampling period and to a greater sensitivity than can be estimated by the bioanalytical methods alone.
In Sullivan et al. [14], we performed Kaplan-Meier analysis across a population of patients to determine the ‘time-to-wildtype’. ‘Wildtype’ (WT) simply indicated that resistance was present at 20% or less of the viral population. From both a scientific and clinical standpoint, this answer is unsatisfactory: 19% of the viral population could be resistant by population sequencing and represent a major undetectable reservoir of drug resistance. Our modeling here addresses this significant shortcoming by allowing calculation of time-to-event analyses across patients based on levels of resistance below the sequencing assay limits of detection (e.g., 1%). Additionally, in our present analysis, we model rates of viral decline within individual patients. This ‘within patient’ analysis allows calculation within an individual of the rate of decline of resistant virus, and therefore allows (1) extrapolation of resistance levels at time points which were not sampled and (2) calculation of the time it takes for a given patient to revert to a given % resistance (e.g., 1%). Significantly, population level analyses can be performed across patients for any of these metrics. As such, our models and pipeline described here form a framework for future antiviral resistance monitoring programs. We propose that this methodology could be applied across both antiviral and antibacterial therapeutics.
To illustrate the concept of resistance monitoring, Figure 1 displays hypothetical viral dynamics for a patient who experiences viral breakthrough before Week 12 of treatment. Up to Week 12, the WT virus is strongly suppressed by the treatment, whereas some resistant variants may be able to replicate even in the presence of the treatment. At Week 12, all treatment is terminated, and the competition between WT and resistant virus continues in the absence of drug selective pressure. By Week 24, the WT virus overtakes the resistant variant population as the dominant viral species, and the level of the resistant variant continues to decline (Figure 1A). Resistance monitoring in patients with HCV infection can effectively quantify the frequency of resistant virus over time, when the total viral load exceeds the sequencing limit of detection (Seq. LOD; see Figure 1). In the example shown, sequence analysis can be performed at the beginning of treatment and after treatment has stopped, as denoted by the solid red line. This analysis is focused on the latter portion of this time frame, once treatment is terminated. Dynamic sequence data from 388 genotype 1 HCV patients who did not achieve an SVR with a telaprevir-based regimen were available for the analysis. Figure S1 presents more detailed information on the number of population and clonal sequence data points obtained per patient.
The logistic model given by Equation 3 was used to describe the dynamics of the resistant virus for each individual patient. The model assumed that the treatment-free equilibrium level of resistance is 0%. To fit this model to the available data, a two-step estimation procedure was used. First, we fit the model to the subset of patients who had both clonal (quantitative) and population (qualitative) sequence data after treatment failure. For model fitting, each qualitative (population sequence) data point was converted into a ranged value, with binned resistance values of (1) 0 to 20% (i.e., WT population result), (2) 20 to 80% (i.e., polymorphic population sequence result), or (3) 80 to 100% (i.e., resistant population result). HCV genotypes 1a and 1b were fit independently because their resistance profiles are different [14]. We used this fitted subset to approximate the expected prior parameter distributions for each genotype using a log-normal distribution. We then refit all patients individually, including those who had clonal sequence data (i.e., those that had already been fit), approximating the estimated parameters as distributed according to these priors. The resulting fits for all patients are shown in Figure 2. For patients who had only population sequence (qualitative) data, imposing this prior allowed us to determine a unique set of parameters for those individual patients that fit the data.
To examine how well the model fit the data across all patients, histograms of the objective function values (φ) were generated (Figure 3(A) and Figure 4(A)). Each φ value represents a scalar quantity that can be used to assess how well the model fits the result for a single patient, with larger φ values signifying worse fits to the data. In Figure 3 (B–E) and Figure 4 (B–E), representative fits for individual patients are shown for various quantiles to demonstrate the quality of fit for different φ values.
To assess the predictive capability of the model, we compared model predictions to resistance quantified by massively parallel sequencing (deep sequencing; DS) from a subset of samples included in our modelling analysis using an Illumina platform (LOD: 1%). These data provided a more quantitative measure of resistance as resistance from these samples had previously been determined using only population sequencing. As the patients and time points chosen for quantification were selected and analyzed independently of this modeling work [19], they served as an independent validation of our approach. Because we fit our model (Equation 3) to each individual patient, we were able to predict, for each patient with DS data, the expected resistance at the precise time of the DS. There were 52 time points sequenced, each from a different patient. 32 samples had no detectable resistant virus (below the LOD), and the remaining 20 samples had 1.06%–98.58% resistant virus. For samples with no detectable resistance, the model predicted that the majority (n = 25; 78%) should have resistance values below 1% (Figure 5(A)). For all samples sequenced, the model predictions showed good agreement with the sequence results (Figure 5(B)). To determine the significance of these results, Monte Carlo simulation was used to generate 104 equivalently sized datasets for the DS samples. None of these datasets had a smaller sum of squared errors than that observed in this analysis, indicating that our observations have a <10−4 probability of being generated by chance alone. Additionally, we assessed the null hypothesis that the difference between the actual and predicted % resistance values was equal to 0 (Figure 5(C)). After arcsine square root transformation of the differences, neither a t-test (p = 0.86) nor a Wilcoxon-signed rank test (p = 0.16) suggest a significant difference between the actual and model-predicted results.
The model was used to predict population statistics for reversion of virus from the resistant to the WT, non-resistant state. Specifically, Kaplan-Meier analysis was used to calculate the median time to reversion for each HCV subtype. Previously, the estimated time it takes for a population of patients to revert from resistant to WT virus was calculated using only population sequence data [14]. Given the 20% sensitivity of that sequence method, this time was considered to represent the ‘time-to-20%.’ Here, those results were compared against the model derived time-to-20% estimates. The modeled Kaplan-Meier analysis of this reversion was less than (genotype 1a) or equal to (genotype 1b) the time-to-20% estimated by population sequence data (Figure 6, Table 1). For both genotypes, the upper 95% confidence interval for the median model prediction is lower than that for the population sequence result.
Notably, fewer patients are censored in the modeled results as compared with the population sequence results (Table 1). This difference results from patients whose last population sequence data point is polymorphic (i.e., between 20% and 80% resistant). Because resistance is still present, the direct population sequence-based Kaplan-Meier survival analysis censors these points. In contrast, the model can predict reversion times for these patients. For patients that are considered to have achieved ≤20% resistance by the model and population sequence data (i.e., patients whose last sequenced time point had no detectable resistance by population sequence data), the model consistently predicts shorter times-to-20% resistance than the population sequence results (Figure 7).
The predicted time-to-1% resistance was also determined as this value represents the measure of resistance theoretically obtainable by recent massively parallel sequencing approaches (e.g. [19]). Interestingly, these predictions are similar to the estimated time-to-20% resistance determined using population sequence data alone (Table 1, Figure 6). Note that these predictions did not account for uncertainty in the parameter estimates for individual patients. Monte Carlo simulations of this uncertainty suggested that it minimally affected the median reversion time determined by Kaplan-Meier analysis (see Text S1).
The predicted time-to-1% resistance was also used to assess the effect of a number of covariates on reversion times. The covariates assessed were: (1) baseline resistance status, (2) failure modality, (3) prior treatment status, and (4) the length of time PR treatment persisted after the time of treatment failure (Figure S2). Of these covariates, the presence of resistance at baseline exerted a large effect on resistant variant retention after treatment failure which was statistically significant in the case of genotype 1a infections. The data suggest that resistance retention rates are greater for patients that already had resistance variants present at baseline prior to treatment. Similarly, the data suggest that genotype 1a infected patients who experience on treatment virologic failure (as compared to patients that relapse after the end of treatment) as a group have statistically significantly longer retention times of resistant variants. Significantly, by protocol, a subset of patients that experienced on treatment virologic failure continued to be treated with pegylated interferon and ribavirin (PR) after they experienced viral breakthrough. Consequently, Cox proportional hazards analysis of this covariate is largely overlapping with the analysis of failure modality and suggests that continued dosing with PR after treatment failure also increases the time that it takes for a viral population to revert to 1% resistant. In contrast to these two covariates, prior treatment status did not provide any strong signals for affecting the retention time of resistant variants.
Previously, resistance monitoring of patients who did not achieve an SVR with telaprevir-based treatment in the Phase 3 studies ADVANCE, REALIZE, and ILLUMINATE showed that, after treatment failure in the absence of drug, resistant variants decline over time and are replaced by WT (drug-sensitive) virus [14]. One limitation of this previous work is that the analysis included only qualitative data obtained from population sequencing [14]. To provide a more quantitative understanding of viral dynamics after telaprevir-based treatment, we generated a quantitative clonal sequence dataset for a subset of the patient samples. We then employed mathematical modeling to continuously fit both the quantitative (clonal sequence) and qualitative (population sequence) datasets. The model explored the rate at which resistant virus reverts to WT virus (see Figure 1).
Of note, existing methodology for analysis of population sequence results allows estimation of resistance levels at only discrete time points, and therein can only describe resistance levels in gross bins (e.g., 0–20%; 20–80%, or 80–100%). By developing patient-level models that describe how resistance levels change over time, and by merging population and clonal sequence data, our analysis provides substantial additional advantages:
Overall, the results suggest that the model captures the resistance dynamics for the majority of patients quite well (Figure 3 and Figure 4). We found that numerous patients with the best model fits (φ≤10−10) had only population sequence data. We observed that the resistance dynamics for these patients were well fit by the population average parameters. Of note, one of the reasons why these patients' dynamics were well fit may be explained by the lack of specific information within the population sequence dataset since data are binned into ranges of between 0 and 20%, 20 and 80%, and 80 and 100% resistance. As such, many fits are possible through many of the population sequence curves which are consistent with the observed population sequence results.
Figure S3 illustrates this point as substantial variability in the model fits is observed in the two patients (A and B) having only population sequence data, whereas the variability is notably diminished in the patient C having both population and clonal sequence data.
In order to validate the model, we compared the quantitative model predictions of % resistance against actual quantitative results generated by an independent test set (DS). Monte Carlo simulation analyses suggest that the model predictions of % resistance for this test set are accurate with the null hypothesis of equivalence between the methods not refuted. The consistency between the model predictions and the DS results (1% resistance sensitivity) suggests that the model can accurately predict the resistance dynamics between 100% and 1% resistance even though neither of the sequence data types (population and clonal) used to train the model had sensitivities below 5%.
The model cannot fit two modalities of viral evolution. First, the model cannot fit patients whose % resistance increases over time because the model's logistic expression decreases monotonically over time. For example, as in the rare case shown in Figure 4(E), the measured resistance dynamics start at 0–20% resistance, change to 80–100% resistance, and then revert again to 0–20% resistance. Not surprisingly, this patient had the worst fit viral dynamics for the genotype 1b population. Such phenomena were observed infrequently, and are likely attributable to the stochasticity associated with PCR amplification in populations with HCV RNA levels near the assay LOD.
Second, the model does not accurately fit virus that appears to have a natural resistance level greater than 1% (e.g., Figure 3(E)). While an equilibrium resistance level of ∼20% would result in a better fit of the data from this patient, the final equilibrium resistance was fixed at 0% for all patients. Due to the dynamic nature of viral evolution after the strong selective pressure of the direct-acting antiviral is removed and the potential stochasticity of PCR amplification, the longitudinal sampling for this patient may not have been sufficiently long to capture the dynamics implied by the model's functional formula. Support for the placement of equilibrium resistance levels below 20% and closer to 0% is provided by a previous analysis by Bartels et al., who found that none of 3447 patients assayed for resistance by population sequencing had naturally occurring telaprevir resistant variants present as polymorphisms [18].
To the best of our knowledge, population sequence data have not been explicitly used in any mechanistic modeling analyses thus far. Previously, clonal sequence data were used to construct a multi-variant HCV dynamic model that explained the dynamics of specific telaprevir-resistant variants before and after telaprevir treatment [20], [21]. This model was originally developed by fitting viral kinetics from patients treated with telaprevir monotherapy, with clonal sequence data used to quantify the relative fitness of specific telaprevir-resistant mutants [20]. The model was then refined in order to predict SVR rates by estimating relative fitness rates of different resistant variants using viral kinetics from Phase 2 telaprevir studies, but in this refinement no additional sequence data beyond the Phase 1 clonal sequence data were used to estimate model parameters [21]. Similarly, Rong et al. used an HCV model accounting for drug-resistant and drug-sensitive viruses to explain viral dynamics with telaprevir-based treatment; only clonal sequence data were used to inform the model parameters [22]. These modeling works differ from ours in two primary ways. First, these prior models [20]–[22] used clonal sequence data from a Phase 1 study with small numbers of patients (i.e., tens). In contrast, our work used both clonal and population sequence data from Phase 3 studies with large numbers of patients (i.e., hundreds). Second, these prior models [20]–[22] included substantially more mechanistic detail and greater numbers of free parameters than the approach presented here. However, Ganusov et al. elegantly demonstrated that these more complex models reduce to the simpler model used here under the limiting assumptions of a variant initially present at low frequency (in this case, WT) and a small mutation rate [23]. Thus, both the current analysis and the aforementioned analyses implicitly employ the same basic framework. Namely, viral dynamics are characterized as a competition between viral variants with different fitness levels, and the dynamics of the variant most fit for the environment (e.g., with versus without treatment) outcompeting the other variants are described by exponential growth capped at a specific maximum. The relative simplicity of the approach presented here (only two free parameters per patient) provides a more tractable framework for addressing the quantitative questions surrounding resistance reversion and is less likely to be subject to model over-fitting.
We used Kaplan-Meier estimation to determine the expected time frames for given events (e.g., the time-to-20% reversion, time-to-1% reversion). This data rich Kaplan Meier analysis (Figure 6) used all available data, and included 391 distinct visit results (Figure S1). We found that population sequencing provides a conservative estimate for the time-to-20% reversion (Figure 6, Figure 7, and Table 1). Intuitively, this finding is reasonable: population sequence data discretely sample a continuous process and reversion is marked as occurring at the first single time point it is observed. However, reversion of virus to WT may have occurred prior to this sampled time. Consequently, the estimates based on this approach should become more conservative as the frequency of sampling decreases. The modeling approach presented here offers one means of overcoming these issues by considering the process of reversion along a continuum.
This analysis provides a novel framework for developing a quantitative understanding of resistant variant evolutionary dynamics. The model enabled prediction of the median time-to-1% resistance for HCV subtypes 1a and 1b (11.0 months and 2.1 months, respectively) even though many patients solely had population sequence data available, which can only be used to determine reversion to 20% resistance. These model predictions (median time-to-1% reversion) were comparable to the median time-to-20% reversion as determined by a previous analysis using only population sequence data [14]. Significantly, these predictions suggest that if a patient does not achieve an SVR with telaprevir-based treatment, the viral population is likely to contain less than 1% resistant virus within a year following treatment failure. These findings provide additional quantitative information for patients with HCV infections and health care providers concerned about the ramifications of not achieving SVR with current treatment options.
The dataset used for this analysis was previously reported by Sullivan et al. [14] and is briefly described here. Samples were obtained from 388 patients who had been enrolled in the Phase 3 telaprevir studies (ADVANCE [7], REALIZE [8], and ILLUMINATE [24]) and did not achieve an SVR. The Phase 3 studies evaluated either 8 or 12 week telaprevir treatment durations with 24 or 48 week durations of peg-IFN and RBV. Follow-up assessments monitored the retention of resistant variants for those patients who did not achieve an SVR.
Population sequence analysis was performed with a minimum of 11× coverage of the NS3 protease, and a median of 4 time points per patient as described by Sullivan et al. [14]. In this analysis, a mutation for a given variant is coded as ‘not present’, ‘present as a polymorphism’, or ‘present and monomorphic,’ which quantitatively correspond to frequencies of <20%, between 20 and 80%, and ≥80%, respectively. As an alternative to population sequence analysis, clonal sequence analysis can be used to quantitate each variant with a theoretical 95% confidence limit of detection of ∼5% when 96 clones are sampled. For this analysis, clonal sequence data were obtained for a subset (n = 51) of samples for which population sequence data were already available. Clonal sequencing utilized the 9 KB amplicons used for the direct population sequence analysis. These amplicons were cloned into a TOPO PCR-XL® vector (Invitrogen) and transformed into electrocompetent E. coli as previously described [25]. Plates containing transformed E. coli were sent to Beckman Coulter Genomics (Danvers, MA) or Genewiz (Cambridge, MA) where 96 clones were selected for sequencing of the NS3 protease with a minimum of 3× coverage. If less than 50 clones contained inserts in a given sample, the process was repeated for that sample. A median of 2 time points and a median of 82.5 clones were available per patient with an interquartile range (IQR) of 70–91. From 32 genotype 1a patients, samples from 80 total visits were obtained with a median (IQR) of 79.5 (69.75–91.25) clones analyzed per visit. From 19 genotype 1b patients, samples from 54 total visits were obtained with a median of 84.0 (71.25–90.75) clones analyzed per visit.
For patients that did not achieve an SVR, resistance evolutionary dynamics after treatment were modeled as a competition between the WT virus and telaprevir-resistant variants (an example from a hypothetical patient is shown in Figure 1). Under the assumption that in the absence of telaprevir, the WT virus is more fit than telaprevir-resistant variants [10], [14], [19], the target-cell limited viral dynamics were approximated using a logistic model for the WT virus, V:(1)Here k is the rate at which WT virus out-competes telaprevir-resistant variants. We assumed that V is the percent of the population that is WT virus; therefore Vmax is 100%. Solving Equation 1 for V then yields:(2)Here V0 is the amount of WT virus present at the beginning of resistance monitoring after the patient did not achieve an SVR and t is the relative time from the start of resistance monitoring. Equation 2 has been derived by Ganusov et al. [23] from more complex viral dynamic models under the limiting assumptions of (1) a variant initially present at low frequency (in this case, WT) and (2) a small mutation rate.
The percent resistance (i.e., quantifying the fraction of resistant variant) is simply:(3)The dynamics in each individual patient were characterized via Equation 3 by specifying the parameters V0 and k.
Qualitative and quantitative information on percent resistance (R) were obtained from population and clonal sequence data, respectively. Because clonal sequence data are quantitative, squared deviations between the model predictions and the percent resistance determined by clonal sequencing for a given sample were penalized. Statistically, this approach is equivalent to assuming that the clonal sequence errors are independently and identically distributed. Because population sequence data are qualitative, only model predictions that fell outside of the expected population sequence binned quantitative range (0–20%, 20–80%, and 80–100% resistance) were penalized. In this case, squared deviations were again penalized, with the deviation defined as the difference between the model prediction and the population sequence range extremum closest to the prediction. For example, if the expected range were 20–80% resistance, then a model prediction of 85% resistance at a given time point would have a deviation of 85%–80% = 5%. Similarly, if the expected range from a direct population sequence result were 20–80% resistance, then a model prediction of 10% resistance would have a deviation of 10%–20% = −10%.
The log10 values of the model parameters V0 and k were determined by minimizing the sum of squared errors over all population and clonal sequence data points for a given patient. Computationally, the population sequence data points were handled efficiently as soft constraints, a technique used in advanced process control (see [26] and the references contained within). Assuming there are np population sequence and nc clonal sequence data points, the optimization problem is then:(4)Here:
For example, considering an expected range of 20–80% resistance with a model prediction of 85% resistance, the constraints in Equation 4 dictate that and , and since we minimize over ,. Similarly, the constraints in Equation 4 also dictate that and , and since we minimize over ,.
Data from patients that had both population and clonal sequence data were first fit using Equation 4. The parameters V0 and k for the entire population were approximated using a log-normal distribution; this distribution was selected to enforce positivity of the parameters. The mean () and covariance (Π) for this distribution (θ, consisting of both log10V0 and log10k) were calculated from the subset of patients with both population and clonal sequence data. Finally, data for each individual patient in the population were fit using Equation 4 augmented with a term penalizing deviations of θ from its prior:(5)s.t.: same constraints as Equation 4
Some exceptions were made to this strategy:
Model predictions were compared to results of 52 samples generated by deep sequencing (DS) [19]. The limit of detection for this form of massively parallel sequencing was ∼1%. This dataset was compared against model results generated in this analysis if the population sequence result were ≤80% resistance and clonal sequence results were not available at a given time point (n = 37). Assessment of the model was performed using two methods: (1) Monte Carlo simulation and (2) an assessment of the null hypothesis that the model-predicted results were not different from the observed (actual) results. In both cases, a quantitative result at some time t after treatment failure was obtained from the DS dataset. The modeled patient-specific parameters V0, k, and t were used to solve Equation 3 for R. Given a limit of sensitivity for the DS assay of 1%, if the observed (actual) result were below the assay LOD, the actual % resistance was imputed as 0%.
For the Monte Carol simulation, a sum of squared errors was calculated from the difference between the actual and model predicted % resistance. Monte Carlo simulations assumed a uniform distribution within the measured population sequence range. 10,000 replicate datasets were sampled to generate a distribution of errors. The significance of the model prediction was determined by calculating the percentage of errors in this distribution that were less than or equal to the observed error.
To assess the null hypothesis that the modeled results were not different from the actual results, the difference between the actual and predicted result was determined for each sample, and the null hypothesis that the mean of this difference equaled 0 was tested. Because the resultant differences were not normally distributed, the absolute differences were first arcsine square root transformed, but the original sign of the difference was retained. Parametric (t-test) and non-parametric (rank-sign) tests were performed on the resultant dataset with α set to 0.05 for each test.
Once parameters for all patient samples were estimated, the modeled time required for virus from samples of each patient to reach a specific percent resistance was calculated by setting the % resistance (R) in Equation 3 and solving for t (specific time-to-x% denoted by τx%). Kaplan-Meier curves were constructed to determine statistics for the population τx%. Patients who are modeled to retain resistance throughout their post-treatment follow up were considered right censored in the Kaplan-Meier analysis, with the time of the last observation in the population sequence analysis dataset relative to the time after treatment failure used as the time of the censoring event. To determine the effect of covariates on time-to-1% reversion, Cox proportional hazards models were applied across relevant covariates on the Kaplan-Meier analysis, with this analysis performed separately for genotypes 1a and 1b and with a null hypothesis of no effect.
Population and clonal sequence data were queried using a custom Oracle database with Perl scripts. R (v. 2.15.0) was used to convert these numerical values into a format readable by Octave, which was in turn used to estimate model parameters using the optimization routine sqp.m. R was used to generate figures and calculate Kaplan-Meier statistics using the survival library. Kaplan-Meier statistics were confirmed by independent generation with JMP statistical software (SAS Institute, v. 8.0.1).
To ensure patient confidentiality, an anonymized dataset containing a summatry of the raw data underlying these analyses has been created and is available upon request to [email protected].
|
10.1371/journal.pntd.0007258 | The influence of raw milk exposures on Rift Valley fever virus transmission | Rift Valley fever virus (RVFV) is a zoonotic phlebovirus that can be transmitted to humans or livestock by mosquitoes or through direct contact with contaminated bodily fluids and tissues. Exposure to bodily fluids and tissues varies by types of behaviors engaged for occupational tasks, homestead responsibilities, or use in dietary or therapeutic capacities. While previous studies have included milk exposures in their analyses, their primary focus on livestock exposures has been on animal handling, breeding, and slaughter. We analyzed data from multiple field surveys in Kenya with the aim of associating RVFV infection to raw milk exposures from common animal species. Of those with evidence of prior RVFV infection by serology (n = 267), 77.2% engaged in milking livestock compared to 32.0% for 3,956 co-local seronegative individuals (p < 0.001), and 86.5% of seropositive individuals consumed raw milk compared to 33.4% seronegative individuals (p < 0.001). Individuals who milked and also consumed raw milk had greater odds of RVFV exposure than individuals whose only contact to raw milk was through milking. Increased risks were associated with exposure to milk sourced from cows (p < 0.001), sheep (p < 0.001), and goats (p < 0.001), but not camels (p = 0.98 for consuming, p = 0.21 for milking). Our data suggest that exposure to raw milk may contribute to a significant number of cases of RVFV, especially during outbreaks and in endemic areas, and that some animal species may be associated with a higher risk for RVFV exposure. Livestock trade is regulated to limit RVFV spread from endemic areas, yet further interventions designed to fully understand the risk of RVFV exposure from raw milk are imperative.
| Part of the transmission cycle for Rift Valley fever virus (RVFV) is related to direct human interaction with animals as part of everyday activities, including consumption of animal products for nutritional or therapeutic benefits. Although the vector-borne transmission of RVFV by mosquito populations is well understood, less is known about how human contact with animal tissues and fluids yields direct (non-vector-borne) RVFV transmission. This study describes the risks of RVFV transmission contributed by exposure to raw milk. It analyzed humans’ milk-related activities and their cumulative risk of RVFV infection, as determined by community-based behavioral and serological surveys in four villages in Kenya. Our data suggest that likelihood of exposure is increased both by actively milking live animals and by direct consumption of raw milk. The risk of RVFV exposure varied among the species of animals kept as livestock and utilized for milk production. Further investigations are necessary to fully characterize the dynamics of RVFV in milk. A better understanding of the role of milk in RVFV transmission will contribute to the public health management of RVFV outbreaks and interepidemic infections.
| Rift Valley fever virus (RVFV) is an RNA virus of the Phenuiviridae family [1, 2] that causes a wide range of disease symptoms in both humans and animals [3–6]. Originally isolated in 1930 in the Rift Valley of Kenya [7, 8], RVFV remained within the continent of African until 2000 when it emerged in Saudi Arabia and Yemen [8–11]. RVFV is endemic in much of sub-Saharan Africa [3, 12–19] and the Arabian Peninsula [3, 20–22], but imported cases, fueled by travelers from Europe [23, 24] and more recently, China [25, 26], continue to raise concerns of a future major emergence of RVFV to currently unaffected areas. Many studies theorize that the long-term isolation of this virus combined with changing climate patterns and the ubiquitous availability of vector species suggests it has built substantial momentum for emergence in naïve populations and regions [27–31], as seen recently with other mosquito-borne viruses including West Nile virus (WNV) [32, 33], chikungunya virus (CHIKV) [34–36], and Zika virus (ZIKV) [34, 37, 38].
Infection with RVFV can cause high fever (40°C– 42°C), nasal discharge, vomiting, and injected conjunctivae in animals [24, 39, 40]. Similar symptoms of nonspecific febrile illness are common in humans, with risk of severe sequelae, including retinitis and reduction or complete loss of vision, acute hepatitis, renal failure, hemorrhagic disease, encephalitis, and neurological complications [11, 24, 41–44]. While the spectrum of animal species that are susceptible to RVFV infection is quite extensive, sheep are considerably susceptible, followed by other common domesticated livestock species such as cattle and goats [45, 46]. Newborn and younger animals are at a higher risk for severe disease and death, generally within 5 days, as a result of RVFV infection [46–48]. Humans may experience asymptomatic infection, which is rarely diagnosed or reported, and may contribute to the spread of infection. Causative factors influencing variability of disease presentation in humans have yet to be determined.
Fatality rate also increases in naïve animal populations, which can lead to a sudden devastation of herds. Multiple generations of animals can be lost during outbreaks as pregnant animals often experience spontaneous abortion, making post-infection herd recovery difficult. Rapid reduction in herd size can also have significant financial and resource burdens for families and villages that rely on income from the sale of animal meats, milks, and byproducts. Currently, import and export of meat and livestock is governed by the World Organization for Animal Health (OIE) Terrestrial Animal Health Code [49], which restricts trade after detection of clinical signs or laboratory-confirmed case of RVFV. Under these regulations, formal importation of milk and milk products requires presentation of an international veterinary certificate attesting that the products have been pasteurized and hygienic practices and control measures were met for each product imported from countries not free from RVFV [49]. Such guidelines do not apply to individual behaviors or intercountry trade and sales, and therefore do not restrict products that may be contributing to the maintenance of inter-epidemic.
Vector-borne transmission of RVFV is widely understood [11, 50–53], and there are many opportunities for direct zoonotic transmission in populations that engage in regular animal handling, breeding and rearing, and slaughtering. Direct contact with high volumes of animals for slaughter, such as one would find in an abattoir or slaughterhouse, has been shown to increase risk for exposure to RVFV when compared to regular behaviors associated with keeping animals at the homestead [14]. Similarly, handling abortus from infected animals carries significant risk for exposure, as the aborted fetus may contain a high titer of live virus [54].
While the mosquito-borne transmission cycle of RVFV is well understood[55–59], many gaps in the knowledge of other mechanisms of RVFV transmission persist, specifically underlying risk factors that may contribute to interepidemic transmission and emergence in new regions. Regular exposure to potentially infected animals contributes to the incidence of human RVFV infections, yet the parameters of human behaviors and animal exposure have not been well defined. Our study aimed to characterize the importance of raw milk and behaviors related to milk exposures as a previously understudied method of zoonotic RVFV transmission.
This secondary analysis describes populations in two specific regions of Kenya: Western and Eastern Kenya (Fig 1). Each participating village provides a unique perspective to risks associated with RVFV exposure, as village practices are unique to their surrounding environment, bordering regions, and available resources.
Participants in the Western region resided within 45 kilometers of Busia, Kenya, with notable borders of Uganda and Lake Victoria, as previously described [14]. Enrollment and sample collection occurred as a part of a zoonoses study from 2010 to 2012. The Busia region is primarily rural with considerable representation of Luo, Luhya, and Teso ethnic groups in the area [14]. Keeping livestock is common on the homestead or village level, and many participants confirmed regular contact or behaviors with animals.
Eastern region locations of Kenya included in this analysis had three main clusters: Bodhei, Sangailu, and Masalani. The sampling cluster from Garissa county [60], included a collection of smaller villages within the Sangailu region, including Golabele, Sabenale, Gedilun, Matarba, Korahindi, and Tumtish [44]. Villages within Sangailu are predominantly semi-nomadic pastoralists and herders, many of whom are of Somali ethnicity. Enrollment and data collection within Sangailu occurred between August and November of 2011 [44]. Enrollment and data collection in Bodhei, a predominantly forested area of Lamu County, occurred in 2006 during an interepidemic period for RVFV [60]. Masalani is a semi-arid region located in the Ijara constituency of Garissa County [61, 62]. Populations sampled in Masalani included the rural village of Gumarey and the larger town of Sogan-Godud [61, 62]. Participants enrolled in Masalani were sampled in two phases: during an interepidemic period in early 2006 [61], and a post-epidemic period in late 2009 [62]. Sampling was not conducted during active outbreaks in any of the villages included in this secondary analysis, as the only reported outbreak in Kenya occurred in late 2006 to early 2007 [58, 63, 64].
Ethical approval was obtained for the primary community surveillance studies individually. Studies conducted in Bodhei and Masalani, titled “Late Outcomes of Rift Valley Fever in Kenya: Ijara Clinical Survey” were reviewed and approved by the University Hospitals Institutional Review Board (IRB) for Human Investigation at Case Western Reserve University (Protocol #10-04-09) and the Kenya Medical Research Institute (KEMRI) Ethical Review Committee (SSC Protocol #918). Ethical approval for the study conducted in Sangailu, titled “Innate Immune Factors in Host Susceptibility to Rift Valley Fever Virus” was granted from the University Hospitals Case Medical Center IRB for Human Investigation at Case Western Reserve University (UH IRB #: 11-09-01) and the KEMRI Ethical Review Committee (SSC Protocol #195). Ethical approval for the “People, Animals and their Zoonoses (PAZ) project community and slaughterhouse worker studies conducted in Busia were obtained by the KEMRI Ethical Review Committee (SSC Protocols #1701 and #2086). Written, informed consent was obtained from all participants for each primary study using consent forms that were available in English and Kiswahili. For child and adolescent participants (ages 1–17), a parent or legal guardian provided consent. All approved consent forms presented a section on “future use” of samples and data collected, and all of the participants included in this secondary analysis approved continued usage for future studies.
This study was a pre-specified secondary analysis applied to multiple existing datasets. Participants aged 1 to 87 years were enrolled in previous cross-sectional studies to determine the prevalence of past RVFV exposure using standard serological methods. All participants were administered a questionnaire uniquely designed for the primary goals of each study to detail basic demographic data, health history, and epidemiologic data regarding lifestyle, environmental exposures, and other behaviors.
Serological status was determined by indirect IgG enzyme-linked immunosorbent assay (ELISA) at the time of the original studies, as previously described [14, 15, 44, 61, 62]. All serology was performed according to the same ELISA protocol, using control serum verified by plaque reduction neutralization testing (PRNT).
This secondary study specifically aimed to investigate the statistical importance of raw milk-related exposures for risk of RVFV exposure and transmission. All primary studies were specific to investigating RVFV in Kenya, and utilized questionnaires regarding daily activities and behaviors, occupational behaviors, and specific questions relating to animal exposures and the consumption of animal products, such as meats and milks. For this secondary analysis, data specific to behaviors relating to milk exposure and serology for RVFV from each separate survey study were compiled using Excel. Statistical analyses of the exposures, specifically milking versus consuming or ingesting raw milk from specific animal species were performed using R version 3.3.1 [65].
Exposure methods (milking and consumption) were modeled separately by logistic regression while only adjusting for age and gender. Forest plots were created to illustrate age and gender adjusted odds ratios (OR) of RVFV infection for each exposure with a 95% confidence interval (CI95). Additionally, full models were fit to: (1) examine the potential impact of geographic bias and (2) include an interaction term for milking and consumption to tease out the impact of exposure to one or both exposure methods for Tables 2 and 3, respectively.
Standardized Mean Difference (SMD) was calculated using the “tableone” package in R [65]. SMD was used to visualize the distance between two groups by standardizing variables, specifically those with prior RVFV exposures, or “infection”, and those without prior RVFV exposure, or “no infection”, standardized by animal type.
Demographic data and behavioral data relating to milk were analyzed for potential risk factors for RVFV exposure, as reported in Table 1. Prior exposure to RVFV varied among villages included in this study. Of the four main regions included in this study, prevalence of RVFV infections ranged from 62.9% (Sangailu (n = 168)) to 10.9% (Busia, n = 29), indicating variability across the villages within our study site (p < 0.001, SMD = 1.37). The age of participants included in the total cohort skewed towards young adult (mean = 26.80 years of age, standard deviation (SD) = 19.64), yet the mean age of individuals with history of RVFV infection was 40.70 (SD = 19.41), which is comparatively higher than the mean age of those without prior infection (mean age = 25.04, SD = 19.24) (p < 0.001, SMD = 0.811). The youngest participant to test seropositive for prior RVFV infection was 2 years old, and the oldest was 85 years old. Gender was not found to be a considerable factor in RVFV exposure history, as approximately half of each exposure cohort identified as female (p = 0.072, SMD = 0.118).
Raw milk exposure factors are reported by behavior (milking versus consumption by ingestion), and by specific animal species. Both milking and consumption behaviors were found to be associated with a significant risk of exposure (any milking behavior by the seropositive group: n = 206 (77.2%), versus that of the seronegative group: n = 1,264 (32.0%) (p < 0.001, SMD = 1.024); any raw milk consumption by the seropositive group: n = 231 (86.5%), versus that of the seronegative group: n = 1,320 (33.4%) (p < 0.001, SMD = 1.273)). In order to identify the varied risks associated with exposures to animal types or milk derived from specific animal species, each behavior was also analyzed by specific animal species. Regardless of village or exposure behavior, cows (n = 1,219 (28.9%)), sheep and goats (n = 1,000 (23.7%)) remained the most commonly raised livestock for milk production. Individuals with a history of RVFV infection (n = 267) reported milking cows (n = 193 (72.3%), p <0.001) and sheep or goats (n = 201 (75.3%), p < 0.001) with similar prevalence. Comparatively, individuals without prior RVFV infection (n = 3,956) reported less milking of cows (n = 1,026 (25.9%)), and sheep or goats (n = 799 (20.2%)), overall. Additionally, consumption of cow’s milk (n = 224 (83.9%), p < 0.001), sheep’s milk (n = 227 (85.0%), p < 0.001), or goat’s milk (n = 228 (85.4%), p < 0.001), was reported with similar regularity by individuals with prior RVFV infection (n = 267). Individuals without prior RVFV infection (n = 4,020) reported consumption of raw milk for approximately one-third of respondents, with a similar reporting frequency for cows (n = 1,267 (32.0%)), sheep (n = 1,275 (32.2%)), and goats (n = 1,274 (32.2%)). Very few respondents reported exposure to raw camel milk through milking (n = 9 (0.2%)) or by consumption (n = 19 (0.4%)), compared to other types of animals reported (Table 1). Milking and consumption odds ratios adjusted for age and gender are presented in S1 Table.
Risk of infection, measured by serological data indicating prior infection, was found to be significant regardless of behavior (Fig 2). No significant difference in exposure risk was found between specific milking behaviors with cows, sheep, or goats (Fig 2A). Exposure to camels by milking was observed to have non-significantly lower odds for RVFV exposure (p = 0.71, OR = 0.66, CI95 0.03–4.34) than that of other animal species (milking cows, p <0.001, OR = 5.92, CI95 4.39–8.11); milking sheep or goats, p <0.001, OR = 9.69, CI95 7.02–13.61) (Fig 2A).
Similarly, risk of exposure by consumption of raw milk was comparable between cow’s milk, sheep’s milk, or goat’s milk (Fig 2B). Consumption of raw camel milk was not found to be a significant transmission risk for RVFV exposure (p = 0.24, OR = 2.19, CI95 0.48–7.20) than those of other animal species (cow’s milk: p <0.001, OR = 17.35, CI95 12.20–25.30; sheep’s milk: p <0.001, OR = 19.68, CI95 13.66–29.11; goat’s milk: p <0.001, OR = 20.52, CI95 14.20–30.48) (Fig 2B). Age- and gender- adjusted odds of RVFV infection by method of exposure and milk type are further described in S1 Table.
In an attempt to identify regionally-specific behaviors, populations included in this analysis were grouped by eastern and western geographical regions within Kenya. Odds ratios were then adjusted by region, in addition to gender and age (Table 2). Variables such as age and gender were included for insight into demographic differences, as well as possible risk factors dependent on geographical region.
In all models, both age and region were distinguished as variables that impacted exposure, regardless of behavior (milking versus consumption or ingestion), or animal type (Table 2). These data indicated a lack of difference in gender roles for milk-related behaviors between western and eastern regions of Kenya. Similarly, child participants (aged between 1 and 15 years of age) were not associated with a higher likelihood of exposure when compared to adult participants (aged 16 to 87 years of age), regardless of behavior.
In order to distinguish the importance of each distinct milk-exposure behavior for RVFV transmission in individuals who engage in both milking and consuming raw milk, each exposure was analyzed for effect relative to the other behavior (Table 3). The top section of Table 3 displays the relative effect of performing milking duties, or lack of milking duties, and how such milking duties affect two behavior groups (consumers and non-consumers), individually. Resulting OR and p-values reported for each row are the odds of being seropositive for RVFV for those that perform milking duties compared to those who do not perform milking duties within each group (consumers or non-consumers). The bottom section of Table 3 displays the relative effect of raw milk consumption for two behavior groups (those who perform milking duties versus those who do not perform milking duties). Milking duties appeared to influence risk in individuals who also consume raw milk (OR 2.3, CI95 1.48–3.59, p < 0.001); however, milking duties do not appear to influence risk for those who did not report consumption of raw milk (OR 0.51, CI95 0.17–1.57, p = 0.24), whereas consumption behaviors did not increase risk of exposure whether the individuals also had milking duties (OR 2.1, CI95 0.59–7.43, p = 0.25) or not (OR 0.47, CI95 0.17–1.27, p = 0.13). Each of these effects was observed in general, animal-nonspecific milking and consuming behaviors, as well as animal-specific exposures in cattle, sheep, and goats, with the exception of camels (Table 3).
Our data illustrate the risk of RVFV transmission associated with raw milk consumption and milking behaviors and reveals the act of milking as a likely significant contributor to viral transmission. The act of milking animals is a culturally, nutritionally, and financially important practice that is performed around the world. Identification of raw milk products and milking behaviors as a potential pathway of RVFV transmission may be critical for ongoing efforts, such as predictive modeling[31, 66], cell and molecular research[6], or clinical and public health interventions[67], to mitigate outbreaks and prevent RVFV emergence into new areas of the world.
Public knowledge of RVFV and its diverse transmission processes is relatively limited, and public health efforts are likely to fall short in encouraging thoroughly safe practices. During outbreaks in eastern Africa in 2018, efforts to limit the spread of RVFV were mainly focused on abattoirs and consumption of meat products. Residents in many affected villages in western Kenya were advised to “only eat inspected livestock products, including milk and meat” [68]. While milk and meat are included in this statement, animal handling activities such as milking were not mentioned. Additionally, potential methods for risk mitigation, such as boiling milk products for sterilization, or mosquito abatement outside of standard mosquito net usage, were not mentioned [68].
Milk is routinely considered as a potential route for zoonotic transmission of RVFV, yet the distinction between risks from milk consumption versus risks from direct animal contact through milking has yet to be defined. Milk is acknowledged as a risk factor for RVFV transmission in many publications [54, 61, 62, 67, 69–75], yet the contributive weight or regularity of milk as a route of transmission [76, 77], and the possibility of milk containing live virus [76, 78] is regularly debated. Consumption of raw or unpasteurized milk is often mentioned as a possible but unusual method of exposure [71, 77, 79]. There is a minimal amount of experimental evidence of live virus actively being shed into the milk of lactating animals [80, 81], but these experiments were limited, and further experimental data from more recent experiments have yet to be published [3]. In this study, we found exposure to raw milk to be correlated with prior RVFV infection. All consumption behaviors were found to be significantly associated with seropositivity after adjustments for age and gender (S1 Table).
Our data describe milking behaviors, regardless of species of animal, as a significant risk factor for exposure to RVFV. Few studies report exposure to raw milk as a potential route for RVFV transmission with a distinction between milking and ingestion or consumption [54, 67, 69, 72]. A study in Madagascar by Olive in 2016 describes the variable of raw milk as “contact with raw milk” in their statistical correlation to serological data [69], which does little to define the true route or behavior influencing increased exposure risk. Occupation is often analyzed as a representation of animal exposure, yet keeping domesticated livestock at the homestead or as a village is common practice throughout sub-Saharan Africa. Occupations with specific types of animal handling, such as slaughterman in an abattoir, have been shown to pose an increased risk of exposure to RVFV through handling of a higher volume of live animals, animal carcasses, and bodily fluids during slaughtering [14]. Additionally, methods associated with slaughtering and carcass cleaning may cause the virus to become aerosolized, leading to an alternative inhalational route of transmission. Of note, exposure via aerosolization has been associated with severe neurological sequelae in mouse models of RVF [82].
Attitudes around and behaviors with milk and types of animals kept is likely to vary among villages, as Kenya is comprised of a vast landscape of ecosystems and human populations with diverse cultures and beliefs. Food and beverages often have very strong cultural and/or ritualistic significance, transcending basic usage for sustenance. As with many foodborne illnesses, processing meats and animal byproducts, whether by thoroughly cooking, preserving, or other sterilization methods, reduces the likelihood of transmission of RVFV. For example, processing of meat products results in a change in pH in the tissues and fluids which effectively inactivates live virus [46, 83]. Simulated experiments have shown select viruses can be propagated in milk and milk products, such as heavy cream and ice creams [84]. Many cultures use foods directly as or for the administration of therapeutics. In a study by Mutua et al., consumption of a variety of animal products was common, yet meat and milk specifically were used in the administration of medicines [67], which adds to the complexity of behaviors and potential routes for RVFV transmission. Consideration of the reason for specific animal fluids or tissues ingestion, and whether the products have been processed, cooked, or are being ingested raw should be included in future foodborne RVFV transmission research.
When analyzed by species, cattle and sheep or goats, referred to as “shoats” in this analysis, were found to be a significantly associated with prior infection, as indicated by seropositivity. Behaviors relating to exposure to camel milk were not found to be significantly associated with seropositivity in our analysis. These findings may allude to chemical or molecular differences in milks produced by different animal species that could potentially influence the viability of RVFV in solution. Further research is necessary to properly characterize potential differences between milks from different species. It is also important to note that milk may be a risk factor, not only for humans, but also for nursing animals. Perhaps the high susceptibility of young ruminants for severe and deadly RVF is due to double RVFV exposure in both vector and mother’s milk. While horizontal transmission in animals and humans has yet to be proven, more research is needed to investigate likelihood of non-vector-borne transmission in animals, such as the possible competence of milk to allow horizontal transmission in ruminant species.
It is unclear whether men or women are at higher risk for exposure. Our data did not uncover a direct correlation between gender and milking or milk consumption behaviors, yet others have reported behaviors significantly linked to gender. Mutua et al. describe that while milking is primarily a responsibility of women, women were less likely to engage in milking with sick animals. Conversely, men were more likely to consume milk from sick animals. This suggests that the differences reported in many studies regarding one gender as a risk factor over the other may be referring to gender roles and gender differences by behavior dictated by differing cultures.
This study has number of limitations that should be considered in future studies. The questionnaire responses analyzed were comprised of self-reported data regarding personal behaviors of the participants and may be subject to recall bias. This study only includes data from two geographically opposite regions of Kenya, and other endemic or potentially-exposed regions may experience differential transmission based on the animal species that are most commonly kept for milking or from which animal milk is sourced for consumption. Access to species-specific milks or products may vary in other regions that are endemic for RVFV, or at risk for future RVFV emergence. In addition to species-specific milk access, pasteurized or boiled milk was not discussed in this study, and questionnaires utilized in each study only asked for behaviors around raw milk exposure. Source of milk consumed was not collected in these questionnaires, and exposure risks through milking for one’s own consumption in their homestead or village may be different than that of people who buy raw milk for consumption. Additional considerations will be dedicated to the trade and sale of raw milk in relation to transmission of RVFV. Lastly, only a small number of participants from our study reported exposures to camels with relation to milk behaviors, which may not accurately represent exposure to camel milk throughout Kenya, or in other RVFV endemic regions.
There are many public health strategies for outbreak mitigation that may be improved with the inclusion of raw milk products and milking behaviors as potential risk factors for RVFV transmission. Epidemiological investigations relating to the origin and spread of an outbreak should include questions about milk consumption and milking behaviors. Survey questions relating to milk consumption should relate to specifically raw milk, as well as personal sterilization or pasteurization efforts, such as boiling or fermentation, to ensure proper techniques are being used. Bulletins distributed during outbreaks should include raw milk as a potential risk, and should encourage individuals to avoid milking animals that are showing symptoms of infection or that have been exposed to other animals with suspected or confirmed infection[64, 67]. Current bulletin efforts readily list the consumption of blood, tissues, and raw meat as a route of exposure, yet could improve awareness around other animal-related exposures[11, 64]. As further research is conducted, public education and community awareness efforts administered by local clinics and ministries of health should be updated to include a more extensive list of risks[67]. Specific attention should continue to be directed towards livestock trade as potential routes for viral spread into new communities and territories[25, 26, 76, 85–87]. It may be worthwhile to instate policies and guidelines for economies relating to the sale and shipment of raw milk products at the local level, as seen with current guidelines relating to international and inter-country livestock trade and animal product sales[49, 88].
This study illustrates the potentially significant influence of exposure to raw milk on RVFV exposure, and the importance of direct animal contact in non-vector-borne transmission cycles. Milk exposures are often involved in everyday behaviors, such as occupation, cultural or ritualistic practices, and therefore increase the complexity of transmission risks and the level of detail required for implementing effective risk mitigation. A heightened understanding of the differential risks associated with meats and milk products from various species is necessary to fully describe the transmission cycle for RVFV. The inclusion of raw milk exposure and milking tasks as risk factors for transmission may explain the high level of variability in incidence and prevalence among villages. Further efforts should be dedicated to viral isolation studies for characterization of the role of milk in interepidemic transmission and maintenance of the RVFV. Milk risks may also impact the likelihood of emergence into naïve populations, with risk of importation via trade without regulation of milk products.
|
10.1371/journal.pgen.1003504 | The Majority of Primate-Specific Regulatory Sequences Are Derived from Transposable Elements | Although emerging evidence suggests that transposable elements (TEs) have contributed novel regulatory elements to the human genome, their global impact on transcriptional networks remains largely uncharacterized. Here we show that TEs have contributed to the human genome nearly half of its active elements. Using DNase I hypersensitivity data sets from ENCODE in normal, embryonic, and cancer cells, we found that 44% of open chromatin regions were in TEs and that this proportion reached 63% for primate-specific regions. We also showed that distinct subfamilies of endogenous retroviruses (ERVs) contributed significantly more accessible regions than expected by chance, with up to 80% of their instances in open chromatin. Based on these results, we further characterized 2,150 TE subfamily–transcription factor pairs that were bound in vivo or enriched for specific binding motifs, and observed that TEs contributing to open chromatin had higher levels of sequence conservation. We also showed that thousands of ERV–derived sequences were activated in a cell type–specific manner, especially in embryonic and cancer cells, and we demonstrated that this activity was associated with cell type–specific expression of neighboring genes. Taken together, these results demonstrate that TEs, and in particular ERVs, have contributed hundreds of thousands of novel regulatory elements to the primate lineage and reshaped the human transcriptional landscape.
| Nearly half of the human genome is composed of repetitive sequences, most of which were derived from transposable elements that have replicated in the genome during the evolution of our species. There is growing evidence showing that some of these transposon-derived sequences have been a source of new binding sites for various mammalian transcription factors. Considering that previous studies were targeting only few transcription factors and cell types, a key question that remains is to what extent the transposable elements have contributed to human transcriptional networks. To systematically survey this contribution, we used datasets generated by the international Encyclopedia of DNA Elements (ENCODE) consortium, identifying the location of active regulatory elements in more than 40 distinct human cell types. Using this resource we measured the contribution of all classes of repetitive sequences and systematically characterized the impact that transposable elements have had on the human chromatin landscape. Our results demonstrate that transposon-derived sequences have contributed hundreds of thousands of novel regulatory elements to the primate lineage and reshaped the human transcriptional landscape.
| Nearly half of human DNA is derived from sequences known as TEs that have successfully replicated in the genome during the evolution of our species [1]. Although the parasitic behavior of TEs was initially put forward as a sufficient explanation for their maintenance within genomes [2], [3] there is growing evidence to support the alternative view that TEs have facilitated genomic innovations [4], [5] and contributed critical regulatory elements to their host [6]. Indeed, a number of studies have shown recently that TEs have been the source of binding sites for various mammalian transcription factors (TFs) [7]–[9] and that they have rewired different developmental regulatory networks [10]–[12]. However, given that previous studies were limited either by the number of TFs they surveyed [7]–[10], [13], [14] or by the cell types they explored [10], [11], [15], a key question that remains is to what extent have TEs globally contributed to human transcriptional networks in undifferentiated and differentiated cells. The importance of characterizing the functional role of TEs and other repetitive regions in the human genome is accentuated by the facts that these sequences constitute most of the sequence diversity between mammalian species [16] and are a significant source of human polymorphisms [17] and of somatic mutations in healthy and disease tissues [18]–[20].
To systematically survey the contribution of TEs to human regulatory networks across a range of cell types, we made use of DNase I hypersensitive sites (DHS) data generated at the University of Washington (UW) and Duke as part of ENCODE [21], [22]. The benefit of using these chromatin accessibility maps is that they highlight active DNA sequences [23], [24] independently of a selected set of TFs. Although accessibility does not equate regulatory function, we build upon these data sets to measure the global activity profile of all classes of transposon-derived sequences and systematically characterize the impact that ancient and recent TEs expansions have had on the human chromatin landscape.
Starting from 106 DHS data sets we performed extensive quality control and retained 75 data sets defining a total of 11,848,530 regions of open chromatin in 41 distinct human cell types derived from normal, embryonic, and cancer tissues (Table 1 and Table S1, see Materials and Methods). These DHS data were further grouped across cell types into 1,643,643 distinct regions of open chromatin. By measuring the overlap with repeat elements, we found that 725,610 (44.1%) DHS regions overlapped instances of the 4 major classes of TEs (ERV, also known as LTR, DNA, LINE and SINE). Notably, by partitioning the DHS regions based on the presence or absence of homologous sequences at orthologous loci in other species, we also found that this proportion reached 63.1% for elements embedded in primate-specific sequences (Figure 1A, see Materials and Methods). A large fraction of these primate-specific DHS regions were observed in repeat subfamilies that were themselves specific to the primate lineage as estimated from the divergence of the repeat instances from their consensus (Figure S1, see Materials and Methods).
Given that repeats are ubiquitous in the genome, we wanted to compare the proportion of DHS regions observed in TEs relative to what would be expected by chance. Using annotation-matched random distributions we found that specific repeat subfamilies were significantly over-represented in individual DHS data sets (see Materials and Methods). For example, we observed that 1237 of the 2337 (52.9%) LTR7 repeat instances (a subfamily of the LTR/ERV class) were contributing to open chromatin in the human embryonic stem cell (ESC) line H7 when we would have only expected 60.5 (2.6%). This corresponds to a 20-fold enrichment and is highly significant (p<1.0E-100). We call such repeat subfamilies DHS-associated repeats (DARs) and, using a stringent cutoff (p<1.0E-05), we identified 8937 DARs enriched in various cell types (Table S2). These DARs provided on average 6.7% and up to 11.9% more open chromatin regions than expected by chance in the data sets surveyed (Figure 1B).
We were interested in characterizing further the families of TEs that were contributing to regions of open chromatin. By combining the DAR instances across the various cell types and comparing them to the number of repeat instances of each family across the genome, we found that LINE and SINE repeats were depleted while DNA repeats were observed at levels expected (Figure 1C). In contrast, we found that LTR/ERV, Low complexity, Simple repeats and Others repeat classes were enriched (Figure 1D). For example, although LTR/ERV repeats constitute 13.5% of the repeat instances in the genome, they represent 25.0%, 54.6%, and 33.0% of the DAR instances in normal, embryonic, and cancer cells, respectively. The over-representation of LTR/ERVs in DHS corroborates an observation made previously [22] and did not appear to be a consequence of intrinsic properties of the repeat subfamilies including mappability (Figures S2 and S3, see Materials and Methods). Low complexity, Simple repeats and Others repeat classes were excluded from most downstream analyses because of their extreme GC content (Figure S3) potentially affected by sequencing biases [25].
Next, we looked at the fraction of instances in all repeat subfamilies that were contributing to open chromatin in relation to their estimated age (see Materials and Methods). We observed that for SINE, LINE, and DNA repeats, older subfamilies tended to contribute more often to open chromatin (Figure 1E). Two of the subfamilies contributing the most were AmnSINE1 and MER121, both previously suggested to have acquired functionality in the host [26], [27]. Intriguingly, we observed the reverse pattern for the LTR/ERV repeats with many of the young subfamilies contributing to open chromatin at very high levels (e.g. LTR13 with 379 instances contributing to open chromatin out of 492 (77.0%), LTR2B with 215 out of 332 (64.8%), and LTR7 with 1432 out of 2337 (61.3%)) (Figure 1F). This pattern although dampened was also visible if we restricted the analysis to the data sets derived from normal differentiated tissues (Figure S4).
We noted that DHS overlapping repeats were enriched in chromatin states corresponding to promoters, enhancers, and insulators as defined previously using histone marks profiles [28] (Figure S5). To understand why specific TEs were contributing to open chromatin, we wanted to integrate the DARs with other more targeted functional genomics data sets. For example, it was shown previously that the pluripotency TF OCT4 was bound on LTR9B repeats [10] and it was interesting to see the same repeat subfamily as a DAR in ESCs (Table S2). When we looked in LTR9B for the binding motifs of OCT4 and SOX2, another pluripotency TF, we found them to be specifically over-represented in the repeat instances contributing to open chromatin (p = 6.8E-64 and 3.6E-38 respectively, see Materials and Methods). Notably, the peaks in the aggregate read density profiles of the DHS data in ESCs were also correlating with the localization of the motifs within the repeat instances (Figure 2A).
To characterize more systematically the role of repeat instances in the host genome and to identify putative functional factors associated with the DARs, we used a collection of TF binding sites determined by ENCODE using ChIP-Seq. In the 19 cell types where both DHS and ChIP-Seq data were available, we found that 1014 of the 2784 DARs (36.4%) were statistically enriched for at least one TF (Table S3). This relied on two statistical tests: one that showed that the TF was enriched in the same repeat subfamily and in the same cell type, and one that showed that the number of instances with both DHS and ChIP-Seq signal was also significant (see Materials and Methods). Using this strategy, we found for example that 82.9% of the 210 LTR13 instances that were contributing to open chromatin in K562 were also bound by CTCF (p = 1.1E-13, Figure 2B). Additional DARs supported by specific TFs such as PU.1, BCL11A, and PAX5 in LTR2B are shown in Figure 2C and Figure S6. Predictably, we found that a larger fraction of DARs can be explained by the binding of specific TFs in cell lines where more ChIP-Seq data sets were available (Figure 2D).
To improve on the limited ChIP-Seq coverage in some cell types and in order to characterize the DARs more comprehensively, we developed a classifier to predict TF-repeat associations using Jaspar TF binding motifs (see Materials and Methods). Using this classifier we were able to suggest 3073 high-confidence motif-repeat subfamily associations for 1312 DARs (Table S4). By combining both methods, we were able to predict a total of 2150 unique TF-repeat subfamily associations, which suggest potential functional candidates for 24.1% of the DARs (Figure 2E).
Finally, to further confirm the functional importance of DARs, we also used the annotated conserved non-exonic elements (CNEEs) [29] and assessed the overall sequence conservation of the TEs that were contributing to open chromatin. In total, while only 5.5% of all repeat instances were conserved, we found that 9.0% of the repeats contributing to open chromatin were conserved, a difference that is highly significant (p<1.0E-100, Figure S7). Notably, for almost all repeat subfamilies, we found that the subset of instances contributing to open chromatin was more conserved than expected by chance (Figure 2F).
Next, we were interested in the contribution of repeats to cell type-specific DHS. When we calculated the number of cell types contributing to individual DHS regions, we found that 76.0% of the loci were open in 4 cell types or less (Figure S8, see Materials and Methods). We also observed that regions contributed by few cell types were found more frequently in repetitive sequences (Figure 3A). To determine the cell type-specificity of each repeat subfamily we used the median number of open instances in all DHS data sets as the denominator and calculated the fold enrichment for each repeat subfamily in each cell type (see Materials and Methods). A total of 770 DARs showed a cell type-specific fold enrichment greater than 3 (Figure S8 and Table S2). Notably, we observed that LTR/ERV repeat subfamilies were over-represented in the cell type-specific DARs (Figure 3B) and that on average a higher number of cell type-specific DARs were found in ESCs and cancer data sets (Figure 3C). These patterns were also recapitulated in the top 100 repeat subfamilies with the greatest cell type-specific enrichment (Figure 3D). For example, in the case of the LTR7 repeat subfamily, we observed a remarkable enrichment of 131.6- and 88.7-fold in the ESC lines H7 and H1 respectively. While most cell type-specific DARs were found in ESCs and cancer cell lines, we also found examples, such as the LTR2B and MER121 subfamilies, which had most of their instances in open chromatin from normal differentiated cells (Figure 3E). Additional examples of cell type-specific DARs are shown in Figure S9. We also found that the cell type-specificity of various subfamilies of TEs was supported by the chromatin states previously described [28]. For example, more than 40% of the LTR2B instances were annotated as enhancers in GM12878 while only 10% were annotated as such in H1. In contrast, more than 40% of the LTR7 instances were annotated as enhancers in H1 while only 2.2% of them were annotated as such in GM12878 (Figure S10).
Finally, using a collection of TF binding motifs including novel motifs identified in DNase I footprints [30], we identified tissue-specific motifs enriched in these cell type-specific DARs (Table S5, see Materials and Methods). In particular, we observed that many ESC-specific DARs were supported by ESC-specific motifs that were not enriched in normal- or cancer-specific DARs (Figure S11). The top three ESC-specific motifs found in this way were OCT4, SOX2 and KLF4.
To evaluate the impact of DARs on gene regulation, we used 43 gene expression exon-array data sets from ENCODE and calculated the number of genes in proximity to DAR instances that were up-regulated in the relevant cell type relative to the others (see Materials and Methods). We identified 783 DARs with more proximal up-regulated genes than expected by chance (Table S6). For example, we identified 11 genes in proximity to LTR2B instances that were up-regulated in GM12865 while we would have only expected 4.27 (Figure 4A). Examples of cell type-specific LTR2B associated genes in GM12865 include NAPSB and CLECL1 (Figure 4B and Figures S12, S13), two genes that have been shown to play a role in lymphoblastoid cells [31], [32]. Moreover, we observed that the expression of the DAR-associated genes were frequently highest in the cell type where the DAR had been identified (Figure 4C and Figure S14). We also found that DARs with a higher cell type-specificity score had a higher chance of being associated with cell type-specific expression (Figure 4D). Similar results were obtained using ENCODE RNA-Seq data sets generated by Caltech (Figure S15).
Finally, a recent study combining genotypes with DHS data in 70 lymphoblastoid cell lines has shown that a significant proportion of open chromatin regions, known as dsQTLs, can be influenced by polymorphisms [33]. Having demonstrated that DARs exhibit features associated with regulatory elements, we wanted to test if they also showed this variation across individuals. We found that 36.8% of the reported dsQTLs overlapped repeat instances and that these were contributed by DAR instances in lymphoblastoid cells more than expected by chance (p = 1.1E-6, see Materials and Methods).
In summary, we found that TEs have contributed nearly half of the open chromatin regions of the human genome and the majority of primate-specific elements. This estimate is a lower bound that is likely to grow given that better strategies using longer and paired-end reads will be needed to measure the contribution of young repeat subfamilies and polymorphic sites (Figure S3). An example is the L1PA2 repeat subfamily where, despite the fact that the mappability ratio is 0.08, 117 and 257 of the 4904 L1PA2 instances contributed to the H1 and H7 DHSs respectively. This finding is consistent with previous observations [7], [15], [34], [35] but greatly expands on our understanding of the repeat families contributing to open chromatin in the human genome.
To better understand the regulatory functions that could have been retained in exapted TEs beyond the ones that have already been studied (e.g. [8]–[11], [14]), we predicted a total of 2150 TF-repeat subfamily associations and confirmed that a broad range of functional proteins are targeting these regions (Figure 2 and Tables S3, S4). This resource will be useful to provide insights into the regulation of some of the TE-derived loci that have already been implicated in disease [36]. There is an important distinction between biochemical activity and functional relevance to the host. To help confirm the importance of these regions, we also showed that repeat instances contributing to open chromatin were more conserved than expected by chance (Figure 2F).
Next, we demonstrated that LTR/ERV repeats have contributed a disproportionate fraction of cell type-specific accessible chromatin regions especially in embryonic and cancer cell lines (Figure 3). This is interesting given that network rewiring using ERV elements has already been described in ESCs [10]–[12] and that it has been shown that stem cell potency fluctuates with endogenous retrovirus activity in mouse [37]. The level of activity observed in ERV sequences is likely a consequence of the permissive chromatin state found in ESCs that it sometimes reinstated in cancer [38]. There is fine balance between the successful replication of endogenous retroviruses, from which these repeats are derived, and retrotransposition control in the host [39]. One intriguing possibility is that the manipulations that were initially exerted by the ancestral viruses on their host to by-pass these control mechanisms have also facilitated co-option [40].
Finally, we also reported that repeat subfamilies activated in a cell type-specific manner were also frequently associated with higher expression of neighboring genes. This result corroborates the fact that at the level of expression, TE-derived transcripts, including lincRNAs [41], are also usually tissue-specific [42]. Interestingly, this pattern was observed not only in ESCs but also in differentiated and cancer cells (Figure 4 and Table S6).
Taken together, these results demonstrate that TEs, and in particular endogenous retroviruses, have considerably transformed the transcriptional landscape during primate evolution.
We retrieved 106 ENCODE DHS data sets available from the October 2010 freeze which included replicates for 50 different cell types from a variety of normal differentiated cell types, human ESCs and cancer cell types using the UCSC ENCODE portal (http://genome.ucsc.edu/ENCODE/). These data had been generated from performing DNase I digestion of intact nuclei, isolating DNase I digested fragments and direct sequencing of fragment ends [23], [24]. We discarded 3 data sets involving treatments and performed extensive quality control of the data sets. Specifically, for each peaks file generated by UW corresponds one tagAlign file such that we calculated the average GC content of the tags and removed data sets with GC bias (>55% or <45%). We also used FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/) on these files to remove data sets that failed the per sequence quality criteria (when the most frequently observed mean quality is below 20) or where the total number of overrepresented sequences was above 1 million. Multiple tagAlign files generated by Duke had been combined to call the peaks and we analyzed the individual tagAlign files as described above. The quality control summary is presented in Table S1. In the end, we retained 75 data sets that did not have any unusual deviation based on our quality metrics: 51 produced by UW and 24 by Duke, covering 41 different cell types grouped in 8 “tissues” (Table 1). Most of the UW data sets had a read length of 36 bp while the Duke data sets had a read length of 20 bp. For the downstream analyses, we used the narrow peak files that were generated from the uniquely mapped reads and provided by ENCODE.
The peak regions from the 75 data sets were combined and those within less than 100 bp were grouped in 1,643,643 distinct DHS regions. These DHS regions were resized to 200 bp using their middle point before intersecting with the 5,269,366 repeat instances of RepeatMasker [43] from the UCSC Genome Browser [44] (Figure 1A). In order to calculate the proportion of DHS regions overlapping different classes of repeats based on the age of the sequence in which they are embedded, we sequentially used the liftOver utility from UCSC using default parameters and minMatch set to 0.5. The DHS regions (hg18) were first converted to the mouse genome (mm9), those not converted were then mapped to the marmoset genome (CalJac3), and again those not converted were mapped to the chimp genome (PanTro2) to identify the 23,917 human-specific DHS regions. Primate species divergence time taken from [45]. We also obtained an alignment-free estimate of the age of the repeat subfamilies using the average divergence between the instances and the ancestral repeat consensus (milliDiv value reported by RepBase) and applying the Jukes Cantor method with a substitution rate of 2.2×10-9 per site per year [7]. The ages obtained using this method were largely consistent with previous estimates [16], [46]. Repeat subfamilies with an estimated age <95 Myrs were said to be primate-specific.
For each DHS data set, we calculated the number of overlaps within each repeat subfamily using a 200bp window surrounding the center of the DHS peaks. Next, following a strategy developed on ChIP-Seq data sets [7], [10], we annotated each DHS with respect to its nearest RefSeq genes and we binned the DHS into six categories according to the peak location: TSS (within 1 kbp of a TSS), promoter (up to 5 kbp upstream of TSS), intragenic (within the RefSeq gene boundary), proximal (up to 10 kbp away from the gene boundaries), distal (up to 100 kbp away from the gene boundaries) and desert (more than 100 kbp away from any RefSeq genes). We then generated for each DHS data set a random set of 200,000 regions with the same annotation distribution as the true regions and intersected with the RepeatMasker track to obtain the expected number of overlaps for each repeat subfamily. We used a onesided binomial test to compare the observed number of repeats intersecting the true DHS with the expected numbers from the annotationmatched background. We identified repeat subfamilies with statistically significant contribution to open chromatin (p<1E-5) as DHS-associated repeats (DARs).
We verified that various properties of the repeat subfamilies were comparable across repeat classes. Specifically, we checked for an association between the fraction of repeat instance contributing to open chromatin and the number of instances in a given subfamily, average size and GC content (Figure S2). We only detected an association between open chromatin contribution and GC content affecting Low complexity, Simple repeats and Others repeat classes, which we excluded from most analyses. The majority of the ENCODE processed data sets, such as the narrow peak files that we used in the current study, rely on uniquely mapped reads [47], [48]. To test the impact of such a criteria on the detection of TEs regulatory activity, we extracted 50 million 20 bp and 36 bp sequences (to mimic Duke and UW read lengths respectively) from random location on the human and re-mapped these artificial reads using the Bowtie program [49] and allowing 1 and 2 mismatches respectively. The goal of these simulations was to compute a mappability ratio for each repeat subfamily that is: the ratio between the number of uniquely mapped tags in a particular subfamily and the number of tags that were extracted from this subfamily (Figure S3).
The DHS from the six cell lines (some with two replicates for a total of eight DHS data sets) for which the chromatin states (CS) were available [28] were used for this analysis. The DHS were associated to one of the 15 CS (overlap >50%) using intersectBed from BEDtools [50], then grouped whether they overlap any repeat instance or not. For each data set, the proportion of DHS annotated as each CS was computed, and similar CS were combined together (ex: strong and weak enhancers grouped as enhancers). Similarly, random DHS (using shuffleBed) were generated and overlapped with the CS. Figure S5 is showing the average and standard deviation over the eight data sets. Repeat instances from specific DARs (LTR7, LTR2B and LTR13) were intersected with the CS using a similar strategy (Figure S10).
We used 183 distinct ENCODE ChIP-Seq data sets generated by the Broad Institute, Duke, HudsonAlpha and Yale, covering 87 different TFs from 19 cell lines for which we also used DHS. For each of these data sets we applied the same procedure as to identify the DARs and identified a total of 9367 TF-repeat subfamily pairs. These pairs were then intersected with the DARs and for each combination from the same cell type and the same repeat subfamily we applied a hypergeometric to test the significance of the number of instances with ChIP-Seq peak and DHS. Using a stringent cutoff (p-value<0.001), we identified 2800 statistically significant combinations of DAR-ChIP Seq for 1014 distinct DARs (Table S3).
Using ChIP-Seq data sets obtained previously [51], we trained a classifier that uses the over-representation of TF binding motifs and other features of repeat subfamilies to predict TF-repeat associations. Briefly, five features of repeat subfamilies were used: 1) fraction of repeat instances with motif, 2) fraction of motifs contained within repeat subfamily, 3) motif score ratio between bound and unbound repeat instances, 4) enrichment test for binding motifs within repeat subfamilies, 5) simulations looking at the potential of repeat sequences to generate binding motifs. By combining these individual features using a weighted rank average we were able to achieve an Area-Under-the-Curve (AUC) of 0.81 for this classifier (Jeyakani et al., in preparation). Using 103 JASPAR TF binding motifs derived from human or mouse [52], we applied our classifier to the list of putative motif-repeat subfamily pairs and using a stringent cutoff (top 10%) we identified 2337 potential associations. These motif-repeat subfamily pairs were then intersected with the DARs and for each combination from the same repeat subfamily we applied a hypergeometric to test the significance of the number of instances with motifs and DHS. Using a stringent cutoff (p-value<0.001), we identified 3857 statistically significant combinations of DAR-motif from 1312 distinct DARs (Table S4).
In order to calculate the cell type-specific enrichment for each repeat subfamily, we determined the median number of repeats bound across the DHS data sets (these numbers were further normalized to the total number of sites in each data sets). The median value was computed independently for the UW and the Duke data sets because of the expected differences in mappability given the differences in read lengths. Next, we calculated a cell type-specific fold enrichment for each repeat subfamily in a given cell type by dividing the observed number of repeats contributing to open chromatin in this particular DHS data set by the median number of repeats contributing to open chromatin for this subfamily. This was done for all DARs and non-DARs (Figure S8). Next, we scanned the 56,837 DHS from the 770 cell type-specific DARs for motifs using the FIMO software tool with a maximum p-value threshold of 1×10−5 as was done in [30]. We provided motif templates from Jaspar [52], TransFac [53], Uniprobe [54] and novel de novo motifs identified previously in DNAse I footprints [30]. For each DAR, we then identified the motifs present in >25 repeat instances and in >20% of the instances contributing DHS (Table S5). We also calculated the proportion of DARs from ESC, Cancer and differentiated normal cells with support of at least one of the 28 ESC-specific motif identified in [30] (Figure S11).
From the 70 ENCODE UW Affy All-Exon Arrays expression data sets, only those from the cell types with DHS were selected. These data sets were clustered showing that a few replicates were inconsistent and therefore removed of the downstream analyses, leaving 43 expression data sets (most of them in duplicate). Note that we tried to combine these data sets with the ENCODE Duke Affy All-Exon Arrays data sets but found that the platform correlation was higher than the biological correlation between biological replicates so we therefore decided to only use the UW data (data not shown). Expression data was available in the cell type for 6054 of the 8937 DARs. A gene was called up-regulated in a cell type if it had a Z-score >2 in at least one of the data set compared to the other data sets. For each DAR, the Z-score of cell type-specific expression based on the number of up-regulated genes was computed on permutation tests by randomly picking 10,000 times the same number of genes that associated with the DAR from the set of 35,865 different gene names covered by the arrays. For example, from the 2337 instances of LTR7 in the genome, 788 were contributing to open chromatin in H7 and those were associated to a total of 561 distinct genes. The fact that 85 of these genes were up-regulated in the ESC cell type while only 19.6 (+/−4.3) were expected based the permutation test gives a Z-score of 15.1 (Figure S13 and Table S6).
Similarly, 13 RNA-Seq data sets generated by Caltech from 7 distinct cell types were used to calculate the association of DARs with up-regulation of expression. A 50 kb window centered in the middle of each repeat instance was used and, to estimate the background, the genome was independently segmented in non-overlapping 50 kb windows. For each RNA-Seq data set, the average tag density was calculated in each window. For each window, the mean and SD of the average tag density was then calculated across the 13 RNA-Seq data sets in order to identify up-regulated windows defined as a Z-score >2 in one of the data set of the same cell type compared to the other data sets. For the 2124 DARs for which expression data was available, the Z-score of cell type-specificity expression was computed using permutation tests by randomly picking 10,000 times the same number of windows than the number or repeat instances contributing to open chromatin in this cell type from the background genomic segments (Figure S14).
The 1,034,427 DHS from the 8 lymphoblastoid DHS data sets were first grouped into 430,159 clusters as described above. Using intersectBed, we found as expected that most (4891 of 6070 (80.6%)) short dsQTLs from [33] were overlapping these DHS lymphoblastoid clusters. We also found that 2234 of 6070 (36.8%) short dsQTLs were overlapping a repeat instance. Considering that 995 of the 4891 (20.3%) short dsQTLs overlap one of the 77,135 DHS lymphoblastoid clusters contributed by DAR instances in lymphoblastoid cells (17.9% of all lymphoblastoid clusters), this overlap is highly significant (hypergeometric p = 1.11E-6). Doing the same for the DHS lymphoblastoid clusters that were overlapping DAR instances from the other cell types gave a more marginal enrichment (hypergeometric p = 1.09E-2).
|
10.1371/journal.pcbi.1005742 | Synthesizing developmental trajectories | Dynamical processes in biology are studied using an ever-increasing number of techniques, each of which brings out unique features of the system. One of the current challenges is to develop systematic approaches for fusing heterogeneous datasets into an integrated view of multivariable dynamics. We demonstrate that heterogeneous data fusion can be successfully implemented within a semi-supervised learning framework that exploits the intrinsic geometry of high-dimensional datasets. We illustrate our approach using a dataset from studies of pattern formation in Drosophila. The result is a continuous trajectory that reveals the joint dynamics of gene expression, subcellular protein localization, protein phosphorylation, and tissue morphogenesis. Our approach can be readily adapted to other imaging modalities and forms a starting point for further steps of data analytics and modeling of biological dynamics.
| A wide range of problems in biology require analysis of multivariable dynamics in space and time. As a rule, the multiscale nature and complexity of real systems precludes simultaneous monitoring of all the relevant variables, and multivariable dynamics must be synthesized from partial views provided by different experimental techniques. We present a formal framework for accomplishing this task in the context of imaging studies of pattern formation in developing tissues.
| The need to synthesize data from different observations into coherent multivariable trajectories is discussed in multiple contexts, from physics to social sciences, but systematic approaches for accomplishing this task have yet to be established [1–5]. Here we address this task for imaging studies of developing tissues, where patterns of cell fates are established by complex regulatory networks [6–8]. Advances in live imaging continue to provide new insights into the dynamics of individual components in these networks, but imaging more than three reporters at the same time is still challenging and limited to model genetic organisms [9, 10]. Furthermore, in the absence of reliable live reporters, dynamics of some state variables can only be inferred from fixed tissues. Because of these limitations, extracting the multivariable dynamics from the heterogeneous datasets collected by imaging of live and fixed tissues becomes a non-trivial task [11, 12].
The problem can be illustrated by an imaging dataset from the early Drosophila embryo (Fig 1A and 1B), a model system in which a graded profile of the nuclear localization of transcription factor Dorsal (Dl) establishes the dorsoventral (DV) stripes of gene expression that control cell fates and tissue deformations [13–15]. Current mechanisms of the DV patterning system invoke multiple state variables, such as the levels of gene expression and protein phosphorylation [16] (Fig 1C). These mechanisms were elucidated in studies that reveal only a small subset of the full state space, most commonly 2-3 variables per experiment. Can these partial views be fused into a consistent multivariable trajectory? This is a general question that applies to essentially all developmental systems.
We realized that this question can be addressed by casting the task of data fusion as a matrix completion problem (Fig 1D). Specifically, an image of a fixed embryo or a frame from a live imaging movie can be viewed as a column in a matrix where rows correspond to the relevant variables, such as developmental time or the level of gene expression at a given position. Because of limitations in the number of states that can be accessed simultaneously, the matrix is incomplete. For example, live imaging of gastrulation provides information about nuclear positions as a function of time, but is silent about the levels of gene expression. On the other hand, an image of a fixed embryo reveals the distribution of an active enzyme but has no direct temporal information. Thus, multivariable data fusion requires completing this matrix, filling in the missing components by estimates informed by the rest of the data. Below we show how this task can be accomplished by solving a suitably posed semi-supervised learning problem. We first provide a closed-form solution to this problem and then demonstrate its successful performance on synthetic and experimental datasets.
We assume here that all experiments contain a common variable, which is sufficient to determine all other variables that can be measured or to be predicted. For instance, this variable is revealed by a signal that reports positions of nuclei. This means that the first row in the matrix is complete. To complete other rows, we must establish the mappings between the common variable and each of the target variables. These mappings can be found within a semi-supervised learning framework, in which the values of the variables in the incomplete rows are estimated from a training dataset [17, 18].
As an example, consider images from fixed embryos that are stained to reveal the spatial pattern of an active enzyme, visualized using a phosphospecific antibody (Fig 2). They provide labeled data points that contain information about the common variable and a specific target variable. On the other hand, images without this staining, such as the frames from live imaging of morphogenesis, provide unlabeled data points with only the common variable. By finding a mapping between the common and target variables, we can essentially “color” the frames of a live imaging movie by snapshots of molecular patterns from fixed embryos.
A critical assumption in finding the mappings is that the multivariable dynamics of the patterning process are both low-dimensional and smooth with respect to the underlying parameters. This assumption is supported by studies with mathematical models of specific biological systems and by computational analysis of datasets from imaging studies of development [19, 20]. More formally, we consider a set of data points (x1, …, xl, xl+1, …, xl+u) belonging to a space X. These points correspond to the values of the common variable in the complete row. On the other hand, a row corresponding to any one of the target variables is incomplete. The values in the filled columns of this row are called labels. These are denoted (y1, …, yl) and belong to a target space Y. The semi-supervised learning techniques transfer the information contained in the labeled data points ((x1, y1), …, (xl, yl)) to the unlabeled points (xl+1, …, xl+u), while preserving the intrinsic structure of the dataset [18]. Stated otherwise, these techniques learn the mapping y = f(x) assuming that the considered process is smooth, which means that similar values of x give rise to similar values of f(x).
The missing values, corresponding to the unlabeled data points in each of the incomplete rows, are found by solving the following optimization problem:
f→=argminf→∈Yl+u∀i≤l,fi=yi∑i,j=1l+uwi,j∥fi−fj∥2 (1)
where f → = ( f 1 , . . . , f l + u ) are the values of the target variable on the data points (x1, …, xl+u), the considered norm in Y is the Euclidean distance and the weights wi,j represent the similarity between two data points xi and xj. The norm in the space X can for example be the Euclidean distance in the space where each dimension corresponds to an image pixel or some coordinate in an arbitrary feature transform of that image. Other distances are possible. For example, the one-norm (or L1 distance) can be used, which increases robustness to outliers in the data. That being said, as long as these distances preserve the low-dimensional manifold structure, they will yield similar results as the number of points goes to infinity.
This quadratic optimization problem, known as harmonic extension, has a unique solution that relates the unlabeled data points fl+1, …, fl+u to the labels y1, …, yl where Y = R [17, 21]. The explicit solution reads:
f → u = ( D u - W u u ) - 1 W u l Y (2)
where Y = (y1, …, yl) and f → u = ( f l + 1 , . . . , f l + u ), and d i = ∑ j = 1 l + u w i , j, Du = diag(dl+1, …, dl+u), Wuu = (wi,j)l+1≤i,j≤l+u, and Wul=(wi,j)l+1≤i≤l+u1≤j≤l (S1 Text).
To illustrate our method, we considered a one-dimensional nonlinear trajectory in a three-dimensional space. The trajectory is given by the set of equations
{ x(1)(t)=at(cos(bt)+ϵ(1))x(2)(t)=at(sin(bt)+ϵ(2))y(t)=ctexp(−d(t−e)2)
(3)
where a, b, c, d, e are constants, ϵ(1) and ϵ(2) are Gaussian noise sources and t is a real-valued parameter. The set of points (x(1)(t), x(2)(t)) forms a one-dimensional non-linear manifold embedded in the two dimensional plane and it is parameterized by t. These points are analogs of the embryo morphology. In the absence of noise, this mapping from t to the 2D plane can be inverted as t = 1 | a | ( x ( 1 ) ) 2 ( t ) + ( x ( 2 ) ) 2 ( t ). The signal y(t) is a smooth function of t and is thus a smooth function of (x(1), x(2)) by composition. In this example, y corresponds to the target modality that we would like to estimate.
To mimic the setting of data fusion with three modalities, ((x(1), x(2)), t, y), we consider the following situation: suppose that one acquires a set of labeled points, i.e. a set of l triplets, (((x(1)(t1), x(2)(t1)), y(t1)), …, ((x(1)(tl), x(2)(tl)), y(tl))) and a set of u unlabeled, but timestamped, points, (((x(1)(tl+1), x(2)(tl+1)), tl+1), …, ((x(1)(tl+u), x(2)(tl+u)), tl+u)), as shown in Fig 3A and 3B. The pairwise similarity measures wi,j are computed using Euclidean norm between pairs of data points (x(1)(ti), x(2)(ti)) and (x(1)(tj), x(2)(tj)). In this case, there are no outliers, so the standard Euclidean distance is well suited and there is no need to consider other distance measures.
Then, using Eq (2) it is possible to estimate y = f(x) on the set of unlabeled data points using the harmonic extension algorithm. The results are shown in Fig 3C. We then directly obtain y as a function of t by composition using the known time stamps (tl+1, …, tl+u).
The accuracy of the estimated multivariable dynamics can be assessed using a K-fold validation strategy on the labeled samples (S1 Fig and Materials and methods). For the chosen set of parameters and the size of the dataset, the error is ∼ 1%. As expected for the semi-supervised learning framework, the error decreases with the addition of new unlabeled data points. This example demonstrates how the proposed approach successfully recovers multivariable dynamics from heterogeneous datasets that combine continuous views for part of the state variables and snapshots that report several states without direct temporal information.
As a representative dataset from imaging studies of multivariable dynamics in living systems, we use a collection of ∼ 1000 images each of which reveals the spatial position of the nuclei and either a timestamp or the distribution of one or several components of the DV patterning network (Fig 1D). To apply the semi-supervised learning approach to data fusion to this dataset we need to compute pairwise similarities between the images using the common channel. Prior to this, we took several preprocessing steps that aim to minimize image variability associated to sample handling, microscope calibration and imaging. First, the images were registered to align their ventral-most points. The images were then resized and cropped such that the embryos occupy 80% of the image. All images were resized to 100 by 100 pixels. To overcome local variations of image intensity, we computed a local average using a Gaussian kernel, and then renormalized the image by that value. We also applied a logistic function to the images to handle contrast variability, S2 Fig.
Most importantly, to ensure that pairwise differences between images are insensitive to small translations or deformations, we applied the scattering transform [22] and compared the resulting transform vectors. The scattering transform of an image is a signal representation obtained by alternating wavelet decompositions and pointwise modulus operators. We found that second-order scattering coefficients with an averaging scale of 64 pixels provided sufficient invariance. These are computed using the ScatNet toolbox [23, 24]. The result is a vector of dimension 784 for each image. The point clouds corresponding to each of the 11 datasets were centered separately. It has been shown that the Euclidean distance on the scattering transform is locally invariant to translation and stable to deformation of the original image [22]. For this reason, we compare these 784-dimensional vectors using the Euclidean norm. The corresponding low-dimensional manifold on which the data points lie is shown on S4 Fig.
For each of the 512x512 pixels of each live movie frames, there is a common channel reporting the nuclei spatial position and there are 5 channels that we would like to complete. These channels contain the information about the spatial distributions of one enzyme (dpERK), two transcription factors (Twist and Dorsal), and transcripts of two genes (ind and rho). We thus solved the data fusion problem for each pixel and each channel, leading to 5x512x512 semi-supervised learning solutions. The combination of labeled and unlabeled datasets is described on S2 Table. The result is a multivariable trajectory for the joint dynamics of tissue shape and five molecular components within the regulatory network that patterns the DV axis of the embryo (Fig 4). To evaluate the accuracy of the method, we computed the cross-validation error for each pixel and averaged over the entire images. We found that the normalized absolute error is of 0.9–2.5% of the signal range when considering the various modalities of the entire experimental datasets (S3 Table). We show how the algorithm performs on several examples in Fig 5.
We presented a formal approach to synthesizing developmental trajectories. By posing the task of data fusion as a semi-supervised learning problem, we obtained a closed-form expression for the estimated values of all variables using harmonic extension. The reconstructed trajectories provide the basis for the more advanced mechanistic studies of multivariable processes responsible for the highly reproducible dynamics of developmental pattern formation. Our approach can also be extended using other semi-supervised learning methods [25], if the dimensionality of the intrinsic geometry is greater than one or if there is no unique common channel among all experiments.
Most of the previous attempts to accomplishing this task explored specific features of developmental systems, such as the expression level of a particular gene, and used a discrete number of temporal classes, usually defined in ad hoc way [16, 26]. Our approach reconstructs continuous time dynamics and relies on the intrinsic geometry of multidimensional datasets. Some limitations might appear when considering fluorescent reporters for intrinsically variable processes, and thus not smooth, such as MS2 reporters for nascent transcripts [27]. However, our method is readily applicable to datasets stored in established public databases of gene expression patterns such as the BDGP Resources [28] or the FlyEx database [29] and could serve to animate other pathways such as the segmentation cascade in the early fly development.
We conclude by pointing out two directions for the future extensions and applications of the presented approach. First, while there are no conceptual limitations in using the presented matrix completion framework to studies of pattern formation and morphogenesis problems in three dimensions [30], it is important to increase the computational efficiency of our approach, which can be done at multiple levels, starting with dimensionality reduction at the preprocessing step. At the same time, for a large class of patterning processes that happen on the surfaces of epithelial sheets, one can use the recently developed “tissue cartography” approach to first flatten the three-dimensional images [5], which should make our approach directly applicable. Second, following the step of data fusion, one can attempt to model the observed multivariable dynamics. Here one can employ several modeling methodologies, from mechanistic modeling of specific molecular and tissue-level processes [31–35], to equation-free approaches, which aim to deduce the underlying mechanisms directly from data [36, 37].
Extended Materials and Methods are presented in S1 Text.
All images are cross-sections of Drosophila embryos taken at ∼ 90μm from the posterior pole. Time-lapse movies were obtained using a Nikon A1-RS confocal microscope with a 60x Plan-Apo oil objective. The nuclei were stained with Histone-RFP. A total of 7 movies was acquired with a time resolution of 30 seconds per frame. All movies start about 2.5 hr after fertilization and end after about 20 min after gastrulation starts (about 3.3 hr after fertilization). Four datasets of fixed images were acquired to visualize nuclei, protein expression of dpERK, Twist, and Dorsal, and mRNA expression of ind and rho. Immunostaining and fluorescent in situ hybridization protocols were used as described before [16]. DAPI (1:10,000; Vector laboratories) was used to visualize nuclei. Rabbit anti-dpERK (1:100; Cell Signaling), mouse anti-Dorsal (1:100; DSHB), rat anti-Twist (1:1000; gift from Eric Wieschaus, Princeton University), sheep anti-digoxigenin (1:125; Roche), and mouse anti-biotin (1:125; Jackson Immunoresearch) were used as primary antibodies. Alexa Fluor conjugates (1:500; Invitrogen) were used as secondary antibodies. Stained embryos were imaged using Nikon A1-RS confocal microscope with a 60x Plan-Apo oil objective. Embryos were mounted in a microfluidic device for end-on imaging, as described previously [16, 38]. The first dataset contains 108 images stained with rabbit anti-dpERK and rat anti-Twist antibodies. The second dataset contains 59 images stained with mouse anti-Dorsal antibody, rabbit anti-dpERK antibody, and ind-DIG probe. The third dataset contains 58 images stained with ind-biotin probe, rho-DIG probe, and rabbit-dpERK antibody. The fourth dataset contains 30 images stained with rat anti-Twist antibody, ind-biotin probe, and rho-DIG probe. The distribution of the datasets as labeled and unlabeled data depending on the considered variable is summarized on S2 Table. Raw images can be found in Supplementary Files on the public github repository https://github.com/paulvill/data-fusion-images, see S2 Text.
The affinity matrix W = (wi,j) is computed using a Gaussian kernel w i , j = exp ( - ∥ x i - x j ∥ 2 σ i σ j ) with scaling parameters σi and σj computed locally as the average of the distance with respect to the 10 closest neighbors as described in S1 Text. We used the Euclidean norm in the space of scattering transformed data. The resulting affinity matrix is shown on S3 Fig. The corresponding underlying one-dimensional manifold is shown on S4 Fig.
The K-fold cross validation error was computed by extracting subsamples of the labeled data points and the semi-supervised learning framework was used to predict the value of the labels on them. For the image datasets, we computed the absolute error between the actual value of pixel intensity to the predicted one. The absolute error was then normalized by the range of the signal computed from the entire set of images for a given channel. The number of bins K was chosen so that the number artificially unlabeled data points was about 20. The results for each dataset are shown in S3 Table and described in S1 Text.
The result of data fusion led to multimodal time lapses of developing embryo showing nuclei and the spatio-temporal dynamics of dpERK, Dl, rho, ind, and Twi. The images were colored using the color code shown in S4 Table, i.e. dpERK (red), Dl (pink), rho (yellow), ind (blue), Twi (green). A resulting colored movie is provided in Supplementary Files 2.
The semi-supervised framework used to accomplish the task of data fusion is completely implemented in the open-source MATLAB library and fully runs in GNU Octave. It is available as Supplementary Software on the public github repository https://github.com/paulvill/data-fusion. See S2 Text for a description of the main components of the library.
|
10.1371/journal.pcbi.1005044 | Neuromotor Noise Is Malleable by Amplifying Perceived Errors | Variability in motor performance results from the interplay of error correction and neuromotor noise. This study examined whether visual amplification of error, previously shown to improve performance, affects not only error correction, but also neuromotor noise, typically regarded as inaccessible to intervention. Seven groups of healthy individuals, with six participants in each group, practiced a virtual throwing task for three days until reaching a performance plateau. Over three more days of practice, six of the groups received different magnitudes of visual error amplification; three of these groups also had noise added. An additional control group was not subjected to any manipulations for all six practice days. The results showed that the control group did not improve further after the first three practice days, but the error amplification groups continued to decrease their error under the manipulations. Analysis of the temporal structure of participants’ corrective actions based on stochastic learning models revealed that these performance gains were attained by reducing neuromotor noise and, to a considerably lesser degree, by increasing the size of corrective actions. Based on these results, error amplification presents a promising intervention to improve motor function by decreasing neuromotor noise after performance has reached an asymptote. These results are relevant for patients with neurological disorders and the elderly. More fundamentally, these results suggest that neuromotor noise may be accessible to practice interventions.
| It is widely recognized that neuromotor noise limits human motor performance, generating errors and variability even in highly skilled performers. Arising from many spatiotemporal scales within the physiological system, the intrinsic noise component is commonly assumed to be invariant by most computational models of human neuromotor control. We challenge this assumption and show that after an individual has reached a performance plateau, amplifying perceived errors elicits continued reductions in observed variability. Model-based analyses show that the main driver of this effect is a reduction in the variance of neuromotor noise. Thus, error amplification has the potential to become a key intervention for individuals with increased movement variability due to high levels of neuromotor noise, ranging from children with dystonia, through patients with stroke, to healthy elders.
| A hallmark of human movement is its variability, recognized in the phrase “repetition without repetition” [1]. Even the world champion in dart throwing does not hit the bull’s eye every time—in fact, it is this remaining randomness that creates the thrill in championships. This variability arises from interactions between a number of computational and physiological processes. Computationally, overt movement variability stems from imperfect error corrections and exploratory actions, and can sometimes be decreased by channeling variability into task-irrelevant dimensions [2–4]. Physiologically, some of the observed variability arises from a large number of processes at all levels of the neuromotor system, from noise in ion channel dynamics and action potential firing rates, to changes in the amounts of neuromodulators, such as serotonin or norepinephrine, that themselves depend on systemic factors, such as arousal [5–7]. While world-class champions may be close to a minimum in both computational and physiological sources of variability, older adults and patients with various neurological diseases show greater variability in their limb movements. For the latter, a behavioral intervention or therapy that decreases this overt variability and its underlying sources could have a major impact on motor function and quality of life. However, it is yet unclear whether or how this intrinsic noise can be accessed through behavioral interventions.
Any intervention targeting undesired variability in human movement must consider a fundamental component of motor performance: error. Error information is needed for learning and, consequently, manipulation of error influences learning [8, 9]. Error is typically manipulated physically or visually. Physically, external forces can guide an individual towards or away from a desired trajectory, causing either error reduction or amplification [10–15]. Visually, non-veridical action outcomes can be displayed that are better or worse than the actual performance. Importantly, studies on visual error amplification have shown that the manipulation accelerates learning and achieves enhanced performance [16–19].
Despite the promise of error amplification, the ways that this manipulation affects the human sensorimotor system remain unclear. Mechanistically, error amplification may influence how individuals correct for errors, i.e. changing the error correction gain (the proportion of error corrected on each trial). It has been proposed that error amplification may simply encourage an individual to make larger corrections, thus improving performance [18, 19]. While plausible, there may be more explanations for the observed benefits, which may not be apparent without a different approach. Prior work has used the framework of motor adaptation, where externally imposed errors are central and are gradually diminished to reinstate baseline performance. These previous studies have only assessed the effects of error augmentation on the exponential decline of errors, without separating this decay constant from the error correction gain. Further, the error correction gain may depend on the level of neuromotor noise, as the presence of noise may cause instability with a large gain [20].
To achieve a better understanding of the mechanisms underlying performance improvement with error amplification, computational learning models may be of significant value. Van Beers [21] used a stochastic state space model for a pointing task; the model introduced noise during two sequential stages of the error correction process. Modeling of the time course of this pointing task suggested that the optimal correction gain depended on the relative magnitude of the two noise sources. Using the same model to account for learning a throwing task, Abe and Sternad [22] showed that the error correction gain increased with practice. Hence, if one assumed that humans tend to use conservative gains below the optimal values to avoid instability, then it is reasonable to expect that error amplification improves performance by encouraging larger correction gains. Alternatively, the neuromotor system could lower its level of random noise, by changing computational strategies, such as channeling variability into task dimensions that reduce the error sensitivity of movement strategies [2, 3, 23, 24, 25]. Or, more simply, the system may lower the amplitude of its intrinsic noise via more system-level physiological mechanisms. The present study questions to what degree this is possible.
Physiological sources of noise are ubiquitous on all scales, from molecular to cellular to systemic. However, it is yet unknown whether they are modifiable by practice and interventions. This study explores whether error amplification could act as a simple aid to not only alter control strategies, but also lower neuromotor noise. A simple throwing task in a virtual environment was used as an experimental platform to test whether visual amplification of error affects motor performance, not only by changing how errors are corrected, but also, more importantly, by modifying neuromotor noise. Our focus was on improvements after performance has plateaued: by examining “already good” performance, the effects of error amplification are isolated from initial learning transients and sensorimotor calibration processes.
To further probe the status of this intrinsic noise, we not only used deterministic error amplification but also used stochastic amplification, where the amplification included a random component. We reasoned that layering additional noise on the amplified error would magnify any effects on intrinsic noise. Amplifying error deterministically makes performance appear worse by proportionally increasing both the magnitude and variability of errors, i.e. the coefficient of variation remains relatively unchanged. By disproportionately increasing variability by adding additional randomness, the coefficient of variation will increase. We conjectured that stochastic error amplification would make subjects perceive themselves as more noisy and variable compared to deterministic amplification alone, which would increase the pressure to reduce noise.
The experiment tested the following hypotheses: Hypothesis 1: Amplifying perceived errors improves task performance. Hypothesis 2: Adding noise to the amplified errors improves task performance more than deterministic error amplification alone. Three more hypotheses were tested using three stochastic iterative learning models (including an error correction gain and noise terms) to analyze the temporal structure of the task performance data: Hypothesis 3: Error amplification increases the size of error corrections, i.e. it increases the correction gain. Hypothesis 4: Error amplification reduces intrinsic neuromotor noise. Hypothesis 5: Stochastic amplification reduces intrinsic noise more than deterministic error amplification. Finally, we expected that there is an optimal error amplification magnitude that has the greatest effect on task performance and intrinsic noise.
Forty-two right-handed healthy individuals (22 male; 20 female; age: 24±5 years) participated in the experiment. All subjects received an explanation of the experimental task; however, they were not informed about any manipulations. Subjects read and signed an informed consent document approved by Northeastern University’s Institutional Review Board.
The task was designed to emulate the skill of throwing an object to hit a target. Specifically, the task was modeled after the table version of the “skittles” pub game (or tetherball). In this game, a ball is suspended from a string or chain attached to the top of a vertical post, and the ball is thrown to hit a target skittle on the other side of the post. The trajectory of the ball is fully determined by the ball’s release angle θR and angular velocity θR˙. The game is under-determined or redundant, as there are an infinite number of θR and θR˙ combinations that result in a successful hit for a given skittle location [23, 25].
Participants practiced a two-dimensional version of the skittles throwing task in a virtual set-up with an instrumented lever arm that rotated in the horizontal plane (Fig 1A). A cushioned splint was fixed to the lever arm to secure the participant’s dominant arm in the apparatus. A wooden ball of 6 cm diameter was fixed to the end of the lever arm and subjects grasped this ball with their hand. The distance between the lever arm axis of rotation and the ball was adjustable to the subject’s arm length. The angular position of the skittles lever arm θ was measured with a potentiometer (Bourns, Inc., Riverside, CA). A force-sensing resistor (Trossen Robotics, Westchester, IL) positioned on the surface of the ball was used as a switch to indicate release by converting the analog force signal to a binary open/closed signal using a Schmidt Trigger (Fairchild Semiconductor Corp., South Portland, ME). The switch was positioned so that the index finger closed the switch when the ball was grasped. The switch opened when the finger was extended and this simulated ball release. Data were sampled at 700 Hz using a personal computer and 16-bit analog-to-digital converter (DT300, Data Translation, Inc., Marlboro, MA). A rear-projection screen (width: 1.8 m; height: 1.4 m) was placed 0.6 m in front of the subject and the skittles manipulandum.
A custom-written acquisition program (Visual C++, Version 6.0, Microsoft, WA) sampled θR and the switch state and gave visual feedback to participants (Open GL, Version 1.2, Silicon Graphics, CA). Angular velocity θR˙ was calculated online using a linear regression over the most recent 10 data samples. At ball release, the slope of the regression was used as the velocity estimate. (Regression attenuated the influence of measurement noise.) During task performance the positions of the lever arm, ball, center post, and target skittle were displayed on the rear-projection screen (Fig 1A). The participants saw a top-down projection of the workspace; there were no reports of difficulty with this 90° transformation. When the lever arm was moved and the virtual ball was released (at the switch opening time), the program calculated the ball’s trajectory using the model of Müller and Sternad [23]:
x(t)=Axsin(ωt+φx)e−(tτ)
(1)
y(t)=Aysin(ωt+φy)e−(tτ)
(2)
where τ was the time constant of the decaying trajectory (τ = 20 s), with amplitudes Ax and Ay, given by
Ax=xR2+[ x˙Rω+(xRτ)ω ]2
(3)
Ay=yR2+[ y˙Rω+(yRτ)ω ]2
(4)
where xR, yR and x˙R, y˙R are the positions and velocities of the ball at release, respectively, with Ex and Ey denoting the kinetic and spring potential energies determined by
Ex=0.5(mx˙R2+kxR2)
(5)
Ey=0.5(my˙R2+kyR2)
(6)
where m = 0.1 kg and k = 1.0 N/m. The phases φx and φy of the sinusoidal motions of the two springs were based on the oscillation amplitudes and the x and y release positions (xR and yR, respectively) and are given by
φx=arccos[ 1Ax(x˙Rω+(xRτ)ω) ]
(7)
φy=arccos[ 1Ay(y˙Rω+(yRτ)ω) ]
(8)
and the natural frequency ω was
ω=km−1τ2=3.16 rad/s
(9)
After the ball passed the target skittle, the minimum distance between the ball trajectory and center of the target skittle D-min was calculated (Fig 1B). The color of the target skittle changed from yellow to red, if D-min < 0.011 m, signaling a successful hit to the subject.
The post (radius = 0.25 m) was in the center of the workspace at x = 0, y = 0 m, while the elbow axis of rotation was located at x = 0, y = -1.5 m. The target skittle (radius = 0.05 m) was positioned at x = 0.05, y = 1.05 m. Fig 1B illustrates three different trajectories and the two insets show the coordinates of the arm including the target and error definition. Fig 1C shows the execution space that represents all possible combinations of the execution variables θR and θR˙ and the result variable D-min. Combinations of ball release angle and velocity, θR and θR˙, that give exact target hits, D-min = 0, define the solution manifold, which is shown by the red vertical line. For this study, the target position was chosen such that errors depended primarily on θR and only very little on θR˙, i.e. the solutions were largely insensitive to release velocity. The release angle that yielded a perfect hit was θTarget = 1.44 rad. An example of 240 throws produced by an experienced skittles player is shown in the solution manifold, as well as three artificial numbered data points showing different possibilities. Note that other target constellations are associated with nonlinear U-shaped solution manifolds where the D-min depends on both release angle and velocity, θR and θR˙ [25].
Specific to this target constellation was also that the ball could only pass below the target and the smallest achievable distance error (D-min) was 5 mm, corresponding to θR = 1.44 rad. In Fig 1D, the top left panel illustrates the x-y workspace with the target in red; each black circle represents closest approach of the ball (D-min) for a set of exemplary trajectories. Note that subjects could distinguish between the deviations to the left and the right branch from the perfect release angle by the elliptic shape of the ball trajectories. Therefore, error corrections based on visual information were relatively straightforward. The top right panel of Fig 1D shows that the relation between θR and D-min is nonlinear and approximately parabolic.
Given the geometry of the workspace and the definition of error D-min as always positive, the errors typically assumed a very skewed distribution (Fig 1D, lower left panel). For statistical analysis, we therefore transformed this data using −1/x2, based on Tukey's "ladder of powers" [26]. This choice was made after trying several transformations of increasing severity until an approximately uniform distribution was obtained that afforded calculating means (Fig 1D, lower right panel). For the presentation of the results, the errors were transformed back.
To manipulate the distance error D-min, only the release angle θR had to be modified, as the solution manifold was insensitive to variations in the release velocity θR˙. Therefore, the release angle was measured online to calculate the observed ball trajectory and then modified to amplify the error as follows
θ˜R=θTarget+A(θR−θTarget)
(10)
where θ˜R denotes the manipulated release angle, θTarget the release angle that gave a perfect hit, i.e. D-min = 0, and A the amplification gain. Here, the magnitude of error amplification is held constant (for a given A), and therefore this manipulation is called deterministic error amplification, or DEA. Three magnitudes of A were applied: A = 1.5 (DEA-1.5), A = 2.0 (DEA-2.0), and A = 2.5 (DEA-2.5). Note that the ball trajectory that the participants saw was calculated based on θ˜R, while the angular velocity θR˙ remained unchanged.
To stochastically amplify error (SEA), the manipulation was
θ˜R=θTarget+(A+ξ)(θR−θTarget)
(11)
where ξ was uniformly distributed noise on the interval [−(A−1), +(A−1)]. These interval boundaries ensured that the noise was centered on A and scaled with A. For example, for A = 2.0 the noise interval was [–1, 1] creating random amplification gains in the range of [1, 3]. For A = 1.5 the range was [1, 2], for A = 2.5 the range was [1, 4]. Corresponding to the DEA levels, the three mean or effective gains A¯=A+ξ were: A¯=1.5 (SEA-1.5), A¯=2.0 (SEA-2.0), and A¯=2.5 (SEA-2.5). Errors were only amplified up to a limit; if the manipulated release angle θ˜R fell outside the range of 1.00 to 1.88 rad (optimum = 1.44 rad) subjects saw their true error. This prevented subjects from noticing the manipulations. Pilot work had shown that when subjects perceived very large errors, they “discount” them as external perturbations [27].
The 42 subjects were randomly assigned to one of seven groups: there were six experimental groups that received varying types and amounts of error amplification, and one control group without manipulation (Fig 2). Three groups received DEA and three received SEA. Each amplification type was applied at three levels: 1.5, 2.0, and 2.5. The experiment comprised 6 daily sessions of practice divided into two stages: On the first three days of practice, all participants performed the virtual task without any experimental manipulations. During the second three days, the visual feedback was altered for the six experimental groups (DEA and SEA). Altogether, there were 252 separate practice days with data collection. On each practice day the participants performed four blocks of 60 trials each, to yield 240 trials per day and 1440 total trials for each subject.
Participants placed their right forearm into the cushioned splint attached to the lever arm. The height of the lever arm and the distance between the axis of rotation and the ball were adjusted to place the forearm at a comfortable height with the ball firmly grasped. The participants were instructed to hit the center of the target skittle with the ball and to avoid hitting the center post. Each throw typically lasted 1–2 s with about 3 s between the self-initiated throws. The experiment duration in each daily session was about 20 min.
For each subject/day the 240 trials were parsed into non-overlapping bins of 20 trials. Participants occasionally hit the center post; these trials were replaced using bootstrap resampling [28]. This method computed an estimate of the sampling distribution of the non-post-hit trials and then replaced the post-hit with a new value randomly drawn from the estimated distribution. Trials with post hits were not eliminated because this would have disrupted the iterative trial-to-trial process, which was the basis for the second set of analyses on the temporal structure of the data using system identification. Within each 20-trial bin the data were averaged to obtain the mean distance error D-min. While the -1/x2 transformed D-min was used for statistical analysis, the figures display the more intuitive untransformed D-min values. A preliminary analysis of D-min for Day 3 revealed one outlier subject in the group DEA-2.0 and one subject in the group SEA-2.5, whose means were outside 1.5 times the interquartile range (±2.7 standard deviations or 99.3% of a normal distribution). These two subjects were excluded from the analysis; thus, for DEA-2.0 and SEA-2.5 the number of participants was five. Note that the variations in θR were not analyzed, because after initial practice these variables were centered on the optimal release angle and therefore showed the same patterns as the mean D-min. This is consistent with previous work that showed that data distributions centered on the most error-tolerant location early in practice [2].
Due to the extensive experimental protocol, i.e. seven groups with six days of practice for each subject, the number of subjects in each group was relatively small and did not satisfy the assumptions of ANOVA: normality and equality of variances. Therefore, the data were analyzed with permutation tests, which are a subset of non-parametric statistics [29, 30]. This analysis method uses permutations to create a sample-specific distribution, instead of using an assumed theoretical distribution. A cut-off for a given p-value was obtained from the specific distribution. The permutation analysis involved several steps. First, the data from all subject groups were pooled, based on the null hypothesis that all groups were part of the same population. In the subsequent resampling procedure, the data were randomly shuffled, split into two groups, and the difference between the group means was recorded. This procedure was repeated 1,000,000 times, resulting in a distribution of group mean differences that represented the probability of obtaining a given difference between two groups randomly selected from the subject population.
For all statistical comparisons the difference between the relevant means was compared with the bootstrap distribution. The p-value for each comparison was calculated by dividing the number of bootstrap differences smaller than the actual group difference by 1,000,000 and multiplying by 2 (to give the p-value for a two-tailed test). The critical threshold for significance was set to p < .05, meaning that there had to be 25,000 or fewer bootstrap differences below the actual group difference. For example, if only 1000 bootstrap differences were below the tested group difference, the p-value for a two-tailed test was p = .002.
To evaluate Hypothesis 1—amplifying perceived errors improves task performance—statistical tests were performed to test for differences in D-min between each error amplification group and the control group. These tests were conducted for Day 3, Day 6, and the change or difference between Day 3 and 6. Evaluating D-min on Day 3 tested for between-group baseline differences before error amplification was applied. Evaluating D-min on Day 6 assessed the level reached after three days of practice with amplified error. Examining the change between the days (Day 6 –Day 3) provides a direct assessment of the manipulation effects over time. To test Hypothesis 2—stochastic amplification of perceived errors improves task performance more than deterministic error amplification—the D-min for each pair of DEA and SEA groups (within an error amplification level, i.e. 1.5, 2.0, or 2.5) was compared. Finally, to test for possible differences among error amplification levels, i.e. is there an optimal error amplification gain, each DEA group was compared with each other, and each SEA group was compared with each other. The transformed D-min (−1/x2) was used for all statistical tests.
To test Hypotheses 3, 4, and 5, which were concerned with the effects of error amplification on the error correction gain and neuromotor noise, we analyzed the error time series across practice using system identification techniques with three different learning models. Results of system identification are evidently dependent on the model used. Therefore, we tested three models that presented a step-wise increase in complexity. All three models included the basic component of any learning model—an error correction gain B. Two of the models quantified intrinsic motor noise via a single noise source, and the third model introduced two independent noise sources. Due to the different model structures they required different methods of system identification. The goal of these analyses was to tease apart contributions in the overt performance change due to the error correction gain or from noise. Note that results revealed that Model 1 and 2 were not as suitable as Model 3. We nevertheless present all three models to highlight that the model structure significantly determined the results. However, as the data show, one result was invariant across the three model structures. Details about the models and procedures follow.
The first model contained two iterative steps with the addition of one noise sample, described by the following equations [21]:
θR,i=θPL,i+ηi
(12)
ei=θR,i−θTarget
(13)
θPL,i+1=θPL,i−Bei
(14)
where θPL,i is the planned release angle at trial i, B is the error correction gain, θR,i is the actual release angle, ηi is a sample from a zero-mean Gaussian distribution with variance equal to σ2, ei is the error between the release angle θR,i and the angle that hits the target θTarget. This model assumes that the actual executed release angle θR,i is equal to the internally planned release angle θPL,i with added motor noise ηi (the labels “planning” and “execution” should not be taken literally). The planned angle θPL,i is updated trial-by-trial according to the visual error and correction gain B.
Note that either the actual or manipulated release angle, θR or θ˜R, could be used in the system identification. The interpretation of B was dependent on this choice. Using θR means that if participants fully adjusted the size of their corrections in response to error amplification, then an increase in B should be observed, and this increase should match the error amplification gain. If θ˜R was used, then B remained unchanged if participants increased the size of their corrections in proportion to the amplified errors. For modeling SEA effects, using an amplified θ˜R is non-trivial, because an additional noise term would be needed. To minimize the number of unknown model parameters we used θR in the system identification.
To estimate the two unknown parameters B and σ2, the equations were rearranged and combined into a single equation; θTarget was set to zero. First, Eq 12 was increased by one iteration step:
θR,i+1=θPL,i+1+ηi+1.
(15)
Then, Eqs 12 and 15 were inserted into Eq 14, giving
ei+1=ei(1−B)+ηi+1−ηi.
(16)
System identification was applied to the time series of execution angles θ with the target angle subtracted, according to Eq 13. Model validation procedures showed that the system identification of Model 1 was associated with positive biases in σ2 (see S1 Appendix). In addition, system identification of the experimental data showed that estimates for B were close to zero, and prior research has shown that modeling with only one noise source was inadequate [21, 22]. Thus, we introduced Model 2 that added a second source of noise.
This model is a simple extension of Model 1 by adding a second sample of motor noise into the “planning” stage:
θR,i=θPL,i+ηEX,i
(17)
θPL,i+1=θPL,i−Bei+ηPL,i+1
(18)
K=ηEX,iηEX,i+ηPL,i=ηEX,iηTOTAL,i
(19)
where ηTOTAL,i was a sample drawn from a zero-mean Gaussian distribution with variances equal to σ2. ηTOTAL,i was separated into ηPL and ηEX and their magnitudes were constrained by the ratio K. With this constraint, ηPL and ηEX were not independent. The parameter K was used to describe the noise ratio (Eq 19). B was the error correction gain; the error in the release angle was given by:
ei=θR,i−θTarget.
(20)
To estimate the three unknown parameters B, K, and σ2 Eqs 17 to 20 were rearranged into the form of a regression equation as for Model 1:
ei+1=ei(1−B)+ηTOTAL,i+1−KηTOTAL,i.
(21)
As can be seen, Model 1 is a particular case of Model 2 with K = 1. Again, system identification was performed on the angle errors as defined in Eq 21. The constraint K enabled system identification with a linear regression model. More importantly, this constraint “colored” the noise, i.e. the output noise showed long-range correlations with a 1/f distribution that depended on K. A wide range of studies ranging from brain oscillations to motor behavior, such as tapping, posture and walking, have shown that motor noise is colored [31–34]. In contrast to Model 1, validation of Model 2 showed reliable noise estimates (S1 Appendix). However, simulation results also showed that about 38% of the experimental data rendered negative K values (S2 Appendix). Based on the definition in Eq 19, K should be positive within [0, 1]. These results suggested that Model 2 was not appropriate for those blocks of data. Thus, we introduced Model 3 that separated execution and planning noise into two independent quantities.
In this model, we assumed that execution and planning noise were independently generated from two random processes. This model was previously shown to account for the observed structure in the variability of human motor actions [21, 22]. The equations describing the model’s behavior were:
θR,i=θPL,i+ηEX,i
(22)
θPL,i+1=θPL,i−Bei+ηPL,i+1
(23)
ei=θR,i−θTarget.
(24)
where ηEX and ηPL were random samples from two independent zero-mean Gaussian noises with different noise variances; ei denoted the angle error of sample i, and B was the error correction gain.
System identification was applied on the experimental data using Models 1, 2, and 3. Consistent with the models, the measured release angle θR was converted to error by subtracting the target angle (1.44 rad) from each data point. Note that the error in angle could be zero, unlike the distance error calculated in the x-y workspace. For each subject, the estimations were conducted for each block of 60 trials, yielding four separate estimates per day for each subject. Initial transients were eliminated by excluding the first 10 trials of Block 1 on each day.
For Models 1 and 2 the MATLAB System Identification Toolbox (version 9.2) was used with the function pem.m (Prediction Error Method) to find estimates for the unknown parameters (for Model 1: B and σ2; for Model 2: B, K, and σ2). This algorithm estimated the parameter vector Θ by minimizing the squared prediction error [35]:
Θ^=arg minΘ1tΣi=1tεi2(Θ)
(25)
where εi(Θ) = θR,i − fi|i−1(Θ) is the prediction error with Θ = {K,B,σ2} and fi|i−1(Θ) is an optimal predictor:
fi|i−1(Θ)=(1−B)ei−Kηi−1
(26)
To optimize Eq 25, we applied a nonlinear least-square curve fitting algorithm with the Levenberg-Marquardt Method using the function lsqnonlin.m of the MATLAB Optimization Toolbox (version 7.2). Since Model 3 included two independent noise sources, a different identification method was needed. Thus, a Maximum Likelihood Estimation (MLE) was performed using the Expectation-Maximization (EM) algorithm [36] to identify B, σPL2, and σEX2. The maximum likelihood estimator of Eqs 22 and 23 was given by:
Θ^=arg maxΘ logp(θR,1,⋯,θR,t|Θ; e1,⋯,et)
(27)
where
Θ≡{−B,σEX2,σPL2}.
(28)
To test the validity of the system identification methods, Monte-Carlo simulations with known parameters were performed for all three models, followed by the identification of the parameters. Details and results of the validation are shown in S1 Appendix. For Models 1 and 2, the estimation of B and σ2 had a significant positive bias; K was underestimated in Model 2, especially for higher values. In these cases, we used the Adjusted Yule-Walker (AYW) method [37] to provide an unbiased estimate of B (see S1 Appendix for details). On the other hand, validation of Model 3 showed that the estimation provided unbiased estimates of all model parameters.
Based on the model validation and estimation results, Model 3 was deemed to be most appropriate. Therefore, only the results for this model are presented below and the results for Models 1 and 2 are relegated to S2 Appendix. Focusing on Model 3, three dependent variables were analyzed: the error correction gain B, the execution noise variance σEX2, and the variance of planning noise σPL2. Each parameter estimate was computed for Day 3, Day 6, and the change across the manipulation (Day 6–3). Hence, these measures received the same statistical treatment as the behavioral measures.
To test Hypothesis 3—error amplification increases the size of corrective actions—comparisons were made for B between each experimental group and the control group for Day 3, Day 6, and the change from Day 3 to Day 6. To test Hypothesis 4—error amplification reduces intrinsic neuromotor noise—similar comparisons were made for the two noise estimates. To test Hypothesis 5—stochastic amplification reduces noise more than deterministic error amplification—comparisons were made between DEA and SEA as already described for D-min above. The same tests were also performed to identify differential effects across the different amplification gains.
To further tease apart potential mechanisms underlying the decrease in overt variability, the time series of release variables were examined via stochastic learning models. In particular, three models were used to extricate the contribution of the error correction gain from random noise sources. Model parameters were estimated before and after error amplification using system identification. Parameter estimation was conducted separately for each block of 60 trials; there was a total of 1008 blocks across all days and subjects (42 subjects, 6 days, 4 blocks on each day). The 60 trials presented a sufficiently long time series that also avoided potential drifts that may have otherwise confounded the parameter estimation.
Examples of the raw data of angular error e = θR − θTarget used in the system identification are displayed for three subjects in different groups in Fig 7. Note this error could be both positive and negative, unlike the distance error D-min. In addition, the deviation of the release angle to the optimal angle could be reduced to zero. The open black circles denote errors in unmanipulated trials; the closed colored circles represent the amplified errors as subjects saw on the screen. The SEA condition had clearly a wider range of amplified errors than DEA. The long sequence of 1440 trials showed a very gradual, almost invisible change in release angle error; nevertheless, the average values showed a clear reduction across days as previously seen in D-min (Fig 3).
This study demonstrated how error amplification affected error and variability in the performance of a perceptual-motor task. Performance at asymptote was examined to downplay the error correction processes that naturally predominate in the early stages of learning. Results showed that visual amplification of error achieved decreases in mean error, even after subjects had reached a performance plateau. A model-based analysis of trial-by-trial error corrections revealed that error amplification improved performance mainly by reducing the level of neuromotor noise, rather than by modifying the error correction process itself. Although the observed decrease in the random components of motor behavior cannot be attributed to any specific physiological structure at present, this study represents a critical step towards understanding how error amplification can benefit motor performance. The finding that improvements can be attained after subjects had reached an asymptote is significant for therapy, when patients are unable to make any further progress or may be stuck in a “local minimum”.
All participants underwent extensive practice over three days (720 trials) until their performance had reached a plateau. Six experimental groups that continued to practice for three more days with visually amplified errors (another 720 trials) further decreased their errors, while the control group did not. This improvement was seen for both deterministic and stochastic amplification. Corroborating previous demonstrations [13–19], these findings present clear evidence that amplifying perceived errors improved task performance (Hypothesis 1).
Extending prior work, three different amplification gains were compared to parametrically assess the effect of error augmentation magnitude. As expected, the lowest gain of 1.5 was not effective, neither in stochastic nor in deterministic form. The two larger gains of 2.0 and 2.5 elicited improvements by virtue of a decreased mean absolute error. Although in some cases the gain of 2.0 elicited the greatest improvements, this could have been because this group was worse than others prior to the manipulation. The highest amplification gain 2.5 did not produce any “instability”, as speculated previously. Wei et al. [18, 19] stated that for their perturbed reaching task an error amplification gain of 3.1 would lead to overcorrection and instability based on an error updating model [20], a result that was supported by their experimental data. While the higher amplifications in the present task were successful, one must keep in mind that the specific numerical values are most likely task-dependent.
We also posited that the addition of stochastic noise may enhance the effect of amplification (Hypothesis 2). However, the expectation that the additional noise would add “pressure” to reduce the variability of their performance was not confirmed. We speculate that subjects may simply have “averaged” over the noisy errors, and were therefore insensitive to the immediate error information presented after each trial.
For any account of learning, there are two primary options to improve overt errors in the performance variable: optimize the error correction gain and/or reduce the internal noise variance. To tease these two options apart from the overt variability of the task performance, the present study applied system identification procedures based on three stochastic iterative learning models. All three models included two stages, motivated by previous electrophysiological research that identified significant contributions of noise in the planning processes preceding movement, distinct from execution [40, 41]. A model with two stages of noise was also supported by previous behavioral and modeling studies that assessed structure in the trial-by-trial changes in an aiming task and the skittles task [21, 22].
The system identification results showed that error amplification elicited a small increase in the correction gain or learning rate B, which supports Hypothesis 3, and this increase became larger with larger amplification magnitudes (for SEA, not DEA). This result was provided by Model 3, which was deemed the most appropriate of the three models tested. Although subjects increased the size of their corrections, the increases remained modest, averaging about 5% at most, and fell significantly short of the amplification factors. If participants had fully adjusted the size of their error corrections to the amplification, then there should have been increases of 50%, 100%, and 150% for the three error amplification levels (1.5, 2.0, 2.5, respectively). This under-compensation agrees with prior work showing that humans typically do not respond proportionately to errors, i.e. they respond to larger errors less than would be expected [42]. The relatively small adjustments in participants’ error correction gains in response to error amplification suggest that this gain adjustment was not the prime driver of the observed improvements in performance.
The values for B were mostly between 0.1 and 0.2 (10%–20% of error corrected), which is somewhat lower than other studies, which have reported correction gains of about 0.38 in an aiming task [21] and in the range of 0.20–0.50 for a reaching task [36]. The lower values in the present study could arise from the fact that the actions were well-practiced. However, note that B results depend significantly on the choice of the model; therefore, it is important to compare different models to assess their suitability [e.g. 43]. In our study, Model 1 with only a single noise source and Model 2 with two noise sources were not in agreement (see S2 Appendix). By the last day of practice, both models estimated that half of the error amplification groups had higher B values than the control group, but these groups were not the same. For Model 2 the B values were generally large compared to Model 1 where the gains were often less than 0.1. Further, when noise is part of the dynamics, parameter estimation methods may produce a bias [37]. Given these discrepancies and potential estimation problems, we refrained from further interpreting the gains. Note that amplifying the errors in the modeling itself would not alter the relative changes in B, but would only increase the absolute values of B by a factor equal to the amplification gain (assuming subjects fully respond to the amplified errors).
The most robust result of the system identification, consistent in all three models, was that error amplification reduced the noise sources, supporting Hypothesis 4 (see also S2 Appendix). While execution noise showed a steady decline after error amplification was applied, planning noise declined more abruptly and reached a plateau (Fig 9). Although planning noise remained steady in the control group for most of their practice, there was a precipitous drop on the very last practice day. This behavior was consistent across all subjects in this group. Thus, even without error amplification, planning noise may eventually decline, only at a later stage. This observation may also signal that performance improvements due to error amplification may just accelerate internal learning processes, and not introduce new mechanisms. Related work has shown that even without any external rewards, individuals can eventually reach the same performance level as those that are rewarded, but do so at a slower time scale [25, 44].
While the variance of planning noise was much smaller than execution noise, this does not necessarily imply that planning noise had a negligible impact. For example, for the control group on Day 3, the two noise variances averaged to about 0.00045 and 0.011 rad2. While seemingly small, this equated to standard deviations of about 1.2° and 6.0° for planning and execution noise, respectively. Thus, the planning noise contributed about 21% to the variability in the release angle. Although it was also hypothesized that that stochastic amplification would reduce noise more than deterministic error amplification (Hypothesis 5), this hypothesis was not supported. Potentially, participants did not make trial-by-trial adjustments to their actions in response to the visually amplified errors, but instead may average across multiple trials. It should also be kept in mind that stochastic amplification only added one random number to the release angle of each throw; their arm movements leading up to the release were not continuously perturbed. Hence, the added randomness may have been lost on the subjects.
A decrease in variability is often achieved by slowing down movement, i.e. a decrease in speed is traded for an increase in accuracy [45, 46]. The speed-accuracy trade-off has been proposed as a general signature of learning; a more skilled individual can execute movements faster and with greater precision [46]. Further, it is generally believed that signal-dependent noise is a basic property of neuronal signaling [47]. The effects of this noise are readily seen during isometric force production, with the variability of the exerted force increasing as a multiple of the force [48]. For dynamic tasks and trajectories this manifests itself as velocity-dependent noise. Counter to these expectations, participants in the present study did not exhibit systematic decreases in their release velocity across practice, consistent with previous results on a similar task [24]. This suggests that other avenues were available for the neuromotor system to reduce its overt fluctuations.
Previous research on the same throwing task highlighted three conceptually different avenues to decrease observed variability. Subjects can: 1) find solutions that are error-tolerant, 2) co-vary execution variables to minimize the effect of variability on the task result, and 3) reduce the variance of noise [2, 23, 39, 49, 50]. This work showed that error tolerance was achieved very early in practice, while co-variation was a computational strategy that was gradually exploited with practice. Noise remained the component least accessible to even extended practice [2]. In the present study, which included a relatively long practice schedule, tolerance was of little relevance after the first day of practice. Co-variation between the release angle and velocity could have been exploited to channel noise into task-irrelevant dimension, without necessarily reducing the overall amplitude of noise sources. In the present task, release velocity presented such a task-irrelevant dimension. However, counter to expectation, variability in release velocity did not increase concomitant to the decrease in angle variability. Hence, this computational strategy could not explain the lowering of noise. To improve performance subjects could only reduce the noise component.
The present target constellation rendered a very specific solution manifold that gives release velocity little contribution to the observed error. However, velocity is not entirely irrelevant. Both release angle and velocity are needed to calculate the ball trajectory. Importantly, the error sensitivity of throws with different velocities changes, as seen by a slight broadening of the neighborhood with low errors at higher velocities (Fig 1C). Subjects can visually distinguish between different release velocities as they lead to ball trajectories with different elliptic paths (Fig 1B). As shown in a previous study, subjects seek those more error-tolerant velocities and do not seek the lowest velocity that might be expected due to the least amount of signal-dependent noise [24]. Finally, small changes in target locations will make release velocity an important determinant of the distance error. Hence, the velocity dimension is not quite as redundant as it appears at first sight.
How might subjects have reduced overt noise and possibly physiological noise in response to error amplification? This question is partly motivated by a known mechanisms in songbird learning, where the nervous system can purposefully inject noise into the learning process during early-learning, and then reduce these self-induced perturbations after learning [51]. Might the human nervous system use noise in a similar way? Humans can mechanically reduce the influence of noise on task performance with antagonistic co-activation [52–54]. However, while antagonistic co-activation may increase in early learning, it typically reduces with practice [55]. In the present study, signal-dependent noise has been ruled out as an explanation because the release velocity did not systematically decrease with practice (assuming that velocity presents the “signal”) [47]. Further, there is some evidence that this noise can change with alterations in the physiological properties of motor units, such as reported in aging muscles [56], but it remains unknown whether this noise can be altered on a shorter time scale. One final speculation is derived from studies on the effect of neuromodulators, such as serotonin or norepinephrine, on motor neuron excitability [5, 6, 57]. Animal studies provided significant evidence that the descending drive to muscle contractions is gain-controlled to modulate the required output force. One study on humans showed that force variability increased after the brainstem–spinal cord neuromodulatory system was up-regulated [6]. A complex interplay of neuromodulators can excite or inhibit spinal cord excitability and thereby match precision demands of motor behavior. It is possible that this gain control is modifiable via error amplification. Evidently, more research is needed to solidify these conjectures.
As noted earlier, there could be other factors, such as enhanced perception and correction of errors and motivational factors that may contribute to improved performance with error amplification [18, 19]. In terms of perception, it could be that increasingly smaller errors become simply too small to detect and amplifying error resolves this problem. However, the present results do not support this conjecture: if the control group had been unable to detect their errors, then their error correction gains should have been close to zero. Counter to this expectation, although small, their gains remained significantly above zero, even after six days of practice. This only remains a possibility if one conjectures that subjects corrected their predicted errors based on internal models of the task.
Temporal structure of data can be analyzed with numerous methods, ranging from simple autocorrelation and other linear autoregressive methods to nonlinear methods, such as entropy analysis [58] or recurrence quantification [59]. Autoregressive methods maintain temporal connections and are strictly linear analyses. On the other hand, more sophisticated nonlinear methods, such as multi-scale entropy analyses or recurrence quantification, are useful for more continuous time series, as seen in postural control or heart rate [60–62]. We opted to use iterative learning models to analyze trial-to-trial sequences, where error correction processes explicitly link successive trials and there are explicit parameters of the two processes in focus. However, stochastic iterative models are clearly also not free of limitations.
First and common to all analysis is that temporal analyses select one single variable from a complex movement system. While the neuromotor system is unquestionably high-dimensional and has a highly distributed neural network, task success in the current task was described by a single kinematic variable, angle at ball release. This variable lent itself to be mapped on the single model variable. Second, in the task subjects viewed the distance error between their ball trajectory and the target, whereas in the model the error information was the difference in release angle to the optimal angle. Given the parabolic relation between the two, this may have influenced the result, lowering the estimates of B. However, after two days of practice most subjects were close to the optimal release angle, and therefore operated in a regime where the relation was close to linear. Probably more critical is that the mapping of the perceived error in workspace to the error in release angle may include further inaccuracies.
Another note of caution relates to the models themselves. The comparative analyses of three basic learning models highlighted that parameter estimates are very sensitive to the structure of the model. For example, all three models assumed additive noise. It is conceivable that the learning mechanisms include multiplicative noise. As we do not know the “true” model, results of these estimations should be interpreted with great care. Further, parameter estimation of a stochastic model has computational pitfalls, and the seemingly straightforward estimates of error correction gains become biased when noise is included [37], as we showed in S1 Appendix. We therefore only interpreted the changes in noise that was in agreement in all three models. We also tried to rule out alternative explanations and apply caution in the interpretation. Evidently, stochastic learning models with different forms of noise will have to be a topic for further research in motor neuroscience.
Finally, there is always the question of generalization: do the observed effects of error amplification extend to other motor tasks? For example, do the findings also hold for continuous tasks that involve tracking? How could error amplification be applied without a virtual interface? Do the results from the virtually controlled task generalize to real-world skittles throwing in 3D? These critical questions are clearly not confined to our study, but to almost all controlled experimental studies. While it is difficult to draw any definitive conclusions within the scope of a single study, the present results add to existing evidence. Prior studies have shown error amplification benefits in different reaching tasks [15–19] and a pinball-type task [13]. Further, beneficial effects of error amplification have been reported for different sensory modalities, including visual [16–18] and haptic feedback [10, 12]. Although these studies did not separate improvements due to noise reduction from changes in error correction, it is plausible that similar effects may be found. Of course, further research should substantiate this possibility.
The main results of this study was that error amplification elicited continued improvements in performance that otherwise plateaued, and this effect was mainly driven by a reduction in neuromotor noise. As such, error amplification presents a way of stimulating continued improvements in motor performance. The results challenge the assumption that neuromotor noise is invariant and inaccessible to behavioral interventions. Error amplification has the potential to become an effective intervention for improving motor performance in physical therapy and neuro-rehabilitation to improve motor function. This can be of special benefit for those who have reached a performance plateau, ranging from elite athletes, to patients with neuromuscular disorders, and to the elderly.
|
10.1371/journal.pmed.1002904 | Early initiation of breastfeeding and severe illness in the early newborn period: An observational study in rural Bangladesh | In Bangladesh, neonatal sepsis is the cause of 24% of neonatal deaths, over 65% of which occur in the early-newborn stage (0–6 days). Only 50% of newborns in Bangladesh initiated breastfeeding within 1 hour of birth. The mechanism by which early initiation of breastfeeding reduces neonatal deaths is unclear, although the most likely pathway is by decreasing severe illnesses leading to sepsis. This study explores the effect of breastfeeding initiation time on early newborn danger signs and severe illness.
We used data from a community-based trial in Bangladesh in which we enrolled pregnant women from 2013 through 2015 covering 30,646 newborns. Severe illness was defined using newborn danger signs reported by The Young Infants Clinical Science Study Group. We categorized the timing of initiation as within 1 hour, 1 to 24 hours, 24 to 48 hours, ≥48 hours of birth, and never breastfed. The analysis includes descriptive statistics, risk attribution, and multivariable mixed-effects logistic regression while adjusting for the clustering effects of the trial design, and maternal/infant characteristics. In total, 29,873 live births had information on breastfeeding among whom 19,914 (66.7%) initiated within 1 hour of birth, and 4,437 (14.8%) neonates had a severe illness by the seventh day after birth. The mean time to initiation was 3.8 hours (SD 16.6 hours). The proportion of children with severe illness increased as the delay in initiation increased from 1 hour (12.0%), 24 hours (15.7%), 48 hours (27.7%), and more than 48 hours (36.7%) after birth. These observations would correspond to a possible reduction by 15.9% (95% CI 13.2–25.9, p < 0.001) of severe illness in a real world population in which all newborns had breastfeeding initiated within 1 hour of birth. Children who initiated after 48 hours (odds ratio [OR] 4.13, 95% CI 3.48–4.89, p < 0.001) and children who never initiated (OR 4.77, 95% CI 3.52–6.47, p < 0.001) had the highest odds of having severe illness. The main limitation of this study is the potential for misclassification because of using mothers’ report of newborn danger signs. There could be a potential for recall bias for mothers of newborns who died after being born alive.
Breastfeeding initiation within the first hour of birth is significantly associated with severe illness in the early newborn period. Interventions to promote early breastfeeding initiation should be tailored for populations in which newborns are delivered at home by unskilled attendants, the rate of low birth weight (LBW) is high, and postnatal care is limited.
Trial Registration number: anzctr.org.au ID ACTRN12612000588897.
| Severe illness, including sepsis, is one of the leading causes of newborn deaths in low-and-middle-income countries and is responsible for 42% deaths in the early neonatal period (0–7 days).
WHO currently stresses the importance of initiating breastfeeding within the first hour of birth.
We identified studies conducted in Egypt, India, Nepal, Ghana, Tanzania, and Ethiopia in the past 2 decades that specifically examined the role of timing of breastfeeding initiation and neonatal survival.
We found no direct evidence reporting the mechanism by which initiation within the first hour of birth can reduce early newborn (0–7 days) deaths.
We report the effect of early initiation of breastfeeding on severe illnesses in the early newborn period using data from a large population-based cohort.
We defined severe illness using newborn danger signs reported in The Young Infants Clinical Signs Study Group.
Our results show that the earlier the initiation of breastfeeding, the lower the risk of having severe illnesses in the early newborn stage.
By accounting for possible reverse causality from infants too ill to initiate breastfeeding, we have established that starting breastfeeding beyond the first hour of life can double the likelihood of having severe illness.
Early initiation of breastfeeding within 1 hour of birth reduces neonatal mortality, and a reduction in the rate of severe illnesses, including suspected sepsis, likely mediates this effect.
Our work highlights the need to design and evaluate interventions to promote and support early initiation of breastfeeding that engage women who are at the highest risk of delaying initiation of breastfeeding, as well as those assisting them at delivery.
Policymakers need to prioritize the effective implementation of interventions to support early and sustained exclusive breastfeeding as priorities for global public health.
| Globally 2.7 million newborns died during the first 28 days of life (0–27 days) in 2015 [1]. More than a third of these neonatal deaths occurred on the first day, and three-quarters in the early neonatal period, i.e., 0 to 6 days [2,3]. In Bangladesh, overall neonatal mortality has decreased from 55 per 1,000 live births in 1990 [4] to 23 per 1,000 live births in 2015 [1], with an average yearly reduction of 1 death per 1,000. However, this reduction is 3 times slower than the reduction of under-5 mortality over the same period, and neonatal deaths now account for 62% of all under-5 deaths [1], whereas early-neonatal deaths (19.3/1,000) contribute to over 65% of all neonatal deaths.
Sepsis is considered the final common pathway to neonatal death due to severe illnesses and various invasive infections [5]. Neonatal sepsis is one of the leading causes of neonatal deaths in developing countries and is responsible for 13% of deaths in the neonatal period and 42% of deaths in the first 7 days of life [6]. A quarter of the neonatal deaths in Bangladesh are a result of severe illnesses, including, sepsis, tetanus, and diarrhea [1,3]. Despite being highly preventable, severe illnesses including sepsis is a predominant cause of death among newborns globally [7,8] and in Bangladesh [9].
WHO recommends early initiation of breastfeeding within 1 hour of birth as the first step toward ensuring optimal breastfeeding [10]. Literature [11,12] suggests that successful early initiation of breastfeeding facilitates sustained optimal breastfeeding practices throughout infancy. Yet only about 2 in every 5 newborns worldwide and 40% of newborns in South Asia begin breastfeeding within the first hour of life [10,12,13]. In Bangladesh, although there have been some improvements in other breastfeeding practices, including exclusivity and duration of breastfeeding, 49% of children are still initiating breastfeeding after the first hour of birth [14,15].
Recent systematic reviews [16,17] and global reports by WHO [13,18] have revealed that the risk of neonatal mortality increases by about 33% with the delay in initiation between 2 to 23 hours compared with those who initiated within the first hour of birth. Studies from Nepal, Ghana, and India [11,19,20] have reported a protective effect of early initiation of breastfeeding with a 44% reduction in the risk of death among babies surviving the first 48 hours and a 42% reduction in the risk of death among low birth weight (LBW) babies. Timely initiation of breastfeeding can be crucial in reducing infection associated with neonatal mortality [16,20] by enhancing the newborns’ immune response to infectious pathogens [21,22]. It is still not confirmed whether early initiation of breastfeeding leading to a lower incidence of severe illnesses leading to suspected sepsis is the most likely pathway to reduce neonatal deaths, especially in the first 7 days of life. There exists a gap in the evidence explaining the mechanism by which delayed initiation of breastfeeding may impact mortality in resource-limited settings of developing countries. Early initiation ensures that the infant receives colostrum, which is rich in several immunoglobulins and nutritional factors protecting the newborn from short- and long-term illnesses and death [20,23].
Although several articles report the importance of exclusive breastfeeding and duration of breastfeeding in infancy [24–26], we found no study that looked at the influence of the time of initiation of breastfeeding on newborn danger signs with an emphasis on severe illnesses in the early newborn period, i.e., in the first 7 days. In this paper, we aim to explore the effect of timing of breastfeeding initiation on severe illnesses and newborn danger signs among newborns in the first 7 days after birth in rural Bangladesh.
We collected data for this analysis from a large community-based randomized controlled trial of the effect of iron folic-acid supplementation started early (≤12 weeks) and sustained throughout pregnancy on neonatal mortality. Details of the methods and design of the trial are published as a study protocol by Huda and colleagues [27]. We conducted the trial between 2013 and 2015 in the rural areas of 5 districts in Bangladesh. All women aged between 15 and 49 years at the time of enrollment are permanent residents of the study area and became pregnant during the enrolment period were recruited as study participants. These women were followed throughout pregnancy until 6 weeks after pregnancy outcome. Pregnant women were identified by study personnel from active pregnancy surveillance conducted by BRAC (an international nongovernment organization) field workers (Shyastho Sebikas) to identify women with menses delayed more than 45 days from the first date of their last menstrual cycle. Study personnel then enrolled consenting women following confirmation of pregnancy using a “pregnancy confirmatory dipstick test.” The analysis for this paper was planned at the time the trial was set up, and the questions required for this analysis was designed before the trial was implemented. We registered the trial with the Australia New Zealand Clinical Trials Registry (ANZCTR). The registration ID is ACTRN12612000588897.
Enrollment, follow-up, and data collection was identical throughout the study area. We collected follow-up information from each woman until 42 days after pregnancy outcome. Baseline data including demographic and socioeconomic indicators, intake of nutritional supplements (prior to enrollment and during the study period), parental education, birth history, and maternal health condition during pregnancy. Incentives were given to enrolled women for notifying the research team of the birth within 24 hours of delivery. Study personnel collected information at 48 hours after birth on essential newborn care practices and at the 7 to 10 day follow-up visit on day-to-day neonatal danger signs from birth until the seventh day after birth.
On the first postbirth visit (at 48 hours of birth), mothers were asked if they had ever breastfed their newborn, and if they did, how much time elapsed between birth and initiation of breastfeeding. In case a live-born infant died before the first postbirth follow-up visit, we still collected information on breastfeeding initiation for the child before their death. We also measured the birth weight of the child at this visit using a portable digital scale and infant weighing pouch (WeiHeng WH-A08). At the 7 to 10 day follow-up visit, we asked the mothers to give a day-by-day account of the onset, continuation, and cessation of each of the danger signs during the first 7 days. For infants who died within 7 days of birth, we collected the danger signs before death and included them in the analysis. We estimated the gestational age of the infants in weeks, from the last menstrual period reported by mother at enrollment until the reported date of birth outcome.
We included in the analysis live-born infants with known breastfeeding initiation status, including those who were never breastfed. Breastfeeding initiation was the time in hours after birth, when the newborn was first put to the mother’s breast. To estimate the extent of adherence and nonadherence to WHO recommendation of initiation within an hour of birth, we categorized the variable as within 1 hour, ≥1 hour to <24 hours, ≥24 hours to <48 hours, ≥48 hours, and never breastfed. We excluded women whose breastfeeding status was unknown or missing.
The outcome variable severe illness was defined using newborn danger signs reported in The Young Infants Clinical Signs Study Group [28,29] and the Bangladesh Neonatal Health Strategy [30]. We identified severe illness in newborns if their mother reported the presence of any 1 of the following 6 signs and symptoms: unusually cold/clammy skin, high body temperature, unconscious/no movement or lethargic, caregivers report of convulsions, rapid breathing or difficulty in breathing, and unable to breastfeed. We did not have information on the seventh sign used in previous studies, namely, severe chest indrawing. Use of this definition for severe illness is recommended for studies in low resource settings, in which suspected sepsis cannot be confirmed with the usual clinical parameters and cut-offs [31,32]. We present severe illness as a binary outcome variable indicating presence or absence of the outcome of interest.
Potential confounders were classified as infant, maternal, and household characteristics. Infant characteristics included the sex of the child (male and female); the birth weight [≥2,500 g, 2000–2499 g (LBW), and ≤2000 g (LBW)]; the birth outcome [multiple and single]; colostrum (given and not given); appropriate care of the umbilical cord; application of material after cutting the umbilical cord; timing of the first bath after birth (within and after 72 hours); and timing of drying after birth (within and after 5 minutes). Maternal characteristics included age (<20 years and ≥20 years); gestational age at birth (<28 weeks, 28–33 weeks, 34–36 weeks, and ≥37 weeks); education (no education, primary completed, secondary completed, and higher); parity (primiparous and multiparous); stillbirth/miscarriage of previous child; prolonged labor during childbirth; and fever during childbirth. In addition, we constructed a combined variable for type, place, and skilled attendance at delivery (normal vaginal delivery [NVD] at facility by skilled attendant, NVD at home by unskilled attendant, NVD at home by skilled attendant, and caesarean section (CS) at facility by skilled attendant). Household characteristics included the wealth quintiles. We used standard demographic health survey methods to form a list of household assets to construct the wealth index [33]. Most of the selected potential confounders were reported in previous studies that explored the determinants of the timing of initiation of breastfeeding [11,19,20,22,23].
We used frequency distributions to describe the baseline characteristics of the study population and summarized all data on household, maternal, and infant characteristics using descriptive statistics of proportions for categorical variables. All descriptive statistics are presented by the early onset of newborn danger signs for severe illnesses in the newborns. The mean time to initiate breastfeeding is calculated among children for whom we have a record of their breastfeeding initiation time.
To examine the effect of time of initiation of breastfeeding on the primary outcome of severe illnesses in the early neonatal period, we used a multivariable mixed-effects logistic regression model, adjusting for infant, maternal, and household characteristics that showed an independent association (p < 0.20) with the outcome in an univariable model. The variance inflation factor was used to assess collinearity between covariates included in the multivariable model. We examined the outcome variable and found it varied by cluster (data not shown) indicating the need to use multilevel mixed-effects models. All regression models were adjusted for clustering of participants from 100 clusters (from a cluster randomized control trial [RCT]) in 5 districts using multilevel random effects, with clusters nested within district [34]. The variable for treatment arms had been included in all regression models in order to account for the effects of the intervention. We present the odds ratios (ORs) and 95% CIs for severe illness for each category of time to initiate breastfeeding.
To examine the probability of potential reverse causality due to maternal infections and complications at childbirth, we performed a sensitivity analysis excluding the infants whose mother reported prolonged labor or fever around the time of childbirth. We then fitted a restricted version of the adjusted model (restricted model 1) for the same covariates. Further, to explore the probability of reverse causality due to the child being too ill to initiate breastfeeding early, we performed another sensitivity analysis, excluding the children who died in the first 48 hours of birth. We fitted this restricted model (restricted model 2), adjusting for the same covariates of the adjusted model. Finally, we compared the adjusted OR (aOR) of all 3 models to identify any possible influence of reverse causality due to maternal complications during the time of delivery or severe illness of the child immediately after birth.
The population attributable fraction (PAF) of severe illnesses among newborns whose breastfeeding was initiated 1 hour, 24 hours, or 48 hours of birth were calculated using the formula below [35,36]:
PAF=Pb(OR-1OR).
Here, Pb is the proportion of children with breastfeeding initiated after 1 hour, 24 hours, or 48 hours of birth, and OR is the OR of initiating breastfeeding after 1 hour, 24 hours, and 48 hours of birth. For ease of interpretation, we will present the PAF as population attributable risk percent (PAR%) by multiplying PAF by 100. We calculated the crude PAR% using the ORs from the univariable association. To measure the adjusted contribution of delayed initiation of breastfeeding on the total risk of having severe illness, the adjusted PAR% was calculated using the aORs from both the final adjusted model and restricted model. We estimated 95% CI for all PAR%. We performed all statistical analyses using STATA version 15 (Stata Corporation, College Station, TX).
The Ethical Review Committee (ERC) of the International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b), and the Human Research Ethics Committee (HREC) of the University of Sydney have granted ethics approval for the parent study. We obtained written informed consent from pregnant women during enrollment into the study, which provided full disclosure regarding the study.
There were 30,646 live-born infants during the study period, and time to initiation of breastfeeding was known for 29,873 (97.5%) infants. The remaining 773 women had no information on their breastfeeding status, and we excluded them from the analysis. Two-thirds of the newborns (66.7%) with known breastfeeding initiation status had initiated breastfeeding within 1 hour of birth. The mean time to initiate breastfeeding, among children with a record of their breastfeeding initiation time was, 3.8 hours (SD 16.6 hours). By the end of 48 hours, all but 5.7% of infants were breastfed. Severe illness in the neonatal period early (0–6 days) was reported for 4,437 (14.9%; Fig 1) infants of whom 338 (7.6%) subsequently died within the first 7 days. Among the severe illness cases, hyperthermia was the most common danger sign reported by 36.6% mothers, followed by respiratory distress reported by 37.4% mothers. Convulsion was the least reported danger sign and was reported by 7.8% mothers. Fig 2 shows the percentages of newborns who had any of the danger signs in the first 7 days after birth. Table 1 summarizes some infant, maternal, and household characteristics of the newborns by their time of breastfeeding initiation.
Table 2 presents the unadjusted ORs and aORs for the association between timing of initiation of breastfeeding and severe illness in early neonatal stage using univariable and multivariable models from the unrestricted and restricted data. There is a dose response of higher likelihood of severe illness with an increased delay in breastfeeding initiation. Infants who initiated breastfeeding between 1 to 23 hours of birth had significantly higher odds (unadjusted OR 1.45, 95% CI 1.33–1.58) of having signs of severe illness compared with children who initiated breastfeeding within 1 hour of birth. This effect was further seen to have increased for children who initiated between 24 to 47 hours (unadjusted OR 2.94, 95% CI 2.42–3.58) and after 48 hours (unadjusted OR 4.96, 95% CI 4.29–5.73).
When compared with newborns with early initiation of breastfeeding, the aORs for severe illness (Table 2) remained significantly higher among all late breastfeeding initiator groups (1–23 hours: aOR 1.37, 95% CI 1.25–1.50; 24–47 hours: aOR 2.85, 95% CI 2.31–3.52; and 48 hours or more: aOR 4.13, 95% CI 3.48–4.89). The highest risk for severe illness and early newborn danger signs was for those who were unable to initiate breastfeeding (aOR 4.77, 95% CI 3.52–6.47) within the 7 days. The multivariable models were adjusted for confounders that had univariable p < 0.2 (S1 Table). All variance inflation factors were <5, indicating that collinearity was not an issue for the multivariable models.
In the adjusted model using unrestricted data (S1 Table), the odds of having severe illness was lower for mothers who delivered their child by unskilled birth attendants at home (aOR 0.75, 95% CI 0.67–0.84), compared with women who delivered at a facility with skilled attendants. Performing CS at a facility by skilled attendants was associated with a reduced risk of early newborn danger signs (aOR 0.60, 95% CI 0.52–0.68). Infants born with a birthweight of ≥2,500g and who received appropriate thermal and cord care at birth had significantly (p<0.0001) lower odds of having severe illnesses than those who had LBW and inappropriate postbirth care. Infants who were first born, whose mother’s preceding pregnancy ended in stillbirth or miscarriage, and whose mother has prolonged labor during childbirth had significantly higher odds of having severe illnesses in the first 7 days after birth.
Sensitivity analysis of the adjusted association after excluding the children whose mother had prolonged labor or fever at childbirth (Table 2) resulted in slightly higher odds of severe illness among all the breastfeeding initiation categories. The aORs of having early newborn danger signs and severe illness among late initiators slightly increased for children of women who did not have prolonged labor or fever during childbirth. About one-fifth (18.5%) of the 19,914 women who initiated within the first hour had prolonged labor. The increased likelihood of early newborn danger signs in the reference group leads to an overall increase in odds of severe illnesses among all late initiators. Excluding children who died in the first 48 hours (Table 2) had little to no effect on the adjusted risk of early newborn danger signs and severe illnesses among late initiators compared with the final adjusted model. The unadjusted and adjusted associations between severe illnesses in the early neonatal stage and a range of confounders used to generate the models using unrestricted and restricted data are shown in S1 Table. The association between breastfeeding initiation time and severe illness were similar when the regression model was adjusted for all covariates as continuous variables (S2 Table).
Table 3 shows that the PAR% decreases with increased delay in breastfeeding initiation. After adjusting for confounders, about 16% of the early newborn danger signs and severe illness could have been reduced if all newborns initiated breastfeeding within the first hour of birth. The PAR% for initiating breastfeeding after 24 hours indicates that 10.6% of the total risk of early newborn danger signs and severe illnesses were attributable to initiating breastfeeding after the first day of birth.
We found that early initiation of breastfeeding within an hour of birth was significantly associated with a reduced risk of early newborn (0–6 days) danger signs and severe illnesses in rural Bangladesh. We detected a dose-response relationship of increasing odds of severe illnesses in the early newborn stage with greater delay in initiation of breastfeeding after adjusting for confounders. Our findings indicate that the highest risk of presenting with early newborn danger signs following delayed initiation of breastfeeding beyond the first hour was with newborns who were first born, had very LBW (<2,000 g), did not receive appropriate cord and thermal care, were born through NVD at home with unskilled attendance, had a history of stillbirth/miscarriage of a previous sibling, or whose mother had prolonged labor during childbirth. We also found that one-sixth of the reported danger signs in the early neonatal period could be reduced if children who initiated beyond the first hour could initiate within the recommended time. Excluding children whose mother had maternal complications and those who died within 48 hours had a minimum to no effect on the fraction of severe illness cases that could be prevented. The findings from this study highlight the need to encourage women and caregivers to facilitate early initiation of breastfeeding to reduce early newborn danger signs and severe illnesses, especially among high-risk newborns.
Our study is one of the few studies that specifically explores the effect of delayed breastfeeding initiation on severe illness during the early newborn stage (0–6 days after birth), the most crucial stage of the neonatal period [37]. Most studies on the health effects of early initiation of breastfeeding have examined all-cause neonatal mortality [11,19,20,23] and infectious diseases related to neonatal mortality such as diarrhea and respiratory infections [16,38] during the entire neonatal period. In this study, information on the timing of initiation and danger signs was collected within 7 to 10 days of birth, thus reducing the recall period of reporting both the exposure and outcome. As the study sample originates from a large community-based randomized controlled trial, we were able to collect data from a large number of newborns. Sensitivity analyses excluding infants who died within 48 hours of birth, as done in several previous studies looking at mortality outcomes [11,19,20,23], resulted in no change in the risk of severe illnesses in relation to the time of initiation of breastfeeding. Even though our analysis does not provide direct evidence of the absence of reverse causation, it is very unlikely that the reported danger signs were due to the newborn being too sick to initiate breastfeeding within 1 hour of birth. Further sensitivity analysis indicates that maternal complications and signs of maternal infections around the time of childbirth may have contributed to a higher risk of severe illnesses among the newborns in all categories of timing of initiation of breastfeeding, including those who initiated within an hour of birth. The study included all newborns from a defined area of rural Bangladesh over a 12-month recruitment period and thus increased the generalizability of the study results. We present the dose-response relationship of early initiation of breastfeeding associated with a reduced risk of severe illnesses in the early neonatal period after adjusting for all possible confounders known to be associated with the timing of initiation of breastfeeding and early newborn danger signs.
Our study has several limitations. Firstly, the danger signs for severe illnesses used in this study includes 6 out of the 7 signs and symptoms proposed in the Young Infants Clinical Signs Study [28], but we did not collect the seventh danger sign, severe chest indrawing. A study in Bangladesh [39] found that severe chest indrawing has very low specificity and thus contributes to overestimating the proportion of children with severe illness requiring hospital admission. Thus, omitting this symptom in the outcome in our study will not lead to an overestimate of the effect of breastfeeding on newborn danger signs. In addition, there exists a potential for misclassification of the outcome considering that severe illness was defined using the mother’s report of episodes of danger signs, and there were no confirmatory tests using biomarkers to classify the severity. Secondly, mothers of dead infants may have recalled danger signs and breastfeeding initiation times differently compared with the mothers of surviving newborns, thus incorporating a potential recall bias. The short recall period of 7 to 10 days is likely to have minimized any recall bias but not eliminated it. However, there was no change in risk of presenting with newborn danger signs following delayed initiation of breastfeeding when the model was applied to children who survived 48 hours from birth. This implies that any recall bias was minimal and provides evidence that our results were not the result of reverse causality. Considering the short recall period for all mothers, the generic approach to defining severe illnesses, and the lack of evidence of reverse causality our findings provide strong evidence that delayed initiation of breastfeeding increases the risks of early newborn danger signs and severe illness.
Timely initiation of breastfeeding is the first step toward ensuring exclusive breastfeeding and sustaining optimal breastfeeding practices [40]. Furthermore, early initiation and early attachment to the mother’s breast promotes thermal care of the infant and can reduce the risk of hypothermia immediately after birth [11]. Delayed initiation of breastfeeding has a strong biological plausibility of leading to severe illnesses, considering the important role of breastfeeding in enhancing immune functions during the early stages of life [10,41]. Delayed initiation is known to be further compounded with an early introduction of prelacteal feeding [23,40], which is known to have a detrimental effect on the immune system and has the potential to lead to infection and suspected sepsis [42]. Early exposure to breastmilk ensures intake of colostrum, which contains a high concentration of lactoferrin, immunoglobulin A (IgA), leukocytes, and specific developmental factors [43,44]. The amount of these proteins and immunoglobulin (especially IgA) [45] is significantly higher in colostrum than in the transitional milk (6 to 14 days of lactation), and mature milk (after the 15th day of lactation) [46]. Colostrum intake accelerates intestinal maturation, increases the integrity of the epithelial layer, promotes epithelial recovery from infections, has anti-inflammatory and antimicrobial potential, and decreases the risk of microbial translocation [41,47].
Findings in this study are consistent with previous studies that looked at the relationship between early initiation of breastfeeding and all-cause mortality [11,19,20,23] and morbidity [16,38]. Studies in Ghana [11], India [20], Nepal [19], Egypt [22], and Tanzania [47] report a similar dose-response relationship of early initiation of breastfeeding on survival. The rate of initiation of breastfeeding within the first hour is higher in this study than that reported in studies from Nepal (3.4%) [19] and Ghana (43%) [11] and lower than reported in studies from Tanzania (87%) [47]. The study from India [20] looked at the change in risk of adverse outcomes associated with initiation within the first day and later, rather than the association for initiation within the first hour. The rates of LBW reported in studies from Nepal (29.8%) [19] and India (22%) [20] were similar to our study and both present the higher risk of delayed initiation of breastfeeding associated with adverse outcomes among LBW infants (<2,500 g).
In this study, we found that prolonged labor and fever during childbirth contributed to a higher risk of severe illnesses in all breastfeeding initiation groups, especially in the reference group of those who initiated within 1 hour of birth, because two-thirds of the children fall in this group. A higher proportion of mother-infant dyads who never initiated breastfeeding reported having prolonged labor compared with newborns who initiated within an hour of birth (35.7% versus 18.5%). Previous studies [48–50] support these findings and provide evidence that neonatal infection in the first week of life is associated with maternal infection and intrapartum fever. Another study suggests that women who had delivery complications were less likely to initiate breastfeeding on time [51].
We found that women who had caesarean delivery at a facility by skilled attendants had a lower likelihood of presenting with early newborn danger signs in association with delayed breastfeeding initiation compared with those who had a NVD. Previous studies, mostly in high-income countries, compared NVD at a facility with caesarean delivery at a facility and showed more risk of infections in the latter group [40,52–54]. Findings in our study may have resulted from the high percentage (70%) of all births taking place at home, mostly in unhygienic and infection-prone surroundings with limited postnatal support. Hence, the adjusted association showed that babies born through caesarean delivery by skilled attendance at a facility had a lower likelihood of severe illness compared with NVD at a facility.
The findings in this study indicate that the timing of initiation of breastfeeding, especially within the first hour of birth, is significantly associated with reduced risk of severe illness in the early neonatal period. Hence the findings support the need to reinforce the implementation of existing recommendations from WHO and UNICEF [13] to “place the child in skin-to-skin contact with mothers immediately following birth for at least an hour and encourage the mother to initiate breastfeeding within this time,” as the Step 4 of the 10 steps towards successful breastfeeding [16,55]. By improving early initiation of breastfeeding, newborn morbidity and mortality would likely decrease and could contribute to achieving Sustainable Development Goals (SDG) 2030 for reducing child mortality.
Our findings indicate the need for setting a future research agenda around developing comprehensive community-based interventions to promote initiation of breastfeeding within the first hour of birth in low- and middle-income countries to reduce early newborn danger signs and severe illness and thus reduce neonatal deaths. It is crucial for healthcare providers and birth attendants to encourage and support timely initiation regardless of the mode of delivery and birth weight of the newborn [54,56]. Interventions to promote and support early breastfeeding initiation could be designed based on the place and type of delivery and requires involvement of all role players around the time of childbirth. Such role players may include family members and birth attendants (skilled or unskilled) for home deliveries, healthcare providers for NVD at a facility, and a professionally endorsed standard guideline for caesarean delivery. Interventions should be tailored to suit the needs of populations in which the rate of LBW is high and postnatal care is limited with emphasis on first born infants and mothers who experienced pregnancy complications.
|
10.1371/journal.ppat.1006309 | Neutrophils and Ly6Chi monocytes collaborate in generating an optimal cytokine response that protects against pulmonary Legionella pneumophila infection | Early responses mounted by both tissue-resident and recruited innate immune cells are essential for host defense against bacterial pathogens. In particular, both neutrophils and Ly6Chi monocytes are rapidly recruited to sites of infection. While neutrophils and monocytes produce bactericidal molecules, such as reactive nitrogen and oxygen species, both cell types are also capable of synthesizing overlapping sets of cytokines important for host defense. Whether neutrophils and monocytes perform redundant or non-redundant functions in the generation of anti-microbial cytokine responses remains elusive. Here, we sought to define the contributions of neutrophils and Ly6Chi monocytes to cytokine production and host defense during pulmonary infection with Legionella pneumophila, responsible for the severe pneumonia Legionnaires’ disease. We found that both neutrophils and monocytes are critical for host defense against L. pneumophila. Both monocytes and neutrophils contribute to maximal IL-12 and IFNγ responses, and monocytes are also required for TNF production. Moreover, natural killer (NK) cells, NKT cells, and γδ T cells are sources of IFNγ, and monocytes direct IFNγ production by these cell types. Thus, neutrophils and monocytes cooperate in eliciting an optimal cytokine response that promotes effective control of bacterial infection.
| The innate immune system is one of the first lines of defense against invading microorganisms. Many innate immune cell types, including neutrophils and Ly6Chi monocytes, are rapidly recruited to sites of infection. Neutrophils and monocytes are thought to engage both overlapping and distinct effector mechanisms for antimicrobial clearance. Neutrophils are generally viewed as playing a direct bactericidal role, whereas monocytes are thought to be bactericidal, but are also considered to be the major producers of important cytokines that prime or activate subsequent responses. However, neutrophils are also capable of making many of the same cytokines produced by monocytes, but whether neutrophils and monocytes contribute to cytokine responses in a redundant or non-redundant manner remains unclear. Here, we demonstrate that neutrophils and Ly6Chi monocytes contribute to the production of cytokines required for host defense against Legionella pneumophila, which causes the severe pneumonia Legionnaires’ disease. Both neutrophils and monocytes contribute to maximal production of the cytokines IL-12 and IFNγ, and monocytes are also required for production of the cytokine TNF. Finally, we show that monocytes are required for optimal IFNγ production by innate and innate-like lymphocytes, allowing for increased control of L. pneumophila infection. Our work demonstrates that neutrophils and monocytes cooperate in shaping a protective cytokine response that ensures successful host defense.
| The innate immune system is essential for host defense against bacterial pathogens [1,2]. Many critical innate immune functions are carried out by a multitude of cell types, including macrophages, dendritic cells (DCs), neutrophils, and Ly6Chi monocytes and their derivative cells [3]. Some of these cell types are tissue-resident, such as alveolar macrophages in the lung [4]. In contrast, Ly6Chi monocytes and neutrophils exist in low numbers in the periphery during homeostasis, but are rapidly mobilized from the bone marrow and recruited to tissues early during infection [5,6]. The primary role for neutrophils in antibacterial defense is thought to involve direct bacterial killing by means of reactive oxygen species and microbicidal molecules present within granules [6–9], as well as production of neutrophil extracellular traps (NETs) [10]. Conversely, other myeloid cells, such as macrophages, DCs, and monocytes, also synthesize bactericidal molecules, but are predominantly thought to be major producers of proinflammatory cytokines, such as tumor necrosis factor (TNF), interleukin-1β (IL-1β), and IL-12 [3,5]. These cytokines orchestrate anti-bacterial effector responses that are critical for bacterial clearance. For example, Ly6Chi monocytes control bacterial burdens during Listeria monocytogenes, Klebsiella pneumoniae, and Mycobacterium tuberculosis infection [11–14], in large part because they are an important source of IL-1β, IL-12, and IL-18 during infection and can also differentiate into DCs that produce high levels of TNF [11]. Interestingly, neutrophils can also produce TNF, IL-1β, IL-12, IL-18, IFNγ, and other cytokines in response to several bacterial and parasitic infections [15–22]. Although neutrophils and Ly6Chi monocytes produce overlapping repertoires of inflammatory cytokines, it is currently unclear whether these cell types make redundant or distinct contributions to protective anti-microbial cytokine responses.
We sought to address this question in the context of pulmonary infection with the gram-negative pathogen Legionella pneumophila, responsible for the severe pneumonia Legionnaires’ disease [23,24]. L. pneumophila is a pathogen of freshwater amoebae and gains access to the human lung through inhalation of contaminated water aerosols [25–27]. Following uptake by alveolar macrophages, L. pneumophila replicates within these cells by deploying a Dot/Icm type IV secretion system that translocates a large repertoire of bacterial effectors that manipulate host membrane trafficking and other eukaryotic processes [28–32]. A subset of translocated effectors that block host translation elongation in combination with a host response to infection leads to a potent inhibition of global protein synthesis in infected macrophages [33–37]. Thus, infected macrophages are incapable of producing key cytokines, including TNF and IL-12, which are essential for host defense [38–41]. However, infected macrophages still synthesize and secrete the cytokines IL-1α and IL-1β [38,39], which orchestrate neutrophil recruitment to the lung [36,42–44], as well as the production of TNF and other cytokines by uninfected bystander neutrophils, Ly6Chi monocytes, and DCs [38].
Although neutrophils and Ly6Chi monocytes comprise the largest number of cytokine-producing cells and produce overlapping sets of cytokines during L. pneumophila infection [38,45], it is poorly understood whether these cell types make redundant or distinct contributions to the overall cytokine response. This is in part because the role of Ly6Chi monocytes in cytokine production and host defense during L. pneumophila infection has been unknown. As for neutrophils, anti-Gr-1 antibody-mediated depletion suggested these cells were required for maximal IL-12 production during pulmonary L. pneumophila infection [46]. During an intravenous model of L. pneumophila infection, anti-Gr-1 antibody-based depletion suggested that neutrophils were required for IL-12 and IL-18 production and subsequent IFNγ production by natural killer (NK) cells [22]. However, the anti-Gr-1 antibodies used in these previous studies recognize an epitope common to Ly6G expressed on neutrophils and Ly6C expressed on monocytes, and anti-Gr-1 antibodies can deplete both neutrophils and Ly6Chi monocytes [12,47–49], raising the question of whether Ly6Chi monocytes also contribute to some of the phenotypes attributed to neutrophils. Notably, a number of mouse models of L. pneumophila infection in which neutrophil recruitment is impaired due to loss of chemokine or cytokine receptors (CXCR2 or IL-1R) [36,42–44,50] demonstrate elevated bacterial burdens, but these models can also impact recruitment or activation of other cell types [38,51].
Here, we utilized a number of complementary approaches to interrogate the relative contributions of Ly6Chi monocytes and neutrophils to cytokine production and control of pulmonary L. pneumophila infection. Our data indicate that animals lacking the chemokine receptor CCR2, which is required for Ly6Chi monocytes to egress from the bone marrow, exhibited a defect in both TNF and IL-12 production and monocyte-derived DC recruitment to the lung. We further found that Ly6Chi monocytes and DCs produce and serve as critical sources of IL-12. Our data also demonstrate that while neutrophils are dispensable for maximal production of most inflammatory cytokines during L. pneumophila infection, they contribute to maximal IL-12 production. Intriguingly, depletion of either neutrophils or Ly6Chi monocytes resulted in defective control of bacterial burdens. Furthermore, both neutrophils and Ly6Chi monocytes were required for maximal production of IFNγ. NK cells and other innate-like lymphocytes in the lung served as sources of IFNγ, and monocyte-derived IL-12 directed IFNγ production by these cell types. Overall, these findings indicate important roles for both neutrophils and Ly6Chi inflammatory monocytes in shaping key cytokine responses that orchestrate protective immune responses during pulmonary bacterial infection.
Mice treated with anti-Gr-1 antibodies to deplete neutrophils have a defect in bacterial clearance and IL-12 production [46], but the extent to which the anti-Gr-1 antibody might also deplete Ly6Chi monocytes during pulmonary L. pneumophila infection has not previously been assessed. We therefore assayed the numbers of neutrophils and Ly6Chi monocytes in WT C57BL/6 (B6) mice treated with the anti-Gr-1 antibody clone RB6-8C5. The anti-Gr-1 antibody efficiently depleted both neutrophils and Ly6Chi monocytes in the lungs of naïve, uninfected mice compared to mice injected with the isotype control antibody (ISO) (S1 Fig). We then infected anti-Gr-1-treated B6 mice with ΔflaA L. pneumophila, which lacks flagellin, as this is a permissive model of infection that allows for bacterial replication in B6 mice encoding a functional NAIP5 allele [52,53]. When compared to mice injected with isotype control antibody, mice given the anti-Gr-1 antibody had significantly lower numbers of neutrophils in the lung at both 24 and 72 hours post-infection (Fig 1A and 1B). Additionally, anti-Gr-1 antibody treatment did not affect the total numbers of Ly6Chi monocytes at 24 hours post-infection, but there were significantly lower numbers of Ly6Chi monocytes at 72 hours post-infection (Fig 1C), similar to previous observations made during L. monocytogenes or Toxoplasma gondii infection, in which both neutrophils and Ly6Chi monocytes were significantly depleted following anti-Gr-1 treatment [12,48]. In contrast, the total numbers of dendritic cells (DCs) did not decrease following anti-Gr-1 treatment (Fig 1D). Critically, mice that received the anti-Gr-1 antibody during infection displayed significantly higher L. pneumophila colony-forming units (CFUs) by 24 hours post-infection and a two-log increase in CFUs at 72 hours post-infection compared to isotype-treated mice (Fig 1E).
We also measured cytokine levels in the bronchoalveolar lavage fluid (BALF) during anti-Gr-1-mediated depletion at 24 and 72 hours post-infection. We found that levels of IL-1α, IL-1β, IL-4, IL-6, IL-10, and IL-18 were unchanged or even elevated, likely reflecting increased cytokine production by other immune cell types in response to the substantial increase in bacterial load (Fig 2A–2C, 2E, 2H and 2I). In contrast, levels of TNF, the IL-12p70 heterodimer, and the IL-12p40 subunit were significantly reduced in the BALF at 24 hours post-infection in anti-Gr-1-treated mice (Fig 2D, 2F and 2G), in agreement with a previous study using anti-Gr-1-mediated depletion during pulmonary L. pneumophila infection [46]. The reduction in TNF and IL-12 is particularly notable given the increase in bacterial burden in these mice at this time (Fig 1E). Therefore, the anti-Gr-1 antibody depletes both neutrophils and monocytes, and either one or both of these cell types are critical for production of TNF, IL-12, and control of bacterial loads during pulmonary infection with L. pneumophila.
To interrogate the specific role of neutrophils in host protection against L. pneumophila, we employed the more selective anti-Ly6G antibody clone 1A8, which specifically depletes neutrophils and does not target monocytes [49]. The total numbers of neutrophils in the lung at both 24 and 72 hours post-infection were significantly reduced in both infected and uninfected mice treated with anti-Ly6G antibody, although the reduction was not as efficient as with anti-Gr-1 antibody (Fig 3A and 3B, and S1 Fig). The numbers of Ly6Chi monocytes in the lungs of both uninfected and infected mice were unaffected by anti-Ly6G treatment, as expected (Fig 3C and S1 Fig). Importantly, this highly specific, though less robust, depletion of neutrophils resulted in significantly higher bacterial burdens in the lungs of infected mice, with a nearly one-log increase in CFUs at 72 hours post-infection (Fig 3D). Thus, our anti-Ly6G-mediated depletions demonstrate a specific role for neutrophils in host protection during pulmonary infection with L. pneumophila.
The exact mechanisms underlying the contribution of neutrophils to host defense against L. pneumophila remain unclear. Neutrophils have been shown to deploy multiple effector mechanisms, such as direct microbial killing or cytokine production, during L. pneumophila infection and other infection models [6,9,10,15–21,38,45,46]. Neutrophils can produce protective cytokines, such as TNF, IL-1α, and IL-12, during pulmonary L. pneumophila infection [21,38,45,46], but the precise contribution of neutrophil-derived cytokines to the overall cytokine response is unclear. To address this question, we also measured cytokine levels in the BALF during anti-Ly6G-mediated depletion at 24 and 72 hours post-infection. Specific depletion of neutrophils with anti-Ly6G antibody did not reduce levels of other cytokines in the BALF (Fig 4A–4F), but resulted in significantly reduced levels of IL-12p40 at 24 hours post-infection (Fig 4G). Altogether, our findings delineate a specific role for neutrophils in controlling bacterial loads and early production of IL-12 during L. pneumophila lung infection [41].
In addition to neutrophils, a large population of Ly6Chi monocytes is also rapidly recruited into the lung during L. pneumophila infection [38]. Notably, while Ly6Chi monocytes are not productively infected by L. pneumophila [21], they produce important proinflammatory cytokines, including TNF [38,45], which is required for successful control of infection [40]. We therefore sought to examine the contribution of Ly6Chi monocytes to host defense against L. pneumophila. To do so, we used mice deficient for the chemokine receptor CCR2 (Ccr2-/-), because Ccr2-/- monocytes exhibit a defect in the ability to emigrate from the bone marrow to sites of inflammation [54,55]. Indeed, following L. pneumophila infection, Ccr2-/- mice had a significant defect in recruiting Ly6Chi monocytes to the lung at 24 and 48 hours post-infection compared to B6 mice (Fig 5A and 5B). Importantly, we saw no defect in neutrophil recruitment (Fig 5C). Though Ccr2-/- mice showed no reduction in total DC numbers in the lung at 24 hours post-infection, DCs were significantly reduced in Ccr2-/- mice at 48 hours post-infection (Fig 5D), consistent with previous findings that Ly6Chi monocytes differentiate into DCs at sites of inflammation [56]. Critically, Ccr2-/- mice had significantly higher bacterial burdens in the lung at 48 and 96 hours post-infection (Fig 5E), demonstrating that Ly6Chi monocytes play a key non-redundant role in host defense against L. pneumophila.
Ly6Chi monocytes comprise a major fraction of TNF-producing cells during pulmonary infection with L. pneumophila [38,45]. Therefore, we next asked if Ly6Chi monocytes significantly contribute to the cytokine milieu during infection. We examined the levels of proinflammatory cytokines that are normally produced and secreted into the airway space during infection to determine if these cytokines are modulated in the absence of Ly6Chi monocytes at 24 and 48 hours post-infection. The levels of IL-1α, IL-1β, IL-6, and IL-18 were unchanged or increased in Ccr2-/- mice during infection, likely reflecting increased cytokine production by other immune cell types in response to the substantial increase in bacterial load (Fig 6A–6C and 6E). However, TNF levels were significantly reduced at both 24 and 48 hours post-infection in the absence of Ly6Chi monocytes (Fig 6D). This finding indicates that Ly6Chi monocytes are indeed a major source of TNF during pulmonary infection with L. pneumophila, consistent with previous findings showing that Ly6Chi monocytes comprise the majority of TNF-producing cells during infection [38,45]. In addition to TNF, Ccr2-/- mice also had a significant defect in IL-12p70 and IL-12p40 production at both 24 and 48 hours post-infection (Fig 6F and 6G), in agreement with recent findings [57]. The defect in IL-12 production exhibited by Ccr2-/- mice was greater than the defect we observed in anti-Ly6G-depleted mice (Fig 4G), indicating that monocytes are more critical than neutrophils for IL-12 production. Altogether, these data demonstrate that Ly6Chi monocytes are required for maximal production of the protective cytokines TNF and IL-12 during pulmonary L. pneumophila infection.
Thus far, our data suggest that both Ly6Chi monocytes and neutrophils are non-redundantly required for IL-12 production during infection. This contrasted with most of the other cytokines that we examined, suggesting that IL-12 levels are particularly sensitive to perturbation. We next sought to determine whether neutrophils and Ly6Chi monocytes directly produce IL-12 during infection or whether other cell types produce IL-12 instead. In particular, Ly6Chi monocytes can differentiate into DCs when they enter inflamed tissues [56], and we had observed reduced levels of DCs in the lungs of Ccr2-/- mice following L. pneumophila infection (Fig 5D). We thus investigated whether Ly6Chi monocytes or DCs produce IL-12. Compared to naïve mice, we found that in infected mice, there was a significant increase in the total numbers and percentages of Ly6Chi monocytes producing IL-12 at 24 hours post-infection, as determined by flow cytometric analysis of intracellular cytokine staining (Fig 7A and S2A Fig). Though DCs also produced substantial amounts of IL-12 by intracellular cytokine staining (Fig 7B), the total numbers and percentages of IL-12-producing DCs were not significantly higher in infected mice when compared to naïve mice at 24 hours post-infection (Fig 7B and S2A Fig). As another means of tracking the cellular sources of IL-12 during infection, we used IL-12p40-YFP reporter (YET40) mice, which express IL-12p40 and YFP as a bicistronic transcript under control of the IL-12p40 promoter [58]. We found a significant increase in the percentages and numbers of monocytes and DCs that produce IL-12, as measured by YFP production at 48 hours post-infection, when compared to naïve YET40 mice (Fig 7C and 7D and S2B Fig). These data suggest that both monocytes and DCs are major sources of IL-12 during pulmonary infection, in agreement with recent findings [57].
As mice treated with anti-Ly6G antibodies had a significant defect in IL-12 production at 24 hours post-infection (Fig 4G), we next asked if neutrophils also serve as a direct source of IL-12 during pulmonary infection. We found that a significantly higher number of neutrophils in the lungs of infected mice produced IL-12 compared to naïve mice at 24 hours-post infection, as determined by intracellular cytokine staining (S2C Fig), consistent with a prior study indicating that neutrophils stain positive for IL-12 during infection with Legionella pneumophila [46]. However, the level of IL-12 staining observed in neutrophils was relatively modest in comparison to the robust levels of IL-12 that we observed in monocytes and DCs during infection. IL-12 in neutrophils is thought to be premade and stored in granules that are rapidly released upon infection [19], thus making it difficult to detect IL-12 by intracellular cytokine staining. Therefore, we also employed IL-12p40-YFP reporter mice as a complementary method for assessing IL-12 production by neutrophils. Infected YET40 mice had a significant increase in the percentage and number of YFP+ neutrophils compared to uninfected mice at 48 hours post-infection (S2D Fig), further supporting that neutrophils indeed do produce IL-12 during pulmonary infection with L. pneumophila. Additionally, we performed single-molecule RNA fluorescence in situ hybridization (FISH), which enables highly specific and sensitive detection of individual mRNA transcripts, as another method to assess IL-12 production in neutrophils from the lungs of infected mice. While we observed single cell variability in Il12 mRNA expression in neutrophils, some neutrophils were found to have very high ll12 absolute mRNA counts (ranging from 0–100 individual mRNAs per cell) (S3A and S3B Fig). Overall, these data indicate that Ly6Chi monocytes, DCs, and neutrophils all serve as cellular sources of IL-12 during L. pneumophila infection.
As our data thus far indicate that both neutrophils and Ly6Chi monocytes produce IL-12 and are required for maximal IL-12 responses during pulmonary L. pneumophila infection, we next sought to determine the functional roles of neutrophil- and Ly6Chi monocyte-derived IL-12 during infection. IL-12 can elicit production of the cytokine IFNγ [59,60], and IFNγ is required for host defense against pulmonary L. pneumophila infection [61]. Therefore, we next examined whether neutrophils and Ly6Chi monocytes are also required for IFNγ production during L. pneumophila infection.
In mice depleted of both neutrophils and monocytes with the anti-Gr-1 antibody, we observed a significant and sustained decrease in IFNγ levels in the BALF at 24 and 72 hours post-infection (Fig 7E), in agreement with previous studies also using anti-Gr-1 antibody-mediated depletion [22,46]. In mice depleted of neutrophils with the more specific anti-Ly6G antibody, we also observed a significant decrease in IFNγ production at 72 hours post-infection compared to isotype-treated mice (Fig 7F). Because Ccr2-/- mice also exhibited a profound defect in IL-12 production, we next examined secretion of IFNγ into the airway space of Ccr2-/- mice during infection. Strikingly, Ccr2-/- mice also produced significantly less IFNγ at both 24 and 48 hours post-infection when compared to WT mice (Fig 7G), in agreement with recent findings [57]. Furthermore, exogenous administration of recombinant IL-12 significantly enhanced control of bacterial loads in anti-Gr-1-treated mice, anti-Ly6G-treated mice and Ccr2-/- mice (Fig 8A–8C). Recombinant IL-12 treatment also significantly restored IFNγ production in anti-Gr-1-treated mice and Ccr2-/- mice (Fig 8A and 8B). Overall, these data indicate that neutrophils and monocytes are required for maximal production of both IL-12 and IFNγ, and that IL-12 from monocytes and/or neutrophils directs other cell types to produce IFNγ, thus promoting bacterial clearance.
We next asked which cell types produce IFNγ in response to either neutrophil- or monocyte-dependent IL-12. NK cells produce IFNγ early during pulmonary L. pneumophila infection [61]. Thus, we first examined the impact of neutrophil or Ly6Chi monocyte deficiency on IFNγ production by NK cells and focused on 24 hours post-infection, as the defect in IFNγ production was already evident at this timepoint (Fig 7E). In infected B6 mice, we observed a significant increase in the numbers and percentages of IFNγ+ NK cells compared to naïve B6 mice, in agreement with previous findings (Fig 9A and 9B and S4B Fig). In contrast, significantly fewer NK cells produced IFNγ in Ccr2-/- mice (Fig 9A and 9B and S4B Fig). Though a subset of NK cells express CCR2 [62,63], we did not observe a defect in NK cell recruitment to the lung, as B6 mice and Ccr2-/- mice had similar total numbers of lung NK cells during infection (S4A Fig). Additionally, Ccr2-/- NK cells from infected mice robustly produced IFNγ when activated with PMA and ionomycin (P/I), indicating that Ccr2-/- NK cells do not have a cell-intrinsic defect in the ability to produce IFNγ (S4C and S4D Fig).
IFNγ is critical for controlling L. pneumophila infection, but NK cells do not account for all of the IFNγ produced during infection [61]. We therefore sought to identify additional lymphocyte populations that produce IFNγ early during pulmonary infection and determine whether IFNγ production by these other cell types also relies on neutrophils and Ly6Chi monocytes. We found that lung αβ T cells do not produce detectable amounts of IFNγ at 24 hours post-infection, and the absence of Ly6Chi monocytes did not alter their ability to make IFNγ (S5A and S5D Fig). We next examined γδ T cells and NKT cells, as they rapidly produce IFNγ in response to other bacterial pathogens [64,65]. We observed a significant increase in the percentage and total numbers of IFNγ-producing γδ T cells and NKT cells in the lungs of infected B6 mice compared to uninfected mice (Fig 9C–9F and S6A and S6B Fig). CCR2 deficiency did not affect IFNγ production by NKT cells at 24 hours post-infection (Fig 9C and 9D and S6A Fig), but was associated with a significantly reduced percentage and total number of IFNγ+ γδ T cells (Fig 9E and 9F and S6B Fig). The decreased number of IFNγ+ γδ T cells in Ccr2-/- mice was not due to a defect in the total number of γδ T cells, as Ccr2-/- and B6 mice had equivalent numbers of lung γδ T cells during infection (S6C Fig). Additionally, γδ T cells from infected Ccr2-/- mice produced IFNγ when stimulated in vitro with PMA and ionomycin (P/I) (S6D and S6E Fig), suggesting that CCR2 deficiency did not cause a cell-intrinsic defect in IFNγ production. These data suggest that NK cells, NKT cells and γδ T cells all produce IFNγ during pulmonary L. pneumophila infection, and that Ly6Chi monocytes direct IFNγ production by these cell types, in agreement with recent findings [57].
In infected mice treated with anti-Gr-1 antibody, the total number and percentage of IFNγ-expressing NK cells was significantly reduced compared to mice treated with isotype control antibody (S4E–S4G Fig). Furthermore, anti-Gr-1 treatment led to a significant decrease in the percentage and total numbers of IFNγ+ NKT cells (S7A and S7B Fig), but did not affect the percentage or total numbers of IFNγ+ γδ T cells (S7C and S7D Fig). Specific depletion of neutrophils using anti-Ly6G antibody did not significantly affect IFNγ production by any cell type, as measured by intracellular cytokine staining (S7E Fig). It is therefore difficult to conclude whether neutrophils direct optimal IFNγ production by a particular cell type, despite our observation that neutrophil depletion decreases total IFNγ levels in the BALF (Fig 7F). We also considered the possibility that neutrophils themselves serve as a source of IFNγ during L. pneumophila infection, as neutrophils have been shown to produce IFNγ during T. gondii, Salmonella Typhimurium, Streptococcus pneumoniae, and other infections [16–18,20]. Our attempts to analyze IFNγ production by neutrophils using antibody-based methods, such as immunofluorescence microscopy, were inconclusive. We found that the anti-IFNγ antibody exhibited highly variable and non-specific staining that was comparable in neutrophils from WT infected mice or in neutrophils from IFNγ-deficient infected mice (S8A Fig). Thus, we employed single molecule RNA FISH as a more specific method for assessing IFNγ production by neutrophils. Although Ifng mRNA levels were variable in neutrophils from infected mice, many neutrophils exhibited high Ifng absolute mRNA counts (ranging from 0–100 individual mRNAs per cell) (S3A and S3B Fig), indicating that neutrophils produce IFNγ during L. pneumophila infection.
During pulmonary L. pneumophila infection, neutrophils and Ly6Chi monocytes are both rapidly recruited to the lung and produce overlapping sets of proinflammatory cytokines. However, the individual contributions of neutrophils and monocytes to the overall cytokine response and bacterial clearance remained unclear. Using antibody-based and genetic strategies to selectively ablate neutrophils, singly or in combination with Ly6Chi monocytes, our data suggest essential and non-redundant roles for both neutrophils and Ly6Chi monocytes in driving cytokine responses and control of bacterial burdens. Notably, we show that Ly6Chi monocytes and neutrophils are critical for host defense against L. pneumophila infection. We also find both neutrophils and monocytes contribute to IL-12 production, and monocytes are additionally required for TNF production. Furthermore, both neutrophils and monocytes are required for maximal IFNγ production, and monocyte-derived IL-12 directs IFNγ production by NK cells and innate-like NKT and γδ T lymphocytes. Therefore, our data suggest that neutrophils and Ly6Chi monocytes cooperate in shaping an optimal cytokine response that contributes to successful control of L. pneumophila infection. These findings highlight the functional crosstalk required between various innate immune cell populations to generate a protective cytokine response against bacterial infection.
Ly6Chi monocytes are recruited from the bone marrow into sites of infection and are appreciated as having an important role in host defense against several pathogens [11–14]. Notably, our study, along with another recently published study [57], are among the first to examine the contribution of CCR2-dependent Ly6Chi monocytes and their derivative cells to control of pulmonary L. pneumophila infection. Consistent with the requirement for CCR2 in emigration of Ly6Chi monocytes from the bone marrow, Ccr2-/- mice had a severe defect in Ly6Chi monocyte recruitment to the lung following L. pneumophila infection. Critically, despite robust recruitment of neutrophils to the lung of Ccr2-/- animals, which was indistinguishable from WT mice, Ccr2-/- mice exhibited significant defects in TNF and IL-12 production and delayed clearance of bacteria from the lung. In addition, levels of IL-12 were more substantially reduced in anti-Gr-1-treated mice and Ccr2-/- mice than in anti-Ly6G-treated mice, suggesting that monocytes are the major producers of IL-12 during L. pneumophila infection. Previously, during L. monocytogenes infection, Ly6Chi monocytes were shown to produce IL-18, which then drives IFNγ production by T cells and NK cells [66]. Our data indicate that during L. pneumophila infection, Ly6Chi monocytes are dispensable for IL-18 production. Instead, we found that monocytes are required for IL-12 production and subsequent IFNγ production by NK cells and γδ T cells. As monocytes are also phagocytic cells that produce nitric oxide and reactive oxygen species [67], and can exert direct bactericidal activity against L. pneumophila in an IFNγ-dependent manner [57,68], it is likely these bactericidal activities, in addition to cytokine production, also contribute to control of L. pneumophila infection.
We found that Ccr2-/- mice were not only deficient in Ly6Chi monocytes but also lacked CD11c+MHCII+ DCs by 48 hours post-infection with L. pneumophila, consistent with previous findings demonstrating that Ly6Chi monocytes differentiate into DCs in inflamed tissues [56]. We show here that both Ly6Chi monocytes and DCs produce IL-12 during L. pneumophila infection, and we have previously found that both Ly6Chi monocytes and DCs also produce TNF [38]. Thus, absence of both cell types in the Ccr2-/- mice may be responsible for the defect in TNF and IL-12 observed in these mice. As total DC numbers in the Ccr2-/- mice were not significantly different than those observed in B6 mice at 24 hours post-infection, but were significantly different at 48 hours post-infection, it is likely that monocytes are the major producers of TNF and IL-12 at 24 hours post-infection, with an additional contribution from DCs occurring at later timepoints. Our findings are in agreement with a recent study that also identified an essential role for Ly6Chi monocytes in host defense against L. pneumophila [57]. They similarly observed that Ly6Chi monocytes were a critical source of IL-12 and were required for optimal IFNγ production by NK cells, NKT cells, and γδ T cells. They also found that monocytes instructed memory αβ T cells to produce IFNγ [57], whereas we did not observe robust IFNγ production by αβ T cells. This likely reflects a difference in the timepoints analyzed in the two different studies, as we analyzed an earlier timepoint (day 1) post-infection, whereas they examined day 2 post-infection, which likely allowed for increased IFNγ production by these cells.
In addition, our study clarifies the role of neutrophils in host defense against pulmonary L. pneumophila infection. The anti-Gr-1 antibody clone RB6-8C5, previously used to deplete neutrophils during pulmonary L. pneumophila infection [46], is now known to bind and deplete cells expressing either Ly6G or Ly6C, which are expressed by neutrophils, monocytes, and activated T cells [47]. Comparing Gr-1-mediated depletion to neutrophil-specific Ly6G-mediated depletion has revealed that Ly6Chi monocytes, rather than neutrophils, are required for host defense against L. monocytogenes [12] and T. gondii [48]. Thus, we chose to compare anti-Gr-1-mediated depletion and anti-Ly6G-mediated depletion in the current study to examine the effects of anti-Gr-1 depletion on Ly6Chi monocytes and to elucidate the role of neutrophils in host defense against L. pneumophila. Our data revealed that anti-Gr-1-mediated depletion with high doses of RB6-8C5 antibody led to a decrease in total Ly6Chi monocyte numbers in the lung following L. pneumophila infection, similar to what we found in uninfected mice administered the anti-Gr-1 antibody. We found that depletion of Ly6Chi monocytes was significant at 72 hours post-infection, but there was minimal depletion at 24 hours post infection. Although we did not observe depletion of monocytes at the 24 hour timepoint, we cannot rule out the possibility that the anti-Gr-1 antibody affects the function of these remaining monocytes.
Anti-Gr-1 antibody treatment was more efficient at removing neutrophils than the more neutrophil-specific anti-Ly6G antibody, but also depleted monocytes. We observed increased bacterial burdens and defective IL-12 and IFNγ production in anti-Ly6G-depleted mice, indicating that neutrophils contribute to production of IL-12 and IFNγ and control of bacterial loads in the lung. Given that neutrophils are required for maximal IL-12 and IFNγ production in infected mice, we asked whether neutrophils directly produce these cytokines. Our flow cytometry data suggest that neutrophils produce IL-12, albeit at low levels relative to monocytes and DCs. Our attempts to further determine whether neutrophils produce IL-12 and IFNγ using immunofluorescence microscopy were inconclusive, as the anti-IL-12 and anti-IFNγ antibodies resulted in comparable staining in neutrophils from WT mice or IL-12- or IFNγ-deficient mice, leading us to conclude that anti-IL-12 and anti-IFNγ staining was nonspecific under our experimental conditions (S8A and S8B Fig). However, single molecule RNA FISH, which is a more specific and sensitive method for assessing gene expression, revealed that some neutrophils from infected mice exhibited high Il12p40 and Ifng absolute mRNA counts (ranging from 0–100 individual mRNAs per cell) (S3A and S3B Fig). Although the flow cytometry and RNA FISH data suggest that the majority of neutrophils do not produce high levels of IL-12 and IFNγ during L pneumophila infection, the large numbers (2x107-4x107) of neutrophils infiltrating the lung may enable these cells to significantly contribute to overall cytokine production. Neutrophils are classically thought to control bacterial burdens through direct anti-microbial mechanisms involving phagocytosis and degradation of microbes or production of NETs [6–10]. During L. pneumophila infection, neutrophils are major producers of reactive oxygen species, which is critical for effective control of infection [45]. Thus, in addition to contributing to cytokine production, neutrophils carry out important bactericidal functions that also contribute to control of L. pneumophila infection.
Our data shed additional light on the cell types involved in production of IFNγ, which is critical for control of L. pneumophila infection [57,61]. We found that NK, NKT, and γδ T cells serve as cellular sources of IFNγ, in agreement with recent findings [57]. IFNγ produced by NKT and γδ T cells is likely to be functionally important, and presumably accounts for the published finding that NK cell depletion had no impact on control of pulmonary L. pneumophila infection, despite reducing IFNγ levels [61]. IL-18 is required for maximal IFNγ production by NK cells during L. pneumophila infection [61] [65], but the critical cellular sources of IL-18 during pulmonary infection remain unknown. A previous study using an intravenous model of L. pneumophila infection found that neutrophil-derived IL-18 was critical for IFNγ production by NK cells [22]. In contrast, during pulmonary infection, our data suggest that neutrophils are not an essential source of IL-18, suggesting one or more cell types in the lung act as a redundant or critical source of IL-18. Instead, both neutrophils and monocytes were required for maximal IL-12 production, and monocytes were required for NK cells to produce IFNγ. Our data support a model in which IL-12 and IL-18 act in concert to elicit optimal IFNγ production, consistent with previous findings showing that both IL-12 and IL-18 are critical for maximal IFNγ responses [41,61,69]. Although our data suggest that neutrophils are required for maximal IFNγ production, we were unable to determine whether neutrophils direct IFNγ production by NK cells and innate-like lymphocytes. This may be due to inefficient depletion of neutrophils following anti-Ly6G treatment. Intriguingly, our data suggest that neutrophils express high levels of Ifng mRNA, suggesting that neutrophils themselves may serve as a source of IFNγ, as has been observed in other infection models [16,20,70].
Overall, our findings provide new insight into the roles of Ly6Chi monocytes and neutrophils during pulmonary L. pneumophila infection. We find that both monocytes and neutrophils are critical for control of bacterial infection. We further show that both neutrophils and monocytes were required for maximal IL-12 and IFNγ production, but monocytes were distinct in their essential contribution to TNF production. Finally, our data demonstrate that Ly6Chi monocytes instruct NK cells and innate-like lymphocytes to produce IFNγ, a key cytokine required for eventual control of L. pneumophila infection. Thus, our findings reveal critical but nuanced roles for neutrophils and Ly6Chi monocytes in shaping an optimal cytokine response that ensures successful host defense against pulmonary bacterial infection.
All animal studies were performed in compliance with the federal regulations set forth in the Animal Welfare Act (AWA), the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, and the guidelines of the University of Pennsylvania Institutional Animal Use and Care Committee. All protocols used in this study were approved by the Institutional Animal Care and Use Committee at the University of Pennsylvania (protocols #804714 and #804928).
C57BL/6J, B6.129S7-Ifngtm1Ts/J (Ifng-/-), and B6.129S1-Il12btm1Jm/J (Il12p40-/-) mice were purchased from The Jackson Laboratory. Ccr2-/- mice [54], YET40 mice [58], and Il12p40-/- mice were purchased from The Jackson Laboratory and maintained and bred in a specific pathogen-free facility at the University of Pennsylvania.
All experiments used the Legionella pneumophila serogroup 1 JR32-derived (rpsL and hsdR) strain lacking FlaA (ΔflaA Lp) [52,71]. L. pneumophila was cultured on charcoal yeast extract (CYE) agar plates containing streptomycin for 48 h at 37°C before infection.
8–12 week old mice were anesthetized by intraperitoneal (i.p.) injection of a ketamine/xylazine solution (100mg/kg ketamine and 10mg/kg xylazine) diluted in PBS. Mice were then infected intranasally (i.n.) with 40μl of 1x106 bacteria suspended in PBS by instilling 10μl at a time into the nostrils. For experiments involving addition of recombinant IL-12 (Peprotech), bacteria were instilled into the nostrils 10μl at a time and 500 ng of IL-12 or PBS was administered in between doses of bacteria (alternating 20μl bacteria, 20μl cytokine, 20μl bacteria). To collect BALF, 1mL cold PBS was slowly instilled into the lung through a catheter (Jelco) and retrieved immediately. To quantify CFUs, lungs were excised, weighed, and a portion was mechanically homogenized in sterile distilled H2O with a gentleMACS dissociator (Miltenyi Biotec). Lung homogenates were plated on CYE plates containing streptomycin, and CFUs were enumerated.
For anti-Gr-1-mediated depletions, mice were injected i.p. with 250μg of either α-Gr-1 (clone RB6-8C5) or rat IgG2b isotype control antibody (clone LTF-2) (Bio X cell). Mice were injected 16 hours before infection and then again 48 hours post-infection. For anti-Ly6G-mediated depletions, mice were injected i.p. with 250μg of either α-Ly6G (clone 1A8) or rat IgG2a isotype control antibody (clone 2A3) (Bio X Cell). Mice were injected 48 hours prior to infection and then again every 24 hours until the completion of the infection. 24 or 72 hours post-infection, lungs were harvested and CFUs were enumerated. The efficiency of neutrophil or monocyte depletion was monitored by flow cytometry.
A portion of the lung was weighed, cut into small pieces, and digested in PBS + 5% (vol/vol) FBS, 20 U/mL DNase I (Roche) and 200–300 U/mL collagenase type IV (Worthington Biochemical) at 37°C for 40 minutes, with shaking every 5 minutes. The tissue was then mechanically homogenized with a gentleMACS dissociator (Miltenyi Biotec) and red blood cells (RBCs) were lysed with RBC lysis buffer (7.44g/L NH4Cl and 2.06g/L Tris-HCl, pH 7.2 in distilled H2O), followed by quenching with cold PBS. After filtering through a 40μM cell strainer, cells were first treated with the Zombie Yellow fixable cell viability kit (BioLegend) at room temperature in PBS + 2mM EDTA for 15–20 minutes and then stained at 4°C in PBS + 2mM EDTA, 2% BSA, 0.1% sodium azide, 5% normal rat serum, and 5% normal mouse serum (Jackson ImmunoResearch) for 40 minutes with antibodies specific for the cell surface antigens CD45 (eBioscience, clone 30-F11, e650NC), CD11c (BioLegend, clone N418, BV785), Ly6G (BioLegend, clone 1A8, PE-Cy7), Ly6C (BioLegend, clone HK1.4, APC-Cy7), Gr-1 (eBioscience, clone RB6-8C5, PE-Cy5 or APC), CD11b (BioLegend, clone M1/70, PacBlue), MHCII (BioLegend, clone M5/114.15.2, AF700), Siglec F (BD Biosciences, clone E50-2440, PE or PE-Texas Red), NK1.1 (BioLegend, clone PK136, PE-cy5), CD3ε (BD Biosciences, clone 145-2C11, PE-Texas Red), TCRβ (BioLegend, clone H57-597, APC-Cy7), and TCRδ (BioLegend, clone GL3, PE-Cy7). Data were collected on an LSR II flow cytometer (BD Biosciences) and post-collection data were analyzed using FlowJo (Treestar). Cells were always pre-gated on live, CD45+ singlet cells. Neutrophils were identified as live, CD45+, Ly6G+, Ly6Cint cells. Ly6Chi monocytes were identified as live, CD45+, Ly6Glo, Ly6Chi, CD11b+ cells; in some experiments where indicated in the figure legend, an exclusion gate for B, T, and NK cells (CD19+, CD3+, NK1.1+) was also applied (S1B Fig). DCs were identified as live, CD45+, Ly6G-, Ly6C-, SiglecF-, CD11c+, MHCII+ cells. αβ T cells were identified as live, CD45+, NK1.1-, CD3ε+, TCRδ-, TCRβ+ cells. NK cells were identified as live, CD45+, CD3ε-, NK1.1+ cells. γδ T cells were defined as live, CD45+, NK1.1-, CD3ε+, TCRβ-, TCRδ+ cells. NKT cells were defined as live, CD45+, CD3ε+, NK1.1+ cells.
Cells from the lung were obtained as described above and were resuspended in RPMI + 10% (vol/vol) heat-inactivated FBS, 2 mM L-glutamine, 100 IU/mL penicillin, 100 μg/mL streptomycin, and 3.3 μg/mL brefeldin A for 3.5 hours (for IFNγ ICS) or 6 hours (for IL-12 ICS) at 37°C. Cells were then washed and stained for surface markers as above. After extracellular staining, cells were fixed and permeabilized using the Cytofix/Cytoperm buffer set (BD Biosciences) and stained in Perm/Wash buffer (BD Biosciences) at 4°C for IFNγ (eBiosciences, clone XMG1.2, APC) or IL-12p40 (eBiosciences, clone C17.8, eFluor660) for 1 hour. Flow cytometric data acquisition and analysis was performed as above.
Single-molecule RNA FISH was performed as described previously [72]. Briefly, cells were fixed in 3.7% (vol/vol) formaldehyde in 1X PBS for 10 min at room temperature and stored in 70% ethanol at 4°C until imaging. Pools of fluorescently labelled Stellaris Custom RNA FISH probes (Ifng–Quasar 570; Il12p40 –CAL Fluor Red 610, Gapdh–ATTO 488) (Biosearch Technologies) were hybridized to samples, followed by DAPI staining and wash steps performed in suspension. Samples were cytospun onto slides for imaging on a Nikon Ti-E inverted fluorescence microscope. For image processing, boundaries of cells were manually segmented from brightfield images and RNA spots were localized using custom software written in MATLAB [73].
Harvested BALF from infected WT, Ifng-/-, and Il12p40-/- mice were collected and incubated in RPMI + 10% heat-inactivated FBS, 2 mM L-glutamine, 100 IU/mL penicillin, 100 μg/mL streptomycin, and 3.3 μg/mL brefeldin A for 3 hours. For anti-IFNγ staining, following brefeldin A treatment, cells were then cytospun onto Superfrost Plus microscopy slides (Thermo Fisher Scientific) and fixed in 4% paraformaldehyde, followed by washes with PBS + 0.2% Triton X-100 and blocking in PBS + 20% horse serum. Anti-mouse IFNγ antibody (eBioscience, clone XMG1.2, AlexaFluor 488) was applied in PBS + 2% horse serum + 0.1% Triton X-100, followed by washing and coverslip mounting with ProLong Gold Antifade with DAPI (Thermo Fisher Scientific). For anti-IL-12 staining, following brefeldin A treatment, cells were stained using the protocol described above for intracellular cytokine staining with anti-mouse IL-12p40 antibody (eBioscience, clone C17.8, AlexaFluor 488). Cells were then cytospun onto Fisher Scientific Superfrost Plus microscopy slides, followed by washing and coverslip mounting with ProLong Gold Antifade with DAPI (Thermo Fisher Scientific). Images were acquired on a Leica DM6000 microscope.
Harvested BALF from infected mice was assayed using kits specific for murine IL-1α (Biolegend), IL-1β (Biolegend), IL-6 (Biolegend), TNF (Biolegend), IL-18 (MBL International), IL-12p70 (Biolegend), IFNγ (BD Biosciences), IL-4 (Biolegend), IL-10 (Biolegend), or paired capture and detection antibodies specific for IL-12p40 (BD Biosciences).
Plotting of all data and statistical analyses were performed using GraphPad Prism software. For comparisons between more than two groups, statistical significance was determined using a one-way ANOVA with Tukey post-test. For comparisons between two groups, statistical significance was determined using an unpaired Student’s t-test. Differences were considered statistically significant if the P value was <0.05.
|
10.1371/journal.pcbi.1000835 | The Energy Landscape, Folding Pathways and the Kinetics of a Knotted Protein | The folding pathway and rate coefficients of the folding of a knotted protein are calculated for a potential energy function with minimal energetic frustration. A kinetic transition network is constructed using the discrete path sampling approach, and the resulting potential energy surface is visualized by constructing disconnectivity graphs. Owing to topological constraints, the low-lying portion of the landscape consists of three distinct regions, corresponding to the native knotted state and to configurations where either the N or C terminus is not yet folded into the knot. The fastest folding pathways from denatured states exhibit early formation of the N terminus portion of the knot and a rate-determining step where the C terminus is incorporated. The low-lying minima with the N terminus knotted and the C terminus free therefore constitute an off-pathway intermediate for this model. The insertion of both the N and C termini into the knot occurs late in the folding process, creating large energy barriers that are the rate limiting steps in the folding process. When compared to other protein folding proteins of a similar length, this system folds over six orders of magnitude more slowly.
| Proteins are chains, which must fold into a compact structure for the molecule to perform its biological function. There are a large number of ways the molecule can move into this final shape. Proteins have evolved sequences that perform this difficult task by having strong biases toward the final shape, while not getting stuck in different structures along the way. One way proteins can be trapped is by forming a knot in the chain. For the most part, proteins are remarkable in avoiding knotting. However, in order to function a few proteins form knots. We show how a model protein is able to knot itself, and estimate how fast this process occurs. Our goal is to treat a small and uncomplicated protein to estimate the fastest rate possible for the folding of a knotted protein. This rate is interesting when compared to the speed of folding of other proteins. We have visualized how the molecule changes shape to its functional position, and examined other paths the molecule may take.
| The wide range of kinetics that characterizes protein folding has attracted interest from both experimentalists and theoreticians for decades [1]. Proteins fold on time scales that vary from microseconds to minutes [2], even though the corresponding energy landscape directs folding towards its native state. This wide range of rates can be explained by the diverse size and shapes of the free energy barriers between the unfolded and folded ensembles, which are largely determined by the pattern of contacts, often called the protein topology. Some of the slowest folding proteins, such as the green fluorescent protein, are both long and have complicated contact patterns [3]. In the past 15 years, a set of proteins has been discovered that fold slowly and have knotted topologies [4]. Topological constraints can lead to large energy barriers that are difficult to characterize, so we employ a reduced description of the protein and explore the landscape using geometry optimization techniques. Topological problems have been investigated previously using G models [5], and have shown that local unfolding may be required in some cases to organize the sequence of the folding of structural units. This type of potential energy landscape analysis with transition state theory has been used for describing kinetic phenomena in systems as diverse as molecular clusters, glasses, and proteins [6]. An advantage of this approach is that the topological constraint limits the utility of a simple energy based reaction coordinates (like , see Methods section), which ordinarily work well to describe folding.
We studied a tRNA methyltransferase (PDB code 1UAM), which contains a deep trefoil knot in the C terminus domain [7]. The goal of this study was to estimate the fastest speed possible for folding a small knotted protein, and we therefore truncated the system to residues 78–135 to limit the number of atoms not included in the knot. Mathematically knots are defined in closed loops. In proteins links are used to connect the termini, and the structure is topologically classified by the determination of its Alexander polynomial [8], [9]. Recently knotted proteins have been identified [10] and their kinetics explored. Knotted systems with a larger number of residues beyond the knot will exhibit slower kinetics, because of the need to break a larger number of contacts to fold properly. Protein models based on random contacts produce knotted proteins with greater frequency than is seen in protein structure databases [11], [12]. Knotted proteins are likely avoided during evolution, while some have remained and are an evolutionary curiosity. Knotting can also occur in other biopolymers such as DNA, and these systems exhibit significantly slower kinetics [13].
A connected path of minima and transition states between an unfolded structure and the native state was created with the discrete path sampling method (DPS) combined with the associative memory Hamiltonian [14], [15]. After obtaining an initial connection, this database of structures was expanded using schemes to systematically reduce the length and barriers associated with the largest contribution to the rate coefficient [16]–[19]. Here, the SS superscript refers to an approximate formulation where the steady-state condition is applied to minima outside the product and reactant sets. This formulation provides a convenient framework for analysis because can be written as a sum over discrete paths [16], [20]. Once this pathway appeared to have converged, the database was further refined by connecting undersampled minima with a large ratio of the free energy barrier to free energy difference from the global minimum [21], [22]. This choice is motivated by the idea of optimizing folding kinetics for topologically constrained system in a similar way to the minimization of energetic frustration obtained by comparing the folding temperature to the glass transition temperature [23], [24].
The resulting database contained 212054 minima and 206923 transition states, and the corresponding disconnectivity graph [25], [26] is shown in Figure 1, as rendered by VMD [27]. Here we remind the reader that every vertical line in the graph terminates at the energy of a local minimum, and that the minima are progressively connected together as the threshold energy, , increases, according to the lowest barrier between them. The graph exhibits three distinct color coded features corresponding to potential energy basins, with properly knotted minima occupying the lowest-lying states in the center of the figure. Branches corresponding to minima with the knotted topology are colored blue, while those with the C and N termini still free are colored green and red, respectively. The kinetic coefficients for interconverting minima within these basins are relatively fast, so that local equilibrium is achieved on the time scale of the slow kinetics determined by the barriers between the different basins. This figure represents an unusual folding energy landscape, where large energy barriers occur despite the lack of favorable non-native interactions in the Hamiltonian. The two higher energy sets of structures correspond to local minima where either the C or N terminus has the native topology, but the other terminus is still unknotted, and we will refer to these as N-free and C-free geometries, respectively (Figures 2 and 3).
A useful descriptor of these ensembles is their structural overlap. The value (a measure of structural similarity, see Methods section) between the N-free and native minima is , between the C-free and native minima is , and between N-free and C-free minima is . The small variation in shows that only a few contacts are different, where a contact is restricted to be less than 9 Å in order to distinguish structures that have a high degree of similarity. Most of the contacts in each basin are identical, except for a few important differences near the termini. In the N-free minima these interactions are between residues 7–8 and 45–46 as shown in Figure 4, while in the C-free minima they are between 48–52 and 30–34 as shown in Figure 5, and define the interactions that prevent unphysical chain crossings. These non-native contacts are energetically neutral with respect to the interaction Hamiltonian (see Methods section) due to the native-only form of the Hamiltonian, but they affect the calculated pathways through an excluded volume repulsion due to the backbone interaction .
A kinetic analysis of the DPS database using transition state theory requires choices for the value of and the mass associated with each site in the AMH potential. For simplicity a value of 12 atomic mass units (amu) was assigned to each site. To assign a value for , we compared the normal mode frequencies for the AMH potential with typical values associated with heavy atom motion in all-atom representations of proteins. This comparison suggests that ε should be around one kcal/mol. The discrete path that makes the largest contribution to the phenomenological two-state rate coefficient, which we use to define the overall reaction mechanism, exhibits the same qualitative features over a wide range of values for . The estimated rate coefficients themselves are more sensitive, as discussed below. If we take kcal/mol and the values of length and mass in the AMH potential as 1 Å and 12 amu then the reduced value of at room temperature is approximately 0.59. The pre-exponential factor for each minimum-to-minimum rate coefficient scales as , while the reduced value for room temperature decreases linearly, lowering the corresponding Boltzmann factors exponentially.
The choice of the reactant and product states can have a significant effect on the calculated rates. One way of selecting the states is to calculate an order parameter for all the local minima, and simply assign reactant and product states on this basis. However, an alternative method is possible using the known characteristics of the kinetic transition network and a self-consistency condition. Here we take the two endpoints that were used to calculate an initial path, and assign these minima to reactant and product sets. We then regroup the database using a recursive scheme [21], combining free energy minima that can interconvert without encountering a barrier higher than a chosen threshold value, . This approach is attractive, because we require a separation of time scales for equilibration in the product and reactant regions, compared to the folding transition time, in order to recover a two-state description of the kinetics [20], [28], [29]. In this case, we expect to see a range of values for that give a similar value for the calculated rate coefficient.
Rate coefficients were calculated for three different choices of the reactants, namely a fully unfolded minimum and low-lying minima from the C-free and N-free regions of the landscape. In each case a low-lying minimum from the set of knotted configurations was chosen as the product, and rate coefficients were calculated for a range of and values. Following the recursive regrouping of states according to the value of , mean first-passage times were calculated from each minimum in the reactant set using a graph transformation procedure [17], [30]. A phenomenological two-state rate coefficient is then obtained using appropriate occupation probabilities for the starting minimum [16], [17], [20], [30]. These values of tested (1.0, 0.9 and 0.7 kcal/mol) are close to the magnitude suggested by examining the normal mode frequencies. For kcal/mol the rate coefficient varies from 0.04 to 0.4 s for kcal/mol using a fully unfolded or C-free minimum as the reactant. Therefore the folding time is between 25 seconds and 2.5 seconds.
A movie of the C-free minimum to the folded state is shown in Video S1. With a low-lying N-free minimum as the reactant, the calculated rate coefficient is 0.02 s for kcal/mol, jumping to about 50 s for kcal/mol. So the folding time becomes 50 seconds to 0.02 seconds. A movie of the N-free minimum to the folded state is shown in Video S2. The values obtained rise by about two orders of magnitude starting from N-free as reactant if we set kcal/mol.
For each choice of reactant, the discrete path making the largest contribution to the rate coefficient [16] was extracted, and snapshots of the intervening structures are superimposed upon the energy as a function of integrated path length in Figures 6, 7, and 8. Here the path length is defined from the Euclidean distance between successive configurations in the folding reaction. These pathways are based on the rate coefficients and associated free energy barriers calculated at , and the entropy terms that derive from alternative discrete paths through the stationary point database are all included in the estimates of the overall rate coefficients. This kinetic analysis suggests that the local maxima in the energy profiles shown in Figures 6, 7, and 8 generally correspond to kinetic bottlenecks. Starting from an unfolded state, we see that the N terminus first forms a loop that threads through the middle of the protein, and then opens (Figure 6). The structures involved in this process appear very similar to the corresponding event in the pathway starting from an N-free minimum in Figure 8, aside from the state of the C terminus. The final folding events illustrated in Figure 6 are very similar to those shown in Figure 7, with a bend forming at the C terminus, threading through the center of the structure, and straightening. The calculated rate coefficients are also very similar when the reactant is chosen as either the fully unfolded state or a C-free minimum, indicating that knotting of the C terminus is the rate-determining step for this model system. The region of configuration space corresponding to low-lying N-free minima is then interpreted as a kinetic trap, which would probably result in a distinct relaxation time scale.
In this study a truncated sequence from a tRNA methyltransferase was considered with a G model containing only the favorable interactions that are present in the global minimum. In contrast to previous minimally frustrated models, which exhibit only a single potential energy funnel [31]–[33], the landscape for the knotted protein is divided into three distinct regions, corresponding to the correctly folded native state and to structures where either the C or N terminus are not knotted. The potential energy barriers between the lowest minima in these regions are relatively large, with values of order 15 to 20 in units of the associated memory Hamiltonian [34], [35] parameter , for which we estimate a value of around one kcal/mol. The calculated rates for folding are therefore rather slow, in agreement with previous simulations of knotted proteins [36], [37].
The folding reaction is hindered by the complex topology of this protein. Modeling these interactions and mechanisms in a realistic way requires new tools that prevent unphysical chain crossing events from occurring during the interpolation between structures that have an intervening chain. Details of the procedure are given in the Methods section, and an overview is provided here. Our initial aim was to avoid chain crossings by changing parameters of the potential. However, tightening the bond length constraints for covalent bonds does not solve the problem, because the interpolated images simply avoid the chain-crossing region. In the doubly-nudged [38] elastic band [39]–[41] (DNEB) method for identifying useful starting geometries for transition state refinement, a set of images are connected by harmonic springs, and the images can be equally spaced by increasing the corresponding force constant. However, this increase forces the chains into high energy structures that bracket an unphysical crossing. To avoid this situation it is necessary to construct a non-linear interpolation between the end points, and two strategies were implemented. To accelerate the energy evaluation, an elastic network potential was defined based on the two end points, with harmonic restraining potentials for atoms whose separation does not change. This geometrical analysis was also used to diagnose chain crossings for a linear interpolation between the end points, and to distinguish the chain that is moving from one that constitutes a barrier. When crossings were identified the potential was modified to shrink the end of the moving chain and add repulsive interactions to keep it away from the other chain. The DNEB images were then refined following standard procedures and the modified potential was morphed back into the AMH potential slowly enough for sites on one chain to move around the other chain, rather than through it. Overall, this procedure allows paths to be obtained that circumvent chain crossing, while retaining flexibility and providing a solution that is free of constraints.
Both the C and N termini must effectively cross over a chain belonging to the central region of the protein in order to achieve the knotted topology. In the present model it is the C terminus crossing that appears to be the rate-determining step. The rate coefficients reported here are order of magnitude estimates, and correspond to a slow folding process, as expected from previous simulations [36], [37] and experiments [42], [43]. The precise energetics of this truncated model may differ from those of the full protein, but we expect the key steps in the folding and knotting pathways to be retained. Making a meaningful estimate of the scaling behavior with respect to chain length will be addressed in future work. For both the C and N termini the chain cross-over are achieved by formation of a loop, which then inserts through the center of the protein and straightens. While the chain is in the loop conformation the folding process could notionally be reversed by pulling the end of the chain, which is one definition of a slipknot topology, consistent with previous simulations [37]. The region of configuration space corresponding to N-free local minima, where the C terminus crossing occurs first, is therefore likely to give rise to a separate relaxation time scale. The folding pathways exhibit some interesting mechanistic features, which might be transferable to related systems of knotted proteins and polymers. In particular, both the C and N termini crossings are achieved by formation of a loop that threads through the main body of the protein. The order of the knotting of the chain and the folding of the protein may change as the length of the system changes and as the energy function becomes more realistic. Adding non-native interactions would likely lower the free energy barrier of folding [44], but could also stabilize non-native structures and slow local refolding of the loops involved in chain crossings. Introducing non-additive cooperative contacts would increase the energy barriers and likely slow the kinetics [45]. Protein engineering studies of YibK suggest that knotting and formation of native structure are independent events that occur in sequence [46]. These experiments also suggest an early knotting event and slow development of native structure in the knotted region. Similar behavior is seen in DNA, where local unfolding speeds up diffusion of the knot along the polymer chain [47]. To provide meaningful comparisons with these observations will require simulations of longer systems, rather than the truncated sequence considered in the present work. When compared to other protein folding proteins of a similar length, which fold on the microsecond time scale [2], this system folds over six orders of magnitude more slowly.
In order to describe an energy landscape with an exponential number of states, we reduce the atomistic detail of the system and discretize the energy landscape into minima and transition states. The associative memory Hamiltonian (AMH) protein model [34], [35], is a coarse-grained molecular mechanics potential inspired by the physics of protein folding. The energy functions consist of a polypeptide backbone term, , with a molecular interaction term, [48]–[53]. The number of atoms per residue is limited to three (C, C, and O), except for glycine. The interaction parameter , which is the unit of energy, is defined by the native state energy excluding backbone contributions, , via(1)where is the number of residues. All temperatures are quoted in reduced units as . While creates self-avoiding peptide-like stereochemistry, introduces the majority of the attractive interactions that produce folding. Using the interactions described by , we define a pairwise additive G model [54], [55], which is biased toward the native basin. Such models have been shown to reproduce many features of the mechanism and kinetics of protein folding [56], [57]. The interactions between residues were defined by,(2)where the distances in the Gaussian term are determined by the native state. The interactions are defined in this minimal model for residues with greater than three residues sequence separation between the atom pairs. The weights, , corresponding to the depths of the Gaussian wells, are set to (0.177,0.048,0.430) in order to approximately divide the interaction energy equally between the different distance classes, as suggested by previous theoretical models [58]. The width of the Gaussian, , is determined by the sequence separation as Å. The scaling factor is used to satisfy Eq. (1). We measure the quality of the structures encountered with an order parameter, , which measures the sequence dependent structural similarity of two configurations. is calculated from Eq. (3) as a summation of pairwise differences between distances in a target and a reference structure (usually the native state), normalized by the number of contacts, where is sequence length:(3)The resulting order parameter, , ranges from zero, when there is no similarity between structures at a pair level, to unity, which indicates an exact overlap.
We made several changes to the original AMH backbone potential, , in the present work. Eliminating some compromises necessary for rapid molecular dynamics simulations allows the AMH potential to be used with geometry optimization methods to produce tightly converged stationary points. This tight convergence is necessary for the construction of a kinetic transition network [20], [22], [59]–[61]. The terms shown in Eq. (4) are used to reproduce the peptide-like conformations in the original molecular dynamics energy function:(4)For all calculations, we replaced the SHAKE method for bond constraints with a harmonic potential, , between the C-C, C-C, and C-O atoms. This replacement permits the location of local minima without requiring an internal coordinate transformation, and avoids discontinuous gradients [62]. The neighboring residues in sequence sterically limit the positions the backbone atoms can occupy, and this effect is reproduced with a Ramachandran potential, . The planarity of the trans peptide bond is ensured by another harmonic potential, . The chirality of the C centers is maintained using the scalar triple product between neighboring C, C, and N atoms, . Excluded volume repulsion between the backbone atoms is achieved with via a smooth step (hyperbolic tangent) function, , in order to have a continuous potential, and differs from the previous hard sphere potential in the AMH.
For this Hamiltonian, we employed the discrete path sampling (DPS) approach to create databases of local minima and their intervening transition states, starting from two end points. To identify suitable endpoints, we used basin-hopping global optimization [63], [64] to search for the global minimum of the energy landscape, and to create an unfolded conformation. We have previously shown how basin-hopping can be successfully combined with associative memory Hamiltonians for identifying low energy states, and high quality structures [62]. The discrete path sampling approach is a coarse-grained analogue of the transition path sampling method [29], [65], [66], where geometry optimization tools are employed to refine a kinetic transition network. The network consists of local minima and transition states of the energy potential, where a transition state is defined as a stationary point with a single negative Hessian eigenvalue [67]. The connectivity is defined by approximate steepest-descent paths obtained by energy minimization following infinitesimal displacements parallel and anti-parallel to the eigenvector corresponding to the unique negative eigenvalue. A discrete path then refers to a series of minimum-to-minimum connections together with the intervening transition states. The original DPS formulation has been presented in detail elsewhere [16], [20], as have more recent developments [18], [21]. The aim is to enlarge a database of connected stationary points starting from those in the initial path, by adding all the minima and transition states found during successive connection-making attempts for pairs of minima selected from the current database.
The main challenge of DPS calculations is the characterisation of transition states. In contrast, energy minimization and identifying approximate steepest-descent pathways is straightforward; here we used the limited-memory Broyden–Fletcher–Goldfarb–Shanno (LBFGS) algorithm of Liu and Nocedal [68], [69]. The transition state searches are connection attempts for a given pair of local minima. A doubly-nudged [38] elastic band [39]–[41] (DNEB) refinement of interpolated images was first run for each connection attempt, and the images corresponding to local energy maxima were then tightly converged using hybrid eigenvector-following [70]. The missing-connection algorithm [71] was employed to choose subsequent pairs of minima for further connection attempts [15].
To avoid unphysical chain crossed transition states, we made two changes in methodology to generate physical interpolations for finding potential transition state structures. We define two new potentials , which maintains chain connectivity, and , which introduces atomic repulsion. The potential is modified in three stages during the DNEB refinement. During the first third of the DNEB steps and are used with modified distances. In the second third the distances are relaxed to physically meaningful values, and in the final third we switch to the full AMH potential. We also define a simpler potential function, often referred to as an elastic network model [72] to represent the system during some of the DNEB refinement. The two end points for the DNEB calculation are analyzed to identify pairs of atoms within a cutoff distance (10 Å) that are found at the same separation within a given tolerance in both structures. If and are the distances between atoms and in the starting and finishing geometries, then we introduced a harmonic restraining potential for this pair if , where . For such pairs the restraining potential was then(5)where is the distance between the atoms involved in restraint , and was initially set equal to . The parameter was set to in the present calculations, where the DNEB spring constant was set to . has the appearance of an elastic network model [72], which reflects the conserved interatomic distances in the two endpoints. Analyzing the conserved distances is also useful for diagnosing when crossings occur, so that corresponding changes can be made to the potential, as described below. The initial images in the DNEB interpolation were simply placed at regular intervals for a linear interpolation between the specified endpoints, after putting these two structures into optimal alignment [73]. All pairs of atoms corresponding to different restraints with no common atoms were then examined for all pairs of DNEB images. The crossing check was applied for the largest untested image separation of every remaining image. Only pairs of restraints where the separation of the midpoints between the restrained atoms in both images were below a cutoff value of 10 Å were considered. The midpoint separation in one of the two images was also required to change by at least 3 Å from the value in the nearest endpoint structure. For restrained pairs satisfying these criteria a crossing is diagnosed when the dot product between the vectors joining the midpoints between the constraint pairs in the two images is negative. Outer and inner atom pairs are then defined according to how far the midpoints move between the two images: the midpoint that moves the furthest is assumed to belong to the outer chain, which needs to move around the inner chain.
To avoid unphysical crossings in the interpolation, we modify the potential and add repulsive terms through . If atomic contacts within the set of pairs are found to cross, using the above geometrical condition, then repulsive terms are added according to the four distances between the two pairs of atoms. For crossings of restrained distances, the repulsive contribution to the potential is(6)where is a step function, (10 Å) is a cut-off for the repulsive terms, () defines the magnitude of the repulsion, and is one of the four distances between pairs of atoms whose restrained contacts are found to cross. To enable chains to pass around one another when crossings are diagnosed, further changes were made to . For the restrained contact in the outer and inner chains was changed to and , respectively, for each crossing. Hence the outer chain shrinks while the inner chain expands. The first third of the DNEB iterations were run with the modified potential plus . For the middle third of the DNEB optimization the restraint distances were switched back to the value according to the schedule , with . The full AMH potential was then used for the last third of the DNEB iterations.
To describe the global kinetics of the transition network, we calculated the rate coefficients associated with each transition state using transition state theory [74] (TST) with vibrational densities of states obtained from harmonic normal mode analysis. The most important features of the mechanism of folding to the knotted state are relatively insensitive to the values assigned to minimum-to-minimum rate coefficients, while the total rate coefficients that we report are order of magnitude estimates.
|
10.1371/journal.pcbi.1007274 | A benchmark of computational CRISPR-Cas9 guide design methods | The popularity of CRISPR-based gene editing has resulted in an abundance of tools to design CRISPR-Cas9 guides. This is also driven by the fact that designing highly specific and efficient guides is a crucial, but not trivial, task in using CRISPR for gene editing. Here, we thoroughly analyse the performance of 18 design tools. They are evaluated based on runtime performance, compute requirements, and guides generated. To achieve this, we implemented a method for auditing system resources while a given tool executes, and tested each tool on datasets of increasing size, derived from the mouse genome. We found that only five tools had a computational performance that would allow them to analyse an entire genome in a reasonable time, and without exhausting computing resources. There was wide variation in the guides identified, with some tools reporting every possible guide while others filtered for predicted efficiency. Some tools also failed to exclude guides that would target multiple positions in the genome. We also considered two collections with over a thousand guides each, for which experimental data is available. There is a lot of variation in performance between the datasets, but the relative order of the tools is partially conserved. Importantly, the most striking result is a lack of consensus between the tools. Our results show that CRISPR-Cas9 guide design tools need further work in order to achieve rapid whole-genome analysis and that improvements in guide design will likely require combining multiple approaches.
| Modern genome engineering technologies provide unprecedented methods for DNA modifications. CRISPR-based systems generate a high level of interest in the community due to their effectiveness and relative simplicity. However, the design of the guides they use to target specific regions is not trivial. Researchers need to both maximise the likelihood of making the desired modification and minimise the risk of undesired, off-target changes. To assist researchers in making better informed decisions, a number of software tools have been developed to assist with guide design. Here, we analyse the performance of 18 design tools for CRISPR-Cas9 guides. We evaluate each tool based on their runtime, their computational requirements, and the output they generate. We benchmarked tools on datasets of increasing size (to evaluate their scalability) as well as on guides for which experimental data is available. Our results show that there is little consensus between the tools and that improvements in guide design will likely require combining multiple approaches.
| Wild-type CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) is found in archaea and bacteria, and acts as an adaptable immune system [1]. CRISPR is able to provide a method of immunity via three steps [2]: (i) a DNA snippet from an invading phage is obtained and stored within the CRISPR array, making a memory of past viral infection; (ii) the CRISPR region is expressed and matured to produce duplicates of previously obtained DNA snippets (or guides); (iii) in the case of S. pyogenes Cas9 (SpCas9), a guide binds with a SpCas9 nuclease to enable site-specific cleavage due to guide-DNA homology. This last step provides immunity to the host cell and is also the mechanism for CRISPR to be used in a genome engineering context, where a synthetic guide is supplied. CRISPR-based systems have been used for a number of such applications [3–5]. However, guide design is not trivial, as the efficiency and specificity of guides are crucial factors. For this reason, computational techniques are employed to identify and evaluate candidate CRISPR-Cas9 guides.
Here, we analyse 18 CRISPR-Cas9 guide design methods to evaluate whether they are adequate for rapid whole-genome analysis, and potentially whether combining approaches would achieve a solution of better quality. The available tools can be categorised based on algorithmic approach (i.e. procedural or via models trained using experimental data); however, when constructing the tool-set, we considered various factors: (i) whether the source code was easily obtained, (ii) is the installation process straight-forward, (iii) the tool simply does not provide a wrapper for performing a regular expression and (iv) the guide length and PAM sequence can be customised. In our analysis, we consider not only their output (i.e. which targets they identified), but also their ability to process whole genomes in a reasonable time. This is especially important for large genomes, such as some flowering plants. It is also a crucial feature for some applications such as studies of complex pathways or functions that require targeting multiple genes (e.g. sleep [6–8]) or producing whole-genome maps [9].
We selected 18 guide design tools that are released under an open-source license and report candidate guides for the Streptococcus pyogenes-Cas9 (SpCas9) nuclease; these tools are listed in Table 1. The most recent version was used for each tool, see Supplementary data for details.
Python (n = 11) and Perl (n = 5) are the most common programming languages, with CT-Finder and CRISPOR also implementing web-based interfaces via PHP (complemented by JavaScript, CSS, HTML, etc.). To improve run time, mm10db implements some of its components in C. SSC and FlashFry are implemented in C and Java, respectively. PhytoCRISP-Ex is a Perl-implemented tool, however, extensively utilises Linux bash commands for pre-processing. Configuration of tools is most commonly achieved via command-flags, however, tools such as Cas-Designer and CasFinder are configured via a text file. CHOPCHOP is configured via global variables in the main script file. Only Cas-Designer leveraged the GPU for additional computing resources. We ran this tool in both CPU and GPU modes when considering the computational performance.
SciPy [10] (inc. Numpy) and BioPython [11] were common packages utilised by the Python-based tools. CHOPCHOP and WU-CRISPR use SVMlight [12] and LibSVM [13], respectively; both being C implementations of support vector machines (SVM). Similarly, sgRNAScorer2 and TUSCAN utilise the machine learning modules from the SciKit-learn package [14]. The authors of sgRNAScorer2 supplied models for the 293T cell line and the hg19 and mm10 genomes. CHOPCHOP, GT-Scan, CRISPR-ERA, CCTop and CasFinder use Bowtie [15] for off-targeting; similarly to mm10db and CT-Finder with Bowtie2 [16]; while CRISPOR and CRISPR-DO utilise the Burrow-Wheelers Algorithm (BWA) [17]. GuideScan does not depend on external tools for off-targeting, and instead implements a trie structure for designing guides with greater specificity [18]. FlashFry benefits from its guide-to-genome aggregation method which is able to identify off-target sites in a single pass of its database. This achieved greater performance for FlashFry in comparison to BWA, as the number of mismatches and number of candidate guides increases [19]. Interestingly, [20] identified that Bowtie2 lacks the ability to rapidly identify all off-target sites with greater than two mismatches and that Cas-OFFinder [21] is a more suitable solution, however is more time expensive.
CHOPCHOP, Cas-Designer, mm10db, CCTop and CRISPR-DO provide ways to specify annotation files. For each of these tools, we provided the appropriate annotation in each test. CHOPCHOP utilises the annotations to indicate which gene or exon a candidate guide targets, and to allow the user to restrict the search region to particular genes. Cas-Designer utilises a custom-format annotation file, which describes the start and end positions of each exon on a given genome. This is used for designing candidate guides which specifically target exon regions. mm10db requires an annotation file in the UCSC RefGene format in order to generate a file containing sequences for all exons. CCTop utilises annotations for evaluating guides based on their distance to the closest exon, and for passing results to the UCSC genome browser as a custom track. None of the selected tools are restricted to particular genomes; all have been developed to allow for genomes of any organism to be provided. This includes personal genomes, for instance if applying these tools in contexts such as personalised medicine. DNA or RNA bulges were supported by two tools only, as described in Table 1. Genetic variation is not supported by any of the tools which we have included.
Some biological rules are shared across tools, such as: avoiding poly-thymine sequences [22], rejecting guides with inappropriate GC-content [23], counting and possibly considering the position of single-nucleotide polymorphisms (SNPs), and considering the secondary structure of the guides. Most tools report all targets that have been identified (sometimes with a score for specificity and/or efficiency) and rely on the user to determine whether a guide is appropriate for use, while mm10db actively filters guides and only reports ‘accepted’ ones (but ‘rejected’ targets, and reason for rejection, are still available in a separate file). WU-CRISPR and sgRNAScorer2 do not implement any of these rules through procedural-styled programming, but instead utilise machine learning models trained from experimental data. Furthermore, due to the age of some tools and the rapid growth of CRISPR-related research, the specifics of these procedural rules vary. For example, early studies describe the 10-12 base pairs adjacent to the PAM (the seed region) are more significant than the remainder of the guide [1, 24], but recent research contradicts this and suggests that one-to-five base pairs proximal to the PAM are more likely to be of significance [25]. Recent research has shown that specific motifs within the seed region reduce gene knock-out frequencies by up to 10-fold when present [26]. When evaluating tools, these may be factors that researchers would want to consider.
A similar set of candidate guides was generated by each tool, however, the start and end positions of identical guides often differed by up to 4 positions. This was seen due to some tools truncating the PAM sequence, or where zero-based positioning was used. An ellipsis was concatenated during normalisation for those that truncated it. All guides were aligned with one-based positioning as per the UCSC datasets, however, it is noted that Ensembl and Bowtie adhere to a zero-based system, explaining its usage in some tools. Seventeen of the tools produced output using the comma- or tab-separated values (CSV, TSV, respectively) formats, which makes their results easy to exploit for any post-processing. Five of the tools (GuideScan, CRISPR-ERA, mm10db, GT-Scan, CCTop and SSC) did not provide a header line to indicate the meaning of each column. GT-Scan differed as it produced an SQLite database containing a table of all candidate guides. Tests which were terminated did not produce output as the write-to-file routines had not been reached prior to termination.
Here, we discuss the consensus between tools for the datasets derived from chr19. We focus on the ‘500k’ dataset, as this was the only dataset where each tool successfully completed a test. Similar results were observed on the larger datasets, for the tools that still managed to produce an output.
We define the consensus Ci,j between tool i and tool j as the proportion of guides produced by i that are also produced by j. Note that the value is not symmetric: while the intersection between the two set of guides is unchanged, the denominator for Ci,j and Cj,i is the number of guides produced by i and by j, respectively.
There is no ideal, absolute value for the consensus level. If two tools do not filter targets and simply report all candidate sites, one would expect a perfect consensus. This says nothing about the quality of the targets, but confirms that the tools behave as expected. On the other hand, for tools that actively filter targets, it is reasonable to expect a lower consensus level, since they use different criteria to select targets. The consensus level is a first layer of analysis, which needs to be complemented by experimental data.
The consensus matrix C is shown in Fig 1A, and highlights that most tools do not filter targets. They report all possible targets, sometimes with a score. This leads to a high consensus between methods, typically 95% or more. The CHOPCHOP method, on the other hand, checks that a guanine is present at the twentieth nucleotide. If this is enforced, it reports only about a quarter of the targets that other tools produce. CRISPR-DO removes those that contain poly-thymine sequences and those with undesirable off-target effects. PhytoCRISP-Ex only considers as potential targets the guides that satisfy two rules: (i) they have at most two off-target sites with only one mistmatch anywhere else in the input genome, and (ii) their seed region (last 15 bases including the PAM sequence) is unique. These potential guides are then checked for the presence of restriction sites for pre-selected, common restriction enzymes. The consensus of WU-CRISPR with other tools was a maximum of 34.6%. This was due to the tool applying pre-filters to eliminate guides before being evaluated by the SVM model. Some of the pre-filters included: folding and binding properties as well as position specific guide contents. Following this, the tool only reports guides which match the model.
Five tools (CHOPCHOP, Cas-Designer, mm10db, CCTop and CRISPR-DO) focus, or give the option to focus, on identifying targets located on exons. Fig 1B shows the consensus matrix for these regions, across all tools. As before, there is a high consensus between tools that do not reject targets. For tools that more strictly select which targets to report (CHOPCHOP and mm10db), the consensus is again around 15-30%. Interestingly, the consensus between CHOPCHOP and mm10db is low. This highlights that tools are using different selection criteria, and this leads to the identification of distinct targets.
Only one tool (mm10db) provides a detailed report identifying reasons for rejection. These reasons can be used to analyse the output of the other tools and gain some insight into their behaviour. It is worth noting that mm10db applies filters sequentially. For instance, a guide rejected for multiple exact matches in the genome might have also been rejected for its GC content if it had made it to that filter. The results are still informative. The main reasons were high GC-content (approx. 62%), a poor secondary structure or energy (19%), and multiple exact matches (16%). Off-target scoring is the last step so, despite being strict, it necessarily contributes to a small fraction of the rejections. Here, this is made even smaller (< 0.05%) because the small input size limits the risk of having a large number of similar off-target sites.
From Fig 1, we know that up to 83% of guides proposed by other tools are rejected by mm10db. For each tool, we also explored the rejection reasons given by mm10db. This is summarised in Fig 2 for the ‘500k’ dataset. Rejection due to GC content is the most significant reason, followed by secondary structure or energy.
Importantly, a number of guides (on average around 10%) proposed by these tools are rejected by mm10db due to multiple exact matches in the genome. For instance, guide CTCCTGAGTGCTGGGATTAAAGG is reported by 10 tools when analysing the ‘500k’ dataset, which contains the guide at five loci as perfect matches. This guide targets two distinct genes on chr19 alone: Prpf19 at chr19:10,907,131-10,907,153 and Gm10143 at chr19:10,201,455-10,201,477. Ideally, this guide would not appear in the output from a tool due to its lack of specificity: there is no practical application where such a guide would be useful. Furthermore, none of the tools that deal with specificity by reporting the number of sites with k mismatches (k ≤ 5) actually reported the correct number of off-targets for this guide. Cas-Designer and GT-Scan reported zero perfect matches. CHOPCHOP reported 50 perfect matches (Bowtie was used by CHOPCHOP for this task, however, was limited to report 50 alignments). Aligning this guide on the full chr19 using Bowtie (-v 0 -a), we found 241 perfect matches. PhytoCRISP-Ex did not report any guides that we later identified as having multiple matches in the genome. This is because, as mm10db, it is strict about filtering out guides that may have off-target partial matches.
Fig 2 also shows that, for tools that report all guides, roughly a quarter of their output has not been considered (i.e. neither accepted nor rejected) by mm10db. This is due to the regular expression it is using to extract candidates ([ACG][ACGT]20GG), which excludes candidates starting with a T, while other tools do not.
We then considered the performance of tools that either filter targets or provide a score which predicts efficiency. We created two artificial genomes, each containing guide sequences for which experimental data on their efficiency is available (see Data Preparation section). We refer to these artificial genomes based on their originating datasets: Doench and Xu. Note that here, we were only focusing on assessing whether efficiency is correctly predicted. Using an artificial input was therefore crucial. If we were scoring the guides against the whole genome, some could be rejected due to other factors (e.g. off-target considerations) that are not related to their predicted efficiency. A guide rejected because of the off-target risk would tells us nothing about the ability of the tool to predict whether the guide is efficient. By using an artificial genome, we can ensure that each guide has absolutely no off-target risk, and can therefore focus on the efficiency prediction. In this context, a perfect tool would accept all the guides experimentally shown to be efficient, and would reject all the inefficient ones.
The results are shown in Figs 3 and 4. In our analysis we considered the precision of the tools, which is the proportion of guides predicted to be efficient that actually were experimentally found to be efficient.
For the Xu dataset, CHOPCHOP accepted 273 guides, 84.3% of which were ‘efficient’, and mm10db accepted 330, 65.2% of which were ‘efficient’. Incorporating CHOPCHOP’s rule of having a G at position 20 into mm10db would bring its proportion of ‘efficient’ guides to 84.4%. The results also highlight the value in considering multiple selection approaches: only 25.1% of the ‘efficient’ guides identified by mm10db have also been selected CHOPCHOP.
The sgRNAScorer2 user manual states that scores should be used in a relative manner as opposed to absolute: the higher the score, the better the predicted activity, but there is no threshold to classify between predicted efficiency and inefficiency. The 50th, 75th and 90th percentiles used in Fig 3 are therefore for visualisation only. Instead, we extracted every pair (a,b) of guides where a was experimentally considered ‘efficient’ and b ‘inefficient’, and checked whether a had a higher predicted activity than b. This was true for 76.8% of them.
CRISPR-DO and WU-CRISPR did not report all the guides provided in the artificial genome. For the guides where a score is available, the tools had a precision of 87.3% and 81.8%, respectively.
The guides which FlashFry scored higher mostly appear in the efficient region in Fig 3. If we use the recommended threshold for the tool, 84.4% of the selected guides are efficient. Similarly, CRISPOR (which reports the Azimuth score [20]) has a precision of 77.4%, and SSC 85.1%. Overall, the result for these tools suggests that they are able to score guides with some accuracy.
We used the classification model for TUSCAN, which is reported to outperform sgRNAScorer2 and WU-CRISPR [27]. Here, we found 71.5% of the guides that TUSCAN accepted were experimentally shown to be efficient. Among the tools that filter targets (rather than scoring them), TUSCAN had the highest recall (i.e. proportion of efficient guides that are selected).
CRISPR-DO had the highest precision across all tools (87.3%), whilst Cas-Designer had the lowest (61.2%).
For the eleven tools considered, there is a very low overall consensus; only one guide in the entire dataset was selected by all of the tools.
For the Doench dataset, guides from nine transcripts were provided and only guides in the top 20% for each transcript were considered as efficient [28]. For this dataset, we found that the precision for each tool was significantly lower compared to Xu. This was expected because of the high threshold used to define efficiency in this study. Only two tools have a precision above 50%: WU-CRISPR, which was trained on this dataset, and FlashFry, which is also using a scoring method originally derived from this dataset. TUSCAN was partially trained on this dataset but has a precision of 24.5%.
All the other tools have a precision between 20.2% (Cas-Designer) and 30.4% (CRISPR-DO), as summarised in Table 2. Only two tools (mm10db and CHOPCHOP) have an accuracy above 65%.
Fig 4 shows the performance of each tool on this dataset. For tools which provide a score, we show their prediction as a function of the gene rank percentage. For FlashFry, guides with low gene rank percentages are frequently scored low. WU-CRISPR reports high scores for many guides with high gene rank percentages. TUSCAN accepts a large majority of efficient guides. For SSC and sgRNAScorer2, there is a very weak trend for higher scores as the gene rank is increasing. The Azimuth score (as reported by CRISPOR), CRISPR-DO and Cas-Designer are not showing any clear trend. For tools which accept or reject guides, we show the distribution of these two categories over the gene rank percentage. For mm10db higher ranked guides are more likely to be accepted, and lower ranked are more likely to be rejected. For CHOPCHOP, this is not as pronounced and for phytoCRISP-Ex there is no trend.
Again there is a very low overall consensus; only three guides in the entire dataset were selected by all eleven tools.
Even though it is often overlooked when methods are initially presented, their computational performance is an important consideration. There are a number of applications of CRISPR where it can be crucial: large input genomes, time constraints to obtain the results, large number of non-reference genomes to analyse, etc.
In terms of time requirements, our hypothesis was that any efficient scoring or filtering would scale linearly with the number of targets to process (since they are individually assessed), and therefore almost linearly with the input genome size. However, for the specificity analysis, each guide needs to be assessed against any other candidates, which could result in quadratic growth. Another possible limitation is the memory requirement of the tools.
Our results are shown in Table 3 and Fig 5. Only five tools successfully completed the four tests: CasFinder, CRISPR-ERA, mm10db, GuideScan and CRISPR-DO. GuideScan just reports all possible targets, providing no scoring on predicted efficiency, and was therefore expected to be fast. CasFinder and CRISPR-ERA provide a score, but it was identical across the dataset and was therefore not informative. On the other hand, mm10db runs candidate guides through a number of filters and provides a meaningful score for the off-target risk, so its speed is worth noting.
CHOPCHOP saturated both physical memory and allocated swap space (as monitored by SBS), resulting in the out-of-memory killer (OOMK) terminating its run on the full chromosome 19 (approx. 2% of the mouse genome). sgRNAScorer2 was extremely slow, taking more 4 hours to process the 500k dataset (when the other tools took between 5 seconds and 21 minutes). It did not manage to process the 5m dataset in less than 3 days. FlashFry was performing well in the tests below ‘full’, however, the Java Virtual Machine was exhausted of memory when completing the final test.
Four other tools failed to satisfy the time constraint, all on the full chromosome 19 (Cas-Designer, CT-Finder, GT-Scan and CCTop). CRISPOR, GT-Scan and TUSCAN had memory problems similar to those of CHOPCHOP and were terminated on that same dataset.
As part of our analysis we also considered the use of multi-threading. CHOPCHOP, GuideScan, mm10db, FlashFry and sgRNAScorer2 are the only tools which have implemented multi-threaded routines. CHOPCHOP and GuideScan do not allow the user to specify the number of threads to spawn, but instead, spawn threads according to the number of CPU cores. mm10db provides command-flags to specify the number of threads for itself and Bowtie; we specified 128 and 8 threads, respectively. Cas-Finder, CT-Finder and CHOPCHOP utilise Bowtie in single-threaded mode.
We ran Cas-Designer in GPU mode and found this provided a performance boost. However, this was still not enough to allow the full dataset to be processed within the time limit. Ideally, future tools (or future versions of the tools benchmarked here) should also leverage GPU resources.
SBS monitored the physical memory and allocated swap space usage. CRISPOR, CHOPCHOP, GT-Scan and TUSCAN saturated both of these memory spaces, resulting in the out-of-memory killer (OOMK) terminating each tool on the full dataset test. Cas-Designer saturated the swap space, however did not trigger the OOMK in the same test.
A large number of tools have been proposed to assist in the design of CRISPR-Cas9 guides. Many of them, three from our study in particular, have been successfully used experimentally: CasFinder has been used for designing guides from the human genome to target specific genes [29], CRISPR-ERA has been used for designing guides to target the HIV-1 viral genome [30] and mm10db has been used for whole-genome analysis of the mouse genome [6].
However, there has been little comparison of their performance. This paper address this gap by benchmarking 18 tools in terms of their computational behaviour as well as the guides they produce. Our results show that only five tools (CasFinder, CRISPR-ERA, CRISPR-DO, GuideScan and mm10db) can claim to analyse large inputs rapidly and would be readily available to process entire genomes, especially larger ones.
Many tools are not selective in designing guides: they report all candidates, and therefore have a high consensus between them. On the other hand, eleven tools provide clear predictions of guide efficiency, as listed in Table 2. We assessed their performance on two collections of experimentally validated guides. All eleven tools produced a majority of efficient guides for the Xu dataset but much fewer for the Doench dataset. For both datasets, there was limited overlap between tools. This has shown that the tools do have the ability to detect some efficient guides, but also that they are not completely precise and still have low recall. Improved methods are needed.
Taken together, the results suggest that mm10db has currently the best balance between speed and filtering/scoring of the guides for whole-genome analysis. When working on a small number of genes, the much-slower but more predictive CRISPR-DO can be preferred. CHOPCHOP had issues scaling to the full dataset, but would otherwise be a good choice as well. Importantly, the results also emphasise the need for further refinement of CRISPR-Cas9 guide design tools. This benchmark provides a clear direction for future work on optimising computational performance and combining multiple design approaches.
Given the range of rules and implementations, it is crucial to benchmark guide design tools and compare their performance. In this section, we describe how each tool is evaluated based on compute requirements, features and output.
The initial data from our benchmarking is based on the GRCm38/mm10 mouse genome assembly, available via the University of California, Santa Cruz (UCSC). We downloaded chr19, and extracted three datasets of increasing length: 500k, 1m, and 5m nucleotides, all starting from position 10,000,000. These datasets, and the whole chromosome, are used for testing.
For each of these four configurations, we created all the files required by any of the tools: custom annotation file (derived from the refGene table available via UCSC), 2bit compression file, Bowtie and Bowtie2 indexes, and Burrows-Wheeler Aligner file.
To complement these datasets, we have also used two collections of guides for which experimental data is available [28, 31]. One collection, Xu, contains 1169 guides which were used in a screening experiment, with 731 deemed to be ‘efficient’ based on analysis of the gene knock-outs. The other, Doench, contains 371 ‘efficient’ and 1470 ‘inefficient’ guides. Knowing the experimental quality of a guide can give further insight into the quality of computational techniques for evaluating guides.
We constructed two artificial sequences that contain the guides from each dataset, interspaced by 50 N’s to ensure that unexpected overlapping targets cannot be detected. As before, we generated all the supporting files required for this input.
The tools were not optimised for a specific organism, and the choice of mm10 for the initial tests, or of data from human cell lines for the experimental validation, should not impact the results.
Our datasets are available in S1 Dataset.
Our Software Benchmarking Script (SBS) tool is implemented in Python 2.7, and uses the Processes and System Utilisation (PSUtil, version >= 5.4.4) module for process-specific monitoring of system resources. When launched, the user is required to pass a bash command and an output directory via command-line flags. The audit routine begins after the bash command is executed. The parent process and all descendants are monitored at each polling event (PE).
The current wall-time, CPU and memory usage, disk interaction (DIO), number of threads and number of children are recorded. Wall-time was preferred to CPU-time as it is the human-perceived completion time. SBS reports the instantaneous resident set size (RSS) usage and virtual memory usage. DIO includes both the number and size of read/write operations. At each PE, an aggregate of the parent and child data is calculated and written to file. Additionally, the bash commands which launched each child process are logged.
The PE routine continues until the parent process ends or 72 hours is exceeded. This limit was imposed as we aimed to discuss tools with potential for whole-genome analysis, and those that cannot analyse chr19 (2.25% of mm10 total size) within this limit are deemed inappropriate for the task.
All tests were performed on a Linux workstation with Intel Core i7-5960X (3.0 GHz), 32 GB RAM, 32 GB allocated swap space, and Samsung PM87 SSD. We used Python v2.7 and Perl v5.22.1. This machine exceeds the specifications of some workstations, however, it is expected that a user would require a similar machine or better in order to achieve whole-genome analysis.
SBS is available on GitHub at https://github.com/jakeb1996/SBS.
Each tool had its own output format, so we normalised the results as: tool name, candidate guide sequence, start position, end position and strand (written to file in the CSV format). During this process, the start and end values were aligned with one-based positioning, as per the UCSC datasets. An ellipsis was concatenated to guides which were lacking a PAM sequence.
To determine which tools shared common guides, a script aggregated all non-duplicate guides and recorded which tool produced subsequent occurrences of the guide. A guide was considered a duplicate with a previously observed guide when the 3’ positions were equal and when they were reported to target the same DNA strand. A separate script analysed each normalised guide to determine whether it targeted a gene coding region (based on the UCSC annotation data).
|
10.1371/journal.pntd.0002617 | Phytol, a Diterpene Alcohol from Chlorophyll, as a Drug against Neglected Tropical Disease Schistosomiasis Mansoni | Schistosomiasis is a major endemic disease that affects hundreds of millions worldwide. Since the treatment and control of this parasitic disease rely on a single drug, praziquantel, it is imperative that new effective drugs are developed. Here, we report that phytol, a diterpene alcohol from chlorophyll widely used as a food additive and in medicinal fields, possesses promising antischistosomal properties in vitro and in a mouse model of schistosomiasis mansoni.
In vitro, phytol reduced the motor activity of worms, caused their death and confocal laser scanning microscopy analysis showed extensive tegumental alterations in a concentration-dependent manner (50 to 100 µg/mL). Additionally, phytol at sublethal doses (25 µg/mL) reduced the number of Schistosoma mansoni eggs. In vivo, a single dose of phytol (40 mg/kg) administered orally to mice infected with adult S. mansoni resulted in total and female worm burden reductions of 51.2% and 70.3%, respectively. Moreover, phytol reduced the number of eggs in faeces (76.6%) and the frequency of immature eggs (oogram pattern) was significantly reduced. The oogram also showed increases in the proportion of dead eggs. Confocal microcopy studies revealed tegumental damage in adult S. mansoni recovered from mice, especially in female worms.
The significant reduction in parasite burden by this chlorophyll molecule validates phytol as a promising drug and offers the potential of a new direction for chemotherapy of human schistosomiasis. Phytol is a common food additive and nonmutagenic, with satisfactory safety. Thus, phytol has potential as a safe and cost-effective addition to antischistosomal therapy.
| Schistosomiasis is an infectious parasitic disease caused by helminths from the genus Schistosoma, which affects hundreds of millions of people, mainly the poor. Despite schistosomiasis being one of the most prevalent and debilitating neglected tropical diseases, the treatment and control of this disease relies on a single drug, praziquantel. However, there is increasing concern about the development of drug resistance. For this reason, the search for new schistosomicidal agents is a priority. Here, we report that phytol, a molecule from chlorophyll widely used as a food additive and in medicinal fields, shows promising antischistosomal properties against adult Schistosoma mansoni in vitro and in laboratory studies with mice harbouring adult S. mansoni. Phytol is a common food additive and is nonmutagenic, with satisfactory safety. Thus, phytol has potential as a safe and cost-effective addition to antischistosomal therapy. Further studies are needed because our results might have public health relevance.
| Schistosomiasis still constitutes one of the biggest health problems in the world. This disease of poverty has proved difficult to control for centuries and consequently, it still affects hundreds of millions of people. Recent articles document infection of approximately 200 million people in more than 70 developing countries, with approximately 800 million, mostly children at risk of infection [1]. Additionally, the disease burden is estimated to exceed 70 million disability-adjusted life-years [2]. The causative agents of schistosomiasis are parasitic flatworms of the genus Schistosoma. Three species (Schistosoma mansoni, Schistosoma haematobium and Schistosoma japonicum) account for the majority of human infections. The major aetiological agent of human schistosomiasis is S. mansoni and intestinal schistosomiasis caused by this species is present in Africa, the Middle East, the Caribbean, and South America. Typically, the morbidity associated with schistosomiasis results from the immunological reactions launched in response to parasite egg deposition in the liver and other host tissues [3].
Despite the public health importance of schistosomiasis and the risk that the disease might further spread and intensify, schistosomiasis control programmes are based are based mainly on chemotherapy, which is limited to the anthelmintic drug praziquantel [4]. However, due to the widespread and intensive use of praziquantel, there is increasing concern about the development of drug-resistant strains [5], [6]. For this reason, the search for new schistosomicidal agents is a priority.
Plants have always been used as a common source of medicine, both for traditional remedies and in industrialised products [7], [8]. Chlorophylls, found in all green vegetables, constitute an important source of an isoprenoid component, phytol (3, 7, 11, 15-tetramethyl-2-hexadecen-1-ol) [9]. It is an acyclic monounsaturated diterpene alcohol, present in vitamin K, vitamin E, and other tocopherols. Phytol is an aromatic ingredient used in many fragrance compounds and it may be found in cosmetic and non-cosmetic products [10]. In medicinal fields, phytol has shown antinociceptive and antioxidant activities [11] as well as anti-inflammatory and antiallergic effects [12]. Recent studies have revealed that phytol is an excellent immunostimulant, superior to a number of commercial adjuvants in terms of long-term memory induction and activation of both innate and acquired immunity [13]. Additionally, phytol and its derivatives have no cumulative inflammatory or toxic effects even in immuno-compromised mice [14]. Phytol has also shown antimicrobial activity against Mycobacterium tuberculosis [15], [16] and Staphylococcus aureus [17].
Drugs of natural origin have already been used to treat parasitic diseases. In this regard, the search for antischistosomal compounds from natural sources, mainly from plants, has been intensified [18], [19]. We have been specifically interested in phytol since it has well-characterised mechanisms of toxicity, is structurally simple, easily available, and cost-effective. Additionally, phytol is a common food additive and, thus, should be well tolerated by the body [9], [10], [14].
In this paper, we describe the in vitro and in vivo schistosomicidal activity of phytol against Schistosoma mansoni for the first time. As a benchmark, praziquantel was also used in vitro. As a first step, in vitro antischistosomal studies were performed. Subsequently, a trial was designed to test the schistosomicidal activity of phytol in experimental schistosomiasis caused by S. mansoni in a mouse model. We also demonstrated and described the ability of phytol to induce severe membrane damage in schistosomes through the use of confocal laser scanning microscopy. Furthermore, the effects of phytol on pairing and egg production by adult worms were also examined.
Phytol (Fig. 1) was purchased from Sigma-Aldrich (St. Louis, MO, USA) and praziquantel tablets were purchased from Merck (São Paulo, SP, Brazil). For in vitro studies, drugs were dissolved in dimethyl sulfoxide (DMSO, Sigma-Aldrich) to obtain stock solutions of 4 mg/mL. For in vivo studies, phytol was suspended in 3.7 mL of phosphate buffered saline (PBS) and orally administered at final concentration of 40 mg/kg.
The Belo Horizonte strain of Schistosoma mansoni was used in all experiments. The parasite life-cycle is maintained in the laboratory by routine passage through a rodent host and intermediate snail host Biomphalaria glabrata [19]. Infections of rodent host with S. mansoni were initiated by subcutaneous injection of approximately 150 cercariae. Cercariae were harvested from infected snails by exposure to light for 3 h, following standard procedures of our laboratory [19].
For in vivo studies, 3-week-old Balb/c mice were used. Mice were infected with 70 cercariae of S. mansoni by tail immersion and kept under environmentally controlled conditions (temperature, 25°C; humidity, 70%) with free access to water and rodent diet [20].
The present study was approved by the Ethics Committee at Universidade Federal do Piauí, PI, Brazil (approval number 013/11) and Universidade Estadual de Campinas, SP, Brazil (approval number 2753-1). All the animals were handled in strict accordance with good animal practice as defined by the Universidade Federal do Piauí and Universidade Estadual de Campinas guidelines for animal husbandry, according to with the Brazilian legislation (Comissão de Ética de Uso de Animais, CEUA, 11,794/2008).
To observe morphological changes in the tegument of adult parasites after in vitro and in vivo assays, schistosomes were monitored using a confocal laser scanning microscope following standard procedures presented elsewhere [19]. Briefly, at the end of the drug treatment period (120 h) or in the case of death, the parasites were fixed in a formalin-acetic acid-alcohol solution (FAA) and analysed under a confocal microscope (Laser Scanning Microscope, LSM 510 META, Carl Zeiss, Standorf Göttingen, Vertrieb, Germany). Autofluorescence was excited with a 488-nm line from an Argon laser, and emitted light was collected with 505 nm [30], [31].
For assessment of changes in the tegument of parasites, three-dimensional images obtained from confocal laser microscopy were used for a quantitative method. In this quantitative analysis, areas of the tegument of male worms are assessed, and the numbers of tubercles were counted according to standard procedures [19]. Briefly, during the microscopic analysis of the three-dimensional images captured using LSM Image Browser software (Zeiss), areas of the tegument of parasite are assessed, and the numbers of intact tubercles on the dorsal surface of male helminths were counted in a 20,000 µm2 area.
Statistical tests were performed with GRAPHPAD PRISM (version 5.0) software. Dunnet's test was used to analyze the statistical significance of differences between mean experimental and control values. Significant differences were also determined by applying Tukey's test for multiple comparisons. A P value of <0.05 was considered significant.
Adult S. mansoni worms (56-day-old) were cultured in RPMI 1640 medium in the presence of phytol. The parasites were maintained for 120 h and monitored every 24 h to evaluate their general condition: motor activity, changes in pairing, egg production, alteration in the tegument, and mortality rate.
Since phytol had antischistosomal effects in vitro, it was further investigated in vivo. Therefore, the efficacy of phytol was tested against the adult parasite life stage in an experimental mammalian host. Single 40 mg/kg oral dose of phytol was administered to S. mansoni-infected mice at 56 days postinfection. During this period, male and female worms had matured and paired, and eggs were found in the liver, intestine, and faeces.
Phytol is widespread in nature, especially because it occurs ubiquitously as a component of chlorophyll [9]. It is considered a common food additive and information about oral bioavailability of phytol in mice revealed that this drug is well absorbed (30–66% of the administered dose) [32]. Moreover, comprehensive toxicological data are available. For example, the acute oral LD50 of phytol in rats was reported to be greater than 10,000 mg/kg and it was also not considered mutagenic [10].
This study has highlighted S. mansoni as a possible new target for phytol. Initially, we examined its antischistosomal activity on adult worms in vitro and the results encouraged us to examine its efficacy in mice harbouring adult S. mansoni. In vitro assays demonstrated that phytol affected parasite motility, viability, and egg production and it induced severe tegumental damage in schistosomes. Additionally, various parasitological criteria indicated the in vivo antischistosomal effects of phytol: it caused significant reductions in worm load, faeces egg load, and the frequency of egg developmental stages. To the best of our knowledge, we have, for the first time, evaluated the activity of phytol against the laboratory model S. mansoni in vitro and in vivo.
In general, our in vitro experiments on adult schistosomes confirmed the promising in vivo results. Indeed, the in vitro bioassay revealed that phytol acted preferentially against female rather than male worms. Likewise, the effect on worm burden of a single 40 mg/kg oral dose of phytol administered to mice harbouring a 56-day-old adult S. mansoni infection clearly showed that females were more susceptible than males. A similar variation in drug susceptibility between male and female schistosomes both in vitro and in vivo, has been observed with several antischistosomal drugs. For example, Mitsui et al., 2009 [33] reported that female worms of S. mansoni were often more susceptible than males to artenusate in vitro. Keiser et al. (2009) [34] described that female adult worms were more affected by mefloquine than male adults when the drug was administered orally to mice infected with adult S. mansoni. A similar finding was previously reported by Botros et al. (2003) [35] when testing the activity of the acyclic nucleotide analogue 9-(S)-[3-hydroxy-2-(phosphonomethoxy)propyl]adenine [(S)-HPMPA] against experimental schistosomiasis mansoni.
Unlike other recently described schistosomicides such as the cysteine protease inhibitor K11777 [36] or the oxadiazoles [37], which thus far have only been tested intraperitoneally and in multiple doses, phytol at a single oral dose resulted in worm burden reductions. Although our results were moderate with respect to total worm burden reduction (51.2%), high female worm burden reductions (70.3%) were observed in S. mansoni-infected mice treated with phytol. Besides, in contrast to the effects of the recognised antischistosomal drug praziquantel, which kills adult schistosomes (male and female), the killing effect of phytol is weaker. In this respect, in vitro, all adult worm pairs were separated into individual male and female worms after 24 h of incubation with phytol at concentrations of 25 µg/mL and above. Male and female worms were unable to embrace and mate and remained separated, and 100% of the female and male worms were dead after 24 h of incubation with phytol at concentrations of 50 and 100 µg/mL. A single oral dose of phytol administered to mice did not cause a significant reduction in the load of unpaired male worms. Nevertheless, treatment with phytol resulted in the death of most of the S. mansoni unpaired females (no unpaired female was recovered from five mice; one unpaired female was recovered from four mice; and four unpaired females were recovered from one mouse). The reduction of coupled worms and total load of worms may be due to the decrease in the number of female schistosomes.
Since the tegument of schistosomes is an important target for antischistosomal drugs, alterations in the surface topography of schistosome worms were used by several investigators for the evaluation of antischistosomal drug activities in vitro and in vivo [e.g. 25], [38]–[42]. In our in vitro assays, confocal laser scanning microscopy revealed that phytol induced severe tegumental damage in both male and female schistosomes. Additionally, quantitative analysis showed that phytol caused changes on the tubercles of S. mansoni male worms in a dose-dependent manner. Comparable results were obtained by previous works using other antischistosomal compounds, such as piplartine [22], dermaseptin [26], and (+)-limonene epoxide [43]. Based on our in vitro analysis, the anatomical disturbance differed between the reference drug (praziquantel) and phytol. Praziquantel caused severe muscle contractions and the worms became partially curved or swirled. In contrast, phytol caused worm paralysis but not muscle contraction.
Also, the present observations of alteration in the surface architecture of S. mansoni male worms as a result of treatment with phytol are not similar to the tegumental alterations seen in vitro. In this sense, at the confocal microscopic level, the male parasites recovered from mice 48 h after phytol treatment did not show significant morphological alterations, although blebbing was visible on the tegument of male worms. Blebbing is an indicator of stress and has been observed in previous studies evaluating antischistosomal drugs such as carvacryl acetate [23], mefloquine [38], miltefosine [39], artesunate and praziquantel [44], [45]. The discrepancy between the effect of the drug and the onset of action of phytol in vitro and in vivo might be related to the lower phytol concentrations present in the liver and mesenteric veins in mice compared with the in vitro model. Moreover, these differences between in vitro and in vivo results may be explained by the fact that in vitro, the parasite is in a direct contact with the drug and, thus, it is not in direct contact with the host's microenvironment. Pharmacokinetic studies, measuring drug concentrations in the body and the target organs, might aid in the elucidation of these differences observed in vitro and in vivo.
Furthermore, in contrast to adult male schistosomes, which have some blebs on the tegument, a single oral dose of 40 mg/kg phytol resulted in extensive tegumental damage of female worms. These results confirm, at least partially, that phytol is orally bioavailable [10], but possibly the oral dose must be increased to achieve tegumental damage in male worms. We speculate that there is either sex-specific interference of the drug with the target, or that there are different targets for phytol in females compared with males. Nonetheless, further research is needed to provide a better understanding of the schistosomicidal action of phytol.
Tegumental damage may not always result in death [46], but the morphological alterations observed in this study could be a mechanism through which phytol kills the worms. The damage to the tegument along the worm's body would have impaired the functioning of the tegument and also destroyed the defence system of the worm, and so it could easily be attacked by the host's immune system. Further studies are necessary to elucidate the multiple mechanisms of action of phytol, which seem to be involved in the killing of schistosomes. It might also be useful to investigate whether phytol acts synergistically with the host immune response, similar to the chemotherapeutic effect of praziquantel, which has been shown to be dependent on the host antibody response [47], [48]. On the other hand, it is possible that phytol has a direct killing effect without the absolute need for antibody, which occurs with praziquantel [49].
Finally, to see whether phytol affects the sexual fitness of adult worms, we evaluated the number of eggs in vitro and development stages (oogram pattern) and faeces egg load in S. mansoni-infected mice. Reductions in worm recovery and egg density in treated mice and in vitro were observed; this is considered by several authors as strong evidence of the efficiency of antischistosomal drugs [e.g.], [ 19], [39], [40], [42], [50,51]. The reduction of egg load in the tissues and faeces in treated mice may be attributed to the reduction in worm burden as a result of phytol treatment, the low productivity of the female already present, and the active destruction by the host's tissue reaction of the few eggs produced. In vitro, phytol caused a 75% reduction in egg production compared with untreated worms, although it is known that in vitro egg production is spontaneously reduced after a few days in culture [24]. However, more importantly, the inhibition of oviposition was irreversible, as found by examination of the worms following washing and incubation in drug-free RPMI medium, whose effect has been reported in previous studies with other antischistosomal compounds such as epiisopiloturine [25], piplartine and dermaseptin [24]. In vivo, significant alterations in oogram patterns and faeces egg load were found and, thus, phytol affected the fecundity of the worms and/or the viability of the eggs. As described by Sanderson et al. (2002) [52] for in vitro and in vivo studies on the bioactivity of a ginger extract, it is not known whether these anti-fecundity effects were the result of generalised cytotoxic damage or more specific inhibition of reproductive process by phytol. It is known, furthermore, that inhibitors of cholesterol synthesis such as lovastatin (mevinolin) can reduce egg laying by S. mansoni females. For example, Vanderwaa et al. (1989) [53] has demonstrated with lovastatin that egg production by S. mansoni, in vitro and in vivo, is associated with the enzyme hydroxymethylglutaryl-coenzyme A (HMG-CoA) reductase, and that cholesterol precursors, mevalonate and farnesol, stimulate egg production by the female parasite and can reverse mevinolin-induced inhibition of egg production. Subsequently, Chen et al. (1990) [54] demonstrated that mevalonate and/or its metabolite not only plays a vital role in schistosome egg production but is also vital for survival of the parasite.
Importantly, recent investigations have demonstrated that phytol is a cholesterol-lowering agent [55]. Accordingly, we speculated that the reduction in oviposition by phytol may be associated with the inhibition of HMG-CoA reductase. However, the underlying mechanism(s) of the effects remains to be fully elucidated. Presumably, it may also be due to a direct assault on female worms, thus diminishing their numbers or their ability to lay eggs, although a direct ovicidal action cannot be excluded [35]. Results obtained from preliminary morphological investigations (no data shown) indicate that phytol exerts a rapid action on schistosomes, resulting in marked alterations of the reproductive system of the worms.
The main targets of the action of phytol are the female worms in terms of either the load of female worms or their ability to lay eggs. Egg production is contingent on worm maturation, pairing, and the support of the metabolic needs of the female. Phytol clearly disrupted this development process by directly killing female worms or inhibition of oviposition. In schistosomiasis, reductions in worm burden are associated with reduced pathology, and there is no concern about relapse because schistosome parasites do not multiply in the mammalian host. Moreover, reduction in egg burden can reduce egg shedding and the potential for parasite and disease propagation.
In conclusion, the present results suggest that phytol has antischistosomal activities and provide a basis for subsequent experimental and clinical trials. The low toxicity and high bioactivity and tolerance by mammals support the potential of phytol as a new lead compound for human schistosomiasis. However, the effect of phytol in both in vitro and in vivo studies was evaluated using S. mansoni adult worms; thus, further studies are needed to evaluate the efficacy of phytol in different therapeutic regimens (e.g., multiple oral doses) and to evaluate the efficacy of this drug against different life-cycle stages (e.g., schistosomula and juvenile worms) as well as other Schistosoma species. Additionally, the detailed mechanism of action of phytol on schistosomes remains to be investigated.
|
10.1371/journal.pgen.1000651 | Global Analysis of Extracytoplasmic Stress Signaling in Escherichia coli | The Bae, Cpx, Psp, Rcs, and σE pathways constitute the Escherichia coli signaling systems that detect and respond to alterations of the bacterial envelope. Contributions of these systems to stress response have previously been examined individually; however, the possible interconnections between these pathways are unknown. Here we investigate the dynamics between the five stress response pathways by determining the specificities of each system with respect to signal-inducing conditions, and monitoring global transcriptional changes in response to transient overexpression of each of the effectors. Our studies show that different extracytoplasmic stress conditions elicit a combined response of these pathways. Involvement of the five pathways in the various tested stress conditions is explained by our unexpected finding that transcriptional responses induced by the individual systems show little overlap. The extracytoplasmic stress signaling pathways in E. coli thus regulate mainly complementary functions whose discrete contributions are integrated to mount the full adaptive response.
| Bacteria possess various signaling systems that sense and respond to environmental conditions. The bacterial envelope is at the front line for most external stress conditions; its components sense perturbations and transmit signals to induce transcriptional reprogramming, leading to an adaptive response. In Escherichia coli, at least five response pathways, called Bae, Cpx, Psp, Rcs, and σE, are induced in response to envelope stress. To date, these pathways have been studied mainly individually, and the interconnections and/or overlaps between them have not been extensively characterized. The present study establishes two important characteristics of stress response in E. coli: first, that a given stress solicits the combined responses of several pathways; second, that each individual pathway controls a discrete set of genes involved in the response, and shows little overlap with other pathways. Based on previous knowledge and the present data, we propose that an environmental stress probably impacts on the cell envelope by inducing numerous alterations, each of which may be perceived by different pathways of the stress response and contributes to adapting the cell to different aspects of the stress damage. The extracytoplasmic stress signaling pathways in E. coli thus regulate mainly complementary functions whose discrete contributions are integrated to mount the full adaptive response.
| Bacteria possess various stress signaling systems that sense and respond to specific stimuli and allow the cell to cope with changing environmental conditions. One or several stress stimuli may activate multiple stress response pathways to constitute an integrated and complex response. Adaptation to envelope stress illustrates the complexity of these regulatory networks.
The bacterial envelope is involved in necessary processes including nutrient transport, respiration, secretion, adhesion, virulence and maintenance of bacterial integrity. In Gram negative bacteria such as Escherichia coli, the envelope comprises an inner membrane, a periplasmic space that contains the cell wall, an outer membrane and bacterial appendages such as pili and flagella. Being in contact with the external medium, the envelope is the initial target of physical (e.g., hyperthermia, osmolarity), chemical (e.g., ethanol, pH, detergent) or biological (e.g., adhesion, infection) stresses that may alter envelope components, thus inducing an extracytoplasmic stress response. The E. coli σE, Psp, Cpx and Bae signaling pathways are the main elements of this response described to date (reviewed in [1]). The σE and Psp (phage shock protein) pathways are both regulated via sequestration and release of a transcriptional factor in response to specific signals: Accumulation of specific misfolded outer membrane proteins (OMP) within the periplasm induces sequential regulated intramembrane proteolysis (RIP) events leading to degradation of the inner membrane protein RseA, the σE sequestrator [2]–[4], and resulting in σE release in the cytoplasm. Free σE associates with RNA polymerase to allow σE -regulated gene transcription. PspF is a σ54 enhancer binding protein: In the absence of signals, PspF-enhanced transcription is inhibited by PspA binding to PspF [5]. According to the current model, one or both inner membrane proteins PspB and PspC sense the inducing signal (possibly a decrease of proton motive force) and then bind PspA, disrupting its interaction with PspF (reviewed in [6]). PspA, PspB and PspC thus act as regulators and effectors of the Psp response [7],[8], although another cascade might also exist [9]. The two other signal transduction pathways that respond to extracytoplasmic stress, Cpx for conjugative plasmid expression (for a review, see [10]) and Bae for bacterial adaptative response [11], are classical two component regulatory systems. Upon stimulation, the sensor (CpxA or BaeS) autophosphorylates a conserved histidine residue of its transmitter domain. The phosphoryl group is then transferred to a conserved aspartate of the receiver domain of the response regulator (CpxR or BaeR), resulting in its activation. In the absence of signals, sensor proteins are thought to function as phosphatases to deactivate their phosphorylated effector proteins. Additional proteins can participate in signal transduction prior to the sensor step: For example, the outer membrane lipoprotein NlpE stimulates CpxA following bacterial adhesion [12],[13], whereas the periplasmic protein CpxP inhibits CpxA autokinase activity in the absence of signal [14]. In the presence of an extracytoplasmic stress such as accumulation of P pili subunits, CpxP is titrated away from the CpxA periplasmic domain and degraded, together with bound misfolded proteins, by the periplasmic protease DegP [15]. P pili accumulation also induces the Bae pathway [11].
The Rcs system is a complex phosphorelay signaling pathway that also participates in the extracytoplasmic stress response. Initially described as a regulator of colanic acid capsule synthesis [16], mutational analyses later showed that the Rcs regulon also affects envelope composition [17],[18]. Recently, Rcs phosphorelay was shown to be activated by stresses affecting the peptidoglycan layer, and to contribute to intrinsic antibiotic resistance [19]. Rcs phosphorelay was also proposed to sense the extent of phosphorylation of the undecaprenyl carrier lipid, which is also involved in colanic acid synthesis [20],[21]. The Rcs pathway presents several differences as compared to classical two-component systems: RcsC is a hybrid sensor kinase having both a classical histidine kinase transmitter domain and an additional receiver domain with a conserved aspartate. Phosphate transfer from RcsC to RcsB is mediated by RcsD (formerly, YojN), a histidine-containing phosphotransmitter (Hpt). Finally, RcsB, the transcriptional regulator, utilizes an auxiliary cytoplasmic protein, RcsA, to regulate expression of some genes (for reviews, see [21]–[23]). Targets of the Rcs regulon can thus be classified as RcsA-dependent or RcsA-independent. RcsA is degraded by the Lon protease. Its instability has significant regulatory consequences, since the amount of RcsA is generally low. Formation of the RcsA-RcsB heterodimer protects RcsA from proteolysis and leads to transcriptional activation of RcsA-dependent genes and consequent capsule production.
The five pathways described above regulate chaperones, peptidyl-prolyl cis-trans isomerases, periplasmic disulfide isomerases, proteases that are involved in the folding or degradation of misfolded proteins, and also biosynthetic pathways for envelope components (reviewed in [24]). Together, these five envelope stress response systems are involved in the biogenesis, maintenance, and repair of the bacterial envelope, and thus; contribute to cell surface integrity. In addition, they modulate key bacterial life functions, such as motility (e.g., [25]–[27]), colony and biofilm formation (e.g., several cpx, psp, and σE regulated genes are induced during biofilm formation [28]; also see [12], [28]–[32]), conjugation [33],[34], stationary phase adaptation (e.g., [25], [35]–[38]), or virulence (e.g., [6], [36], [39]–[44], for reviews see, [1],[6],[45]).
The five envelope stress response systems have mainly been investigated individually. To gain further insights into the specificity of each pathway and their possible interconnections, here we compare the conditions leading to induction of each of the five pathways, and investigate the global transcriptional responses in parallel. This constitutes the first fully integrated transcriptomic study of extracytoplasmic stress response in E. coli.
Although molecular mechanisms leading to signal transduction and extracytoplasmic stress responses are well documented for several systems, the environmental conditions that act as natural inducers remain obscure. For each stress response system, several genetic defects or different treatments were shown to induce adaptive responses (i.e., Bae [11],[46]; Cpx [47]; Psp [6], [48]–[50]; Rcs [21],[22]; σE [4],[47],[50]). Some conditions are known to concomitantly activate several pathways. For instance, indole induces the Bae and Cpx pathways [11], ethanol, verapamil (calcium channel inhibitor), or dibucaine (an amide local anesthetic that alters membrane fluidity) induce the σE and Psp pathways [50], and antibiotics targeting penicillin binding proteins induce Rcs, Bae, Cpx and σE [19]. However the effects of activating signals have never been simultaneously compared for all five systems.
To determine whether a given stress condition could be specific to a single pathway, we investigated the impact of several stress conditions on activation of the five known extracytoplasmic stress response pathways. To do this, we first used strains with transcriptional gene fusions that place the lacZ reporter gene (encoding β-galactosidase) under the control of a promoter representative of each system: Bae, Cpx, Psp, Rcs and σE pathways were monitored using spy::lacZ, cpxP::lacZ, pspA::lacZ, rprA::lacZ and P3rpoH::lacZ, respectively (Table S1).
Strains were grown under various stress conditions using both external and genetic stimuli, and β-galactosidase activity was determined (see Materials and Methods). In the absence of their corresponding transcriptional regulators, cpxP::lacZ and pspA::lacZ fusions had no detectable activity and rprA::lacZ had strongly reduced activity when compared to the wild type background (data not shown). In the case of the σE reporter, P3rpoH::lacZ, σE is essential [51], and a basal level of β-galactosidase activity was observed in the unstressed condition.
Since the spy::lacZ fusion is dependent upon both BaeR and CpxR, we also analyzed the effects of these signals in a baeR background [52]. To our knowledge, no genes subjected only to baeR regulation have been described. We chose an rprA::lacZ reporter to monitor the Rcs pathway. rprA encodes a small regulatory RNA that stimulates rpoS (encoding σS) translation [53]: This gene was used rather than a cps::lacZ fusion, since the latter fusion is also RcsA-dependent, and thus reports as much on the level of RcsA (which is potentially limiting) as on activation of RcsB [53].
None of the external tested stimuli were found to be strictly specific for a single signaling system (Table 1). Indeed, 5% ethanol and 4 mM indole induced all the transcriptional fusions tested. Dibucaine activated the Cpx, Rcs, Bae, and Psp systems whereas 0.6 M NaCl activated only the Rcs and Psp pathways. However, we observed differences in the response levels of reporter fusions to different signals: Cpx, Bae and σE pathways were induced preferentially by indole and ethanol, Rcs by NaCl in addition to ethanol and indole, and Psp by ethanol and dibucaine, in keeping with previous results [50]. Interestingly, the Rcs pathway was induced in all membrane-altering stress conditions tested, in accordance with a bona fide role of this pathway in extracytoplasmic stress response.
During a genetic screen using an E. coli genomic DNA library [54], we observed that overexpression of yedR (encoding a putative integral inner membrane protein conserved in enteric bacteria) led to a strong mucoid phenotype. This could indicate that YedR is a component of the Rcs pathway, or alternatively, that it acts as an internal inducer of the Rcs response. Further analysis showed that induction of capsule production by yedR depended on RcsB and RcsC proteins (data not shown). However, the rprA::lacZ fusion was responsive to different stresses (0.5 mM dibucaine, 3% ethanol or 0.6 M NaCl) in a ΔyedR background (data not shown), leading us to conclude that YedR is not part of the Rcs signal transduction pathway. Furthermore, multicopy yedR also strongly stimulated the Psp pathway, suggesting that YedR accumulation generates an envelope stress that is sensed preferentially by these two signaling systems (Table 1).
All the above results show that stress activates multiple pathways. Nevertheless, some signal are considered as specific activator of pathways: for example, e.g., exposure of the C-terminal part of certain OMPs to the PDZ domain of DegS activates σE [3],[55],[56], a drop of the proton motive force activates Psp [6], and accumulation of P-Pili subunits activates Cpx [13]; in the case of Rcs and Bae, specific signal sensing mechanisms remain to be identified. For Rcs, although it has been proposed that the signal could be a perturbation of the peptidoglycan [19], alteration in other envelope compartments can also be efficient inducers as previously reported [22] and illustrated by the impact of yedR overexpression (Table 1).
These observations may be reconciled, as the above tested conditions likely alter several aspects of bacterial envelope integrity, generating multiple signals that are in turn specifically sensed by different pathways. In addition, connections between stress response regulons could also account for indirect activation of some pathways.
In conclusion, one exogenous signal induces multiple defects by affecting different envelope components, which may be sensed by specific signal transducing mechanisms that activate all five pathways. It is expected that induction of all five extracytoplasmic stress responses is required to fully protect the cell against the variety of damages caused by a single stress.
To gain further insight into specifically regulated functions and interconnection between all five extracytoplasmic stress signaling systems, we carried out a global analysis of the general transcriptional responses following activation of each pathway. As discussed above, in view of the possible secondary signals generated by the use of inducers, we continued the study using an approach based on overexpression of each five pathway regulators. This methodology was previously used to characterize in detail the σE regulon [57]. We point out a possible limit to this approach, in cases where the regulator requires phosphorylation for its activity (i.e., for BaeR, RcsB, and CpxR); nevertheless, experimental evidence indicates that overproduction of such regulators can be effectively used in such studies, as a proportion of the molecules is phosphorylated in the absence of signal (e.g., [46],[58]).
baeR, cpxR, rcsB, pspF and rpoE were cloned under the control of the PLtetO-1 promoter in cloning vector pZE21, and the corresponding plasmids introduced into the wild-type strain MG1655Z1 (Table S1); a strain containing the plasmid with no insert was used as a control. Expression from the cloned genes was induced by addition of anhydrotetracycline (aTc). Since overexpression of CpxR [59], PspF (data not shown) and σE [35] was toxic, we determined the minimal aTc concentration that resulted in minimal cell toxicity. Accordingly, induction was carried out with 10 ng/mL aTc in exponential phase in LB medium at 37°C for 45 min, to limit indirect effects of regulator overexpression. Western blot experiments indicated that these conditions led to accumulation (approximately between three- to ten-fold) of CpxR, RcsB, and σE as compared to a strain having the control plasmid (data not shown; not determined for BaeR and PspF). Bacteria were harvested and total RNA was extracted for microarray experiments (Materials and Methods). In addition, 30 genes were selected to follow expression by qRT-PCR in six strains (overexpressing BaeR, CpxR, RcsB, PspF and σE, or having the control plasmid). The qRT-PCR tested genes were chosen because: i) they were not previously known to be modulated during one of the studied conditions; ii) they were expected from published studies to be modulated by overexpression of one of the studied regulators, but the predictions were not confirmed by our micro-array experiments or, iii) the statistical significance of our micro-array data for these genes was inconclusive.
As detailed below, the increased transcription of cpxR resulted in induction of only a subset of genes previously shown to belong to the CpxR regulon. We therefore complemented these experiments with a transcriptome analysis of a cpxR mutant relative to the wild type MG1655 parental strain in late exponential phase. Since the Cpx pathway is activated in this condition [52],[59],[60], this comparison was expected to give access to at least some genes of the cpxR regulon.
Principal components analysis (PCA) is a useful statistical technique that removes noise from complex data sets by reducing the dimensionality and helps discriminate the key factors of variations [80],[81]. PCA has proved useful for finding significant patterns in microarray analyses (see for complete explanation, see [82],[83]). Briefly, given m observations (the gene expression ratio) on n variables (our 6 different conditions), the goal of PCA is to find r significant variables, where r is less than n, to select the factors that best explain the observed variance in the observations. PCA was used to analyze our microarray data on the mean log-ratio measures obtained, each condition corresponding to a variable (see Materials and Methods). The first dimension accounting for 33% of the variance, could not discriminate between the conditions, whereas the second dimension axis separated CpxR, PspF and ΔcpxR conditions from σE, BaeR and RcsB conditions (data not shown). The third and fourth components (accounting together for 29% of the variance) discriminated σE overproduction and ΔcpxR conditions from BaeR/RcsB/CpxR/PspF overproduction conditions (Figure 1A). In the case of σE, both the size of the regulon and the nature of the regulation (i.e., by a σ factor or by transcriptional regulators) could account for this result, whereas in the case of ΔcpxR, the result might be explained by the difference in the experimental strategy (deletion vs. overproduction).
A hierarchical clustering that grouped together genes with similar expression patterns was performed on the set of genes differentially expressed in each of the five overexpression conditions (Figure 1B). Results were in agreement with those observed with PCA. The PspF response clustered with the CpxR response, while σE and RcsB responses were further away. But the main conclusion of this analysis is the striking specialization of each pathway, with very limited overlap between responses (Figure 1B). This is also revealed by a Venn diagram representation showing genes regulated by the extracytoplasmic pathways (Figure 2). Results of this analysis are unexpected, since redundancy is often proposed as an important property of robust networks. In addition, in the case of the extracytoplasmic stress response, redundancy was expected because many genes that are regulated by σE or BaeR are also regulated by CpxR. Furthermore, several genes regulated by PspF were also affected by CpxR (Figure 2). One surprising finding in our study is that the overproduction of CpxR had a limited effect on many known CpxR regulated genes, in sharp contrast with the situation in the case of RcsB. We propose that in many cases, CpxR acts in conjunction with other regulators, which are limiting in the absence of a stress signal. Hence, rather than controlling in itself specific genes, an important role of CpxR may be to amplify the response promoted by the other regulons. It should be noted that σS promotes transcription of the cpxRA operon [25] and CpxR can cross-talk with the EnvZ-OmpR response [84],[85]. Thus, CpxR may integrate diverse stimuli associated with growth and central metabolism [60]. In view of these analyses, and of some previous results (e.g., [58]), our results suggest that CpxR functions more as a modulator of the other extra cytoplasmic stress responses, especially σ54E, BaeR and PspF than as a stand-alone regulator.
σE appears to be the major regulator, with at least 69 transcription units affected. It is mostly in charge of envelope biogenesis maintenance, especially genes required for synthesis, assembly, and homeostasis of outer membrane proteins and lipopolysaccharides ([57], Table 2). The other responses are more limited and specialized in certain categories of envelope components. PspF is dedicated to maintenance of energy, and has an important role associated with the cytoplasmic membrane in prokaryotes. RcsB affects additional envelope structures such as capsular exopolysaccharide production and O-antigen ([30],[71], Table 3), and BaeR controls the production of several drug export systems that might be important to extrude toxic compounds during an extracytoplasmic stress ([46], Table 4). Hence, each system has its raison d'être in term of restoring various aspects of envelope physiology.
Given the high specialization of each pathway, genes that are regulated by several of these pathways are likely to play a crucial role in cell physiology in response to an extracytoplasmic stress. Although responses are mainly distinct, handful of genes were found to be in common between some of the pathways. For example, the trans-envelope protein components of the Tol-Pal system have a major role in maintaining envelope integrity, which, driven by proton motive force, bring the inner and outer membranes in close proximity [74],[86]. Genes encoding the Tol-Pal system were positively regulated by the σE, Cpx and possibly Psp pathways. In contrast, the tnaLAB operon was repressed in by these same pathways (Figure 2). This could be in relation with the fact that tnaA encodes tryptophanase, an enzyme that degrades tryptophan and generates indole, itself toxic to the cell and an extracytoplasmic stress inducer. lamB, encoding an outer membrane protein was also found to be repressed in several conditions (Figure 2), which may reflect the extra demands imposed on the cell for folding factors controlled by the extracytoplasmic stress regulons [63].
Extensive studies of stress response in E. coli have established the existence of several extracytoplasmic pathways, and suggest that expression of numerous genes are affected by more than one of these pathways. These findings suggest redundancy and raise questions concerning the reason for such a multiplicity of pathways. For the first time, we explored all five extracytoplasmic stress response pathways under comparable conditions in E. coli. We found that they can be activated simultaneously in response to exogenous or endogenous stimulation. Thus, although activation of a single pathway has been demonstrated experimentally using specific substrates or conditions, our results suggest that natural environmental stimuli provoke bacterial modifications that lead to multiple pathway responses.
To determine the contributions of each stress response pathway, we avoided the use of non-specific inducers, and opted for overexpression of each regulator. Transcriptome analyses show that induction of specific target genes via multiple pathways is an uncommon occurrence. Some genes can also be subject to cooperative regulation between different pathways, or to a cascade of pathway responses. In addition, the pathways might cross-talk through transcriptional regulation, as is the case between σE and σS, and possibly evoked by our results.
Each of for stress response systems (σE, Rcs, Psp and Bae) appears to be specialized in assuring a specific aspect of envelope biogenesis and maintenance, whereas CpxR might have a role as modulator of the response by integrating other endogenous signals. We conclude that all five pathways are needed to mount a full response to extracytoplasmic stress.
The E. coli strains and oligonucleotides used in this study are listed in supplementary tables S1 and S2. Several plasmids constructed for this study were derived from pZE21 that has a ColE1 replication origin, confers kanamycin resistance and has a PLtetO-1 promoter upstream of a multicloning site [87]. pZE21-baeR, expressing baeR under the control of PLtetO-1, was constructed as follows: The MG1655 baeR gene was PCR-amplified using oligonucleotides 450 and 451 (for oligonucleotide sequences, see Table S2). The resulting fragment was digested by KpnI and BamHI and cloned into KpnI-BamHI restricted pZE21-MCS. pZE21-cpxR, pZE21-pspF, pZE21-rcsB and pZE21-rpoE, containing cpxR, pspF, rcsB and rpoE were constructed using the same approach as for pZE21-baeR but with 452/453, 454/455, 456/457 and 458/459 oligonucleotide-pairs, respectively. Plasmid pGem-T-easy is a high copy number plasmid conferring ampicillin resistance (Promega, Madison, WI, USA). pGem-T-easy-yedR expressing yedR was constructed as follows: MG1655 yedR gene was PCR-amplified using oligonucleotides 143 and 144. The resulting fragment was digested by KpnI and BamHI and cloned into KpnI-BamHI restricted pGem-T-easy.
P1 vir-mediated transduction was carried out as described [88]. The chromosomal baeR gene was deleted by targeted gene substitution using a combination of two published protocols as described [89]. The baeR deletion was confirmed by PCR.
Cells were grown in LB broth or on solid LB containing 15 mg.mL−1 agar [90]. When necessary, antibiotics were added at the following concentrations: ampicillin 100 µg.mL−1, chloramphenicol 30 µg.mL−1, kanamycin 20 µg.mL−1, and tetracycline 12.5 µg.mL−1. Growth of strains containing pZE21-MCS or related plasmids was analyzed as followed: 5 µl of overnight cultures adjusted to OD600 of 0.3. Serial dilutions in M9 medium [90] were spotted on solid LB medium containing kanamycin and 0, 2, 10 or 100 ng.mL−1 of anhydrotetracycline (aTc), and incubated at 37°C for 24 h.
Plasmid preparations, DNA cloning and ligation, classical PCR amplification and DNA transformations were carried out according to standard protocols [90] and manufacturers' instructions. Northern blots (see Supplementary Table S2 for information on primers used for probe synthesis) were performed as previously described [90],[91] with 10 to 20 µg of RNA, except that hybridization was performed at 42°C using a NorthernMax prehybridization/hybridization buffer (Ambion, Austin, TX, USA) according to manufacturer's instruction. The ssrA gene was used as a reference to normalize RNA quantities in Northern blot experiments.
Overnight cultures were diluted in LB broth to an OD600 of 0.004. For parallel analyses of inducing and non-inducing growth conditions, cultures were incubated under agitation at 37°C using 96-well culture plates in a total volume of 1 ml. Various growth conditions were investigated: i) standard LB, ii) LB containing ethanol (3 or 5%), iii) 0.5 mM dibucaine, iv) NaCl (0.6 M), v) 5 mM EDTA or vi) indole (2 or 4 mM). The stock indole solution was prepared by dissolving indole in hot LB before use. After five hours, three samples of 200 µl of each culture were taken: One was used to measure OD600 in 96-well plates in the Biolumin (Molecular Dynamics) or Chameleon (Bioscan Inc., Washington DC, USA). The two others were used to evaluate β-galactosidase activity according to Miller's protocol adapted to 96-well plate assays [88]. For other β-galactosidase assays, cultures were prepared in individual tubes in 5 mL LB broth under agitation at 37°C. Strains containing pGem-T-easy and related plasmids were grown with 100 µg.mL−1 ampicillin. β-galactosidase activities were measured in duplicate from 200 µl samples taken at OD600 of 0.4 (exponential phase).
Transcriptome analysis of the effect of transient overexpression of extracytoplasmic stress response regulators was performed with three independent RNA preparations for each of the six biological conditions tested, namely pZE21 (control plasmid), pZE21-baeR, pZE21-cpxR, pZE21-pspF, pZE21-rcsB and pZE21-rpoE. Addition of aTc (for 45 min) to the medium resulted in a 16-, 166-, 225-, 13- and 19- fold increase of baeR, cpxR, pspF, rcsB and rpoE mRNA in strains containing pZE21-baeR, pZE21-cpxR, pZE21-pspF, pZE21-rcsB and pZE21-rpoE respectively, as compared to the reference strain containing the control plasmid, pZE21. Additionally, a ΔcpxR strain was included in the study, using four independent RNA preparations. To assess data reproducibility and minimize dye bias effects, one of the samples (two in the case of ΔcpxR) was measured with Cy3 instead of Cy5. To ensure robustness and comprehensiveness in data analysis, a reference design was used with an equimolar mixture of all the biological conditions serving as a baseline for the comparisons. Such a design does not require pre-definition of the subgroups for comparison, allows discovery of non-anticipated classes among the samples and is compatible with subsequent additional sampling. Strain MG1655 (Table S1) containing pZE21 and derivative plasmids were grown overnight in LB broth supplemented with kanamycin and diluted in 7 mL LB broth at an OD600 of 0.004. After two hours, expression from the PLtetO-1 promoter was obtained by addition of 10 ng.mL−1 aTc to the medium. After 45 minutes, 7 mL of cold absolute ethanol was added to bacterial cultures (at OD600 of about 0.4). Cells were then harvested by centrifugation for 15 min at 3000 g and stored at −80°C to prevent RNA degradation. For parental and cpxR strain transcriptome analysis, LB overnight cultures were inoculated at OD600 of 0.004. When, cultures reached an OD600 of 2 (late exponential phase), they were harvested in the presence of cold absolute ethanol and frozen at −80°C. The next steps were carried as for overexpression transcriptome experiments: Cells were lysed and RNA was extracted three times with an equal volume of acidic hot phenol and once with chloroform. RNA was ethanol precipitated, air dried and dissolved in water. RNA integrity was evaluated using RNA 6000 nano chips and the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA) according to manufacturer's instructions. RNA quality control was performed using user-independent classifiers as described [92].
Ten µg of total RNA from each biological condition and the RNA reference mixture were supplemented with RNA corresponding to known sequences to serve as an hybridization control (Spikes, Universal ScoreCard, GE Healthcare), reverse transcribed and labeled using the Superscript Indirect cDNA labeling System (Invitrogen, Carlsbad, CA, USA) according to manufacturer's instruction, except that purification steps were done using the QIAquick PCR mini column system (Qiagen, Hilden, Germany). Labeling efficiency and product integrity was checked according to [93]. For each condition, the hybridization experiment was performed against the reference sample; a mixture of 0.75 µg Cy3- and 0.75 µg Cy5-labeled targets was incubated at 95°C for 3 min in the presence of a 2× Hybridization Buffer (Agilent technologies, Santa Clara, CA, USA). Denatured targets were placed on an E. coli v2 whole genome array [63] (ArrayExpress accession: A-MEXP-1516, http://www.ebi.ac.uk/microarray-as/ae/) and hybridized for 17 hours at 60°C, in a rotating oven (6 rpm), using an Agilent hybridization chamber system. The hybridized slides were washed once for 10 min in 2×SCC/0.1% SDS at 50°C and once in 0.5×SCC/0.1% SDS at room temperature, then twice for 5 min in 0.1×SSC at room temperature. Any traces of water were eliminated immediately by air drying with ozone-safe dry air (“canned air”). Slides were scanned using a GenePix 4000B scanner (Molecular Devices, Sunnyvale, CA, USA) at 10-µm resolution. All slides were scanned using 100% laser power; PMT voltages were automatically adjusted using the Genepix Pro 6.0 software acquisition system to obtain maximal signal intensities with <0.005% probe saturation.
The resulting 16 bit images were processed using the GenePix Pro 6.0 image analysis software (v6.0.1.26). Data were processed using the MAnGO software [94], an R script that allows integrated analysis of two-color microarrays. Raw data were normalized using the print-tip loess method [95]. The average Log2 expression ratios were then calculated [log2(pZE21-geneX/REF)−log2(pZE21/REF) = log2(pZE21-geneX/pZE21), where geneX is the gene of interest in the condition of interest, REF the value obtained for X in the case of the pool of all conditions (reference sample), and pZE21 the value obtained in the condition corresponding to the vector alone (biological reference)] and used for all subsequent statistical analyses. MIAME-compliant data [96] were deposited in the ArrayExpress database (http://www.ebi.ac.uk/microarray-as/ae/) under the accession number E-MEXP-2139.
Gene functions were assigned using data from EcoCyc (http://ecocyc.org/) and Uniprot (http://www.uniprot.org/). Adjacent genes coordinately regulated, possibly involved in the same function and separated with a short distance with no apparent terminator were considered as belonging to a putative operon.
One microgram of total RNA was reverse-transcribed in a 30 µl final reaction volume using the High Capacity cDNA Reverse Transcription Kit with RNase inhibitor (Applied Biosystems, Foster City, CA, USA) following the manufacturer's instructions. For each sample, negative reverse transcription reaction was done to verify the absence of genomic contamination in subsequent q-PCR. Primer sequences (see supplementary Table S2) were designed using Primer Express 3.0 software (Applied Biosystems). BLAST searches were performed to confirm gene specificity and the absence of multi-locus matching at the primer site. SYBRGreen q-PCR reactions were performed using the ABI Prism 7900 HT sequence detection system (Applied Biosystems) in 384 well optical reaction plates. 3 µl of cDNA (5 ng/reaction), standard or water (no-template control) were used as template for q-PCR reactions with Fast SYBR Green PCR Master Mix (Applied Biosystems) and primers at 500 nM final concentration. Real-time q-PCR amplifications were carried out (95°C for 20 sec, followed by 40 cycles of 95°C for 1 sec and 60°C for 20 sec, and a final dissociation curve analysis step from 65°C to 95°C). Technical replicate experiments were performed for each biological triplicate sample. The amplification efficiencies of each probe were generated using the slopes of the standard curves obtained by a ten-fold dilution series. The efficiency of the q-PCR amplifications for all of the genes tested was higher than 90%. Amplification specificity for each q-PCR reaction was confirmed by the dissociation curve analysis. Determined Ct values were then exploited for further analysis.
The gene expression levels were analyzed using the relative quantification (delta-Ct method). 16 housekeeping genes were tested and GeNorm and Normfinder functions in Genex 4.3.8 (MultiD, Göteborg, Sweden) were used to select the most stable genes. The geometric mean of 5 housekeeping genes (dnaQ, glnD, pcnB, uvrB and gyrA) was used to normalize our samples. Data were analyzed with StatMiner 3.0.0 Software (Integromix, Madrid, Spain). Analyses were done with biological replicates and a relative quantification (RQ) value was calculated for each gene with the control group as a reference. RQ values were adjusted according to specific amplification efficiency. A p-value was computed using a moderated t-test to measure the significance associated with each RQ value. Variations were considered statistically significant when the p-value was <0.05 unless otherwise specified in the tables.
|
10.1371/journal.pgen.1001165 | Conjugative DNA Transfer Induces the Bacterial SOS Response and Promotes Antibiotic Resistance Development through Integron Activation | Conjugation is one mechanism for intra- and inter-species horizontal gene transfer among bacteria. Conjugative elements have been instrumental in many bacterial species to face the threat of antibiotics, by allowing them to evolve and adapt to these hostile conditions. Conjugative plasmids are transferred to plasmidless recipient cells as single-stranded DNA. We used lacZ and gfp fusions to address whether conjugation induces the SOS response and the integron integrase. The SOS response controls a series of genes responsible for DNA damage repair, which can lead to recombination and mutagenesis. In this manuscript, we show that conjugative transfer of ssDNA induces the bacterial SOS stress response, unless an anti-SOS factor is present to alleviate this response. We also show that integron integrases are up-regulated during this process, resulting in increased cassette rearrangements. Moreover, the data we obtained using broad and narrow host range plasmids strongly suggests that plasmid transfer, even abortive, can trigger chromosomal gene rearrangements and transcriptional switches in the recipient cell. Our results highlight the importance of environments concentrating disparate bacterial communities as reactors for extensive genetic adaptation of bacteria.
| Bacteria exchange DNA in their natural environments. The process called conjugation consists of DNA transfer by cell contact from one bacterium to another. Conjugative circular plasmids have been identified as shuttles and reservoirs for adaptive genes. It is now established that such lateral gene transfer plays an essential role, especially for the antibiotic resistance development and dissemination among bacteria. Moreover, integrons, platforms of mobile gene cassettes, have been instrumental in this phenomenon, through their successful association with conjugative resistance plasmids. We demonstrate in this study that the conjugative transfer of plasmids triggers a bacterial stress response—the SOS response—in recipient cells and can impact the cassette content of integrons. The SOS response is already known to induce various genome modifications. Human and animal pathogens cohabit with environmental bacteria, in niches which will favor DNA exchange. SOS induction during conjugation is thus most probably able to impact a wide range of genomes. Bacterial SOS response could then be a suitable target for co-treatment of infections in order to prevent exchange of antibiotic resistance/adaptation genes.
| Free-living bacteria commonly face changing environments and must cope with varying conditions. These adaptive strategies involve temporary physiological responses through various groups of genes gathered in regulons that are induced or repressed according to the surrounding conditions. This is the case for the quorum sensing regulon [1], [2], the stringent response and catabolite repression systems, which allow adjustment of gene expression according to the growth conditions [3]–[5]. In other instances, the only adaptive solution requires a genetic change, and bacteria have developed mechanisms that favour genome modifications either by transiently increasing their mutation rates, inducing re-arrangements, or lateral (horizontal) gene transfer (HGT). One of the better known responses of this kind is the trigger of the SOS regulon, which controls DNA repair and recombination genes [6].
SOS is a bacterial stress response induced when an abnormal rate of single stranded DNA (ssDNA) is present in the cell. ssDNA is the substrate for RecA polymerization. The formation of a ssDNA/RecA nucleofilament stimulates auto-proteolysis of the LexA repressor, leading to de-repression of genes composing the SOS regulon. The SOS response is triggered by the accumulation of ssDNA, for example when cells try to replicate damaged DNA, after UV irradiation or treatment with antibiotics (fluoroquinolones, β-lactams) or mitomycin C (MMC), a DNA cross-linking agent. In addition to these endogenous sources, ssDNA is also produced by several mechanisms of exogenous DNA uptake involved in lateral gene transfer, namely by conjugation, transformation and occasionally transduction.
Conjugation is indeed one mechanism of lateral transfer that leads to the transient occurrence of ssDNA in the recipient cell [7], [8]. The presence of anti-SOS factors in some conjugative plasmids, such as the psiB gene of R64drd and R100-1 [9], suggests that conjugative DNA transfer can induce SOS. In plasmid R100-1, psiB (plasmidic SOS inhibition) was shown to be transiently expressed during the first 20 to 40 minutes of conjugation [10], [11] from a ssDNA promoter [12], and inhibited the bacterial SOS response [13], [14]. Plasmids carrying psiB do not all express it at levels sufficient to alleviate SOS, as seems to be the case in F plasmids for instance [9], [10], [13], [15], [16].
Conjugation is a widespread mechanism in the intestinal tract of host animals where there is a high concentration of bacterial populations [17]–[22]). Lateral gene transfer plays a large role in the evolution of genomes and emergence of new functions, such as antibiotic resistance, virulence and metabolic activities in bacterial species [23].
Bacteria can also possess other internal adaptive genetic resources. Vibrio cholerae carries a superintegron (SI), that can be used as a reservoir of silent genes that can be mobilized when needed. Integrons are natural gene expression systems allowing the integration of an ORF by site-specific recombination, transforming it into a functional gene [24]. Multi-resistant integrons (RI) have been isolated on mobile elements responsible for the assembly and rapid propagation of multiple antibiotic resistances in Gram-negative bacteria through association with conjugative plasmids [25], [26]. An integron is characterized by an intI gene, coding a site specific recombinase from the tyrosine recombinase family, and an adjacent primary recombination site attI [27]. The IntI integrase allows the integration of a circular promoterless gene cassette carrying a recombination site, attC, by driving recombination between attI and attC [28]. The integrated gene cassettes are expressed from the Pc promoter located upstream of the attI site in the integron platform [29]. The discovery of integrons in the chromosome of environmental strains of bacteria, and among these the superintegrons (SI) mentioned above, has led to the extension of their role from the “simple” acquisition of resistance genes to a wider role in the adaptation of bacteria to different environments [30].
The dynamics of cassette recombination and the regulation of integrase expression are poorly understood. Recently, it was shown that intI is regulated by the bacterial SOS response [31]. Since SOS is now known to induce both the RI and V.cholerae integrase expression, an important issue is to understand when and where cassette recombination takes place and how the integrase inducing SOS response is activated.
Our objective was to determine if conjugative ssDNA transfer can trigger the SOS response, and to which extent this affects intI expression and cassette recombination. SOS induction in promiscuous environments can prepare bacteria to face the many threats they can encounter there. In order to understand regulatory networks existing between conjugation and its effect on integron content and cassette expression, we first addressed if conjugation induces SOS using reporter fusions in V. cholerae and E. coli. After quantifying expression from the V. cholerae intIA promoter using GFP fusions, we adopted a genetic approach to test integrase-dependent site-specific recombination in vivo.
We show that conjugative plasmid transfer generally induces the SOS response and up-regulates integrase expression, triggering cassette recombination. However, this is not the case when an anti-SOS factor (psiB) is expressed, as seen for some narrow host range conjugative plasmids isolated from Enterobacteria. We further show that this anti-SOS function prevents up-regulation of the SOS regulon in a host-specific manner after conjugation. We demonstrate that conjugative transfer is sufficient to trigger integron cassette recombination in recipient cells. This study outlines the connections between conjugative lateral DNA transfer, bacterial stress response and recombination of gene cassettes in integrons, and provides new insights into the development of the antibiotic resistance within a population.
During conjugation, plasmid DNA enters the host cell in a single stranded fashion [7], [8]. In order to test whether conjugation induces the SOS response in the recipient cell, we used reporter E. coli and V. cholerae strains carrying sfiA::lacZ (7651) and recN::lacZ (7453) β-galactosidase fusions, respectively. sfiA (cell division in E.coli) and recN (recombinational repair) genes belong to the SOS regulon of E.coli. We also identified a LexA binding box upstream of recN in V. cholerae. We confirmed that induction of SOS in these strains results in expression of the β-galactosidase (β-gal) enzyme (not shown). Table 1 summarizes the conjugative plasmids belonging to several incompatibility groups we used in this study [32]. The donor (DH5α) strain was recA- and ΔlacZ.
The conjugation rates of these plasmids were first measured at various time points after donor and recipient cells were mixed (Figure 1A). In E. coli, all plasmids conjugate approximately at the same rate so that nearly all recipients have received a plasmid after 60 min of conjugation. In V. cholerae transfer rates vary considerably, only 1 in 105 cells have received a plasmid after 4h of mating with R6Kdrd and R388, while RP4 has a transfer rate similar to that of E. coli (10−1 to 1). Neither R64drd nor R100-1 replicate in V. cholerae. In order to address whether R100-1 actually transfers from E. coli into V. cholerae, we used pSU19-oriTF plasmids containing the oriTF (72 bp) of plasmid F. Plasmid F does not replicate in V. cholerae and oriTF is 98% identical to oriTR100. The high oriTF transfer rate observed at 1h of mating confirms that plasmids F and R100-1 (and presumably R64drd) can indeed transfer into V. cholerae and that the lack of R100/R64 transconjugants is due to their inability to establish themselves in this bacterium.
SOS induction linked to conjugation was measured in the total recipient population by counting the actual number of recipient cells plated on selective medium instead of using OD units (Materials and Methods), to obtained an induction value per potential recipient cell. Mating was interrupted at various time points (t0, t40, t60, t120, t180, t240) and β-gal activity was measured in both E. coli and V. cholerae recipients (Figure 1B). The results are represented on the graph as the induction ratios at times t0, t60 and t240 over the induction at t0. When the recipient strain was mixed with empty donor, no SOS induction was observed. A peak of SOS induction in E .coli was detected after 40 min to 60 min of mating with a conjugation proficient donor, 1.7 fold induction for RP4 and R6Kdrd and 2.3 fold for R388. The induction peak was also observed in V. cholerae (2.3 fold for R6Kdrd, 2.7 fold for RP4 and 3.4 fold for R388). To verify that the β-gal activity was due to the SOS induction, we deleted the recA gene in the recipient E. coli strain. No induction of β-gal activity was observed in the ΔrecA strain after conjugation with RP4, R6Kdrd and R388. This confirms that the β-gal induction observed in recA+ strain indeed reflects the SOS induction by RP4, R6Kdrd and R388.
As described above, β-gal induction peaks between t40 and t60 minute of mating. The induction then decreases to reach the level shown at t240, forming bell shaped curves (data not shown). This induction pattern reflects the SOS induction in an asynchronous population of bacteria. It can be explained by the fact that plasmids RP4, R6Kdrd and R388 replicate in recipient cell. Once mating has started and as time goes by, there tends to be less plasmidless recipient cells. Indeed, entry of the plasmid DNA induces SOS, the incoming plasmidic ssDNA then replicates in the conjugant cell and the entry exclusion systems prevents entry of another plasmid [33], [34]. However, cells continue to divide so that the population of kanamycin resistant (kanR) host cells increases. Accordingly, even when the transfer rate remains constant (especially for low rated plasmids), the increase in the number of kanR cells can explain the drop of activity per recipient in the curve. The SOS response is expected to return to normal once all the cells have acquired the plasmid.
Since all cells in the recipient population have not received a conjugating DNA at the time of the β-gal assay, we calculated the SOS induction per conjugant, i.e. per recipient cell that has actually undergone DNA uptake (Materials and Methods). The results are represented as ratios over t0 in Figure 1C. As expected, the induction signal is amplified when one takes into account the conjugation rate for each plasmid. This amplification of several orders of magnitude is likely to be an effect of unsuccessful conjugation: SOS is induced by incoming DNA, that is not always converted into a replicating plasmid. The induction profiles, however, are compatible with Figure 1B: R388 and R6Kdrd strongly induce SOS, RP4 also shows a high induction, however it is lower than the former two plasmids. Once again, no (or very little) induction was observed in E .coli ΔrecA strain, confirming that conjugation induces RecA-dependent SOS response.
SOS induction by RP4 is weaker in both E. coli and V. cholerae, compared to induction during conjugation with R388 or R6Kdrd. We did not find any particular feature in the DNA sequence, or gene order of RP4 that could explain this observation. However, one can imagine that the higher transfer rate during RP4 conjugation is coupled with an early expression of entry exclusion systems, resulting in a quick decrease of ssDNA levels and repression of the SOS response.
To check if lower SOS induction was specific of RP4, we decided to test 2 other plasmids (Rs-a and RIP113) belonging to different incompatibility groups, at t60 - the peak for SOS induction observed for the plasmids mentioned earlier. Rs-a (IncW) induced SOS in E. coli and V. cholerae (Figure 1C and data not shown), confirming that SOS induction can be triggered by this plasmid. Interestingly, Rs-a conjugates at a rate of 10−1 and yields an intermediate induction level (like RP4). Even though we have a small sample of plasmids, SOS induction during mating seems to inversely correlate with conjugation rate (at 1h of mating) or replication of plasmids, except for R6Kdrd. Further study is needed to verify this observation. On the other hand, RIP113 (IncN) induced SOS in V. cholerae only (Figure S2).
An increasing number of non-replicative conjugative elements, generally named ICE, have been described in bacteria. One of the best studied is the SXT element discovered in V. cholerae [35]–[37]. We addressed if conjugative transfer of an SXT element integrated in the chromosome of E. coli [38] to V. cholerae also induces the SOS response. We observed a similar induction profile as for the conjugative plasmids with a peak of induction measured at t210 (Figure 1B). The delay can likely be explained by the very low transfer rate (10−6 after 6 six hours of mating).
Our results show that plasmids lacking the psiB gene (here RP4, R388, R6Kdrd and Rs-a) induce SOS upon conjugation into the recipient cell. On the other hand, RIP113 (IncN) induced SOS in V. cholerae only (Figure S2), thus behaving like R64drd and R100-1 plasmids.
R64drd and R100-1 plasmids do not induce (or very poorly) the SOS response in E. coli (Figure 1B and 1C). This was expected as these plasmids carry a psiB anti-SOS gene. Plasmid RIP113 behaves like R64drd and R100-1 in terms of SOS induction in E. coli. We thus suspected RIP113 to carry a psiB gene as R64drd and R100 plasmids. This was confirmed by PCR amplification with psiB-specific primers (data not shown). This finding is supported by another IncN plasmid which has been sequenced: the R64 plasmid carries a gene named stbA (locus R46_027), presenting 42% DNA sequence identity with psiBF.
We observed a strong induction of SOS by the same 3 plasmids in V. cholerae (Figure 1C), suggesting that the psiB gene is either not expressed in V. cholerae or that its product is not active in this species (R64drd and R100-1 do not replicate in V. cholerae, thus no activity per conjugant could be calculated). Moreover, SOS induction is continuously high for R64drd and R100-1 plasmids after ∼60 min, whereas SOS induction declined after 60 min for RP4, R6Kdrd and R388, as mentioned above. We were unable to delete psiB from R64drd and thus could not check if in its absence SOS induction would be restored in E. coli. The reason for the unsuccessful cloning attempts could be the presence of several genes (such as ssb coding the single strand binding protein, anti-restriction gene ardA, or flm/hok) in the same region where ORFs and regulatory regions overlap [9], [11], [39]–[41], such that deletion of psiB could have unpredicted consequences on plasmid transfer and replication. Instead, psiB from R64drd was cloned and over-expressed from a pBAD plasmid, under the control of the arabinose inducible promoter. SOS induction after mitomycin C (MMC) treatment was measured in E. coli sfiA::lacZ and V. cholerae recN::lacZ containing either empty pBAD or pBAD-PsiB+ plasmids. As previously published [31], MMC treatment induced SOS in E. coli and V. cholerae (Figure 2). SOS induction was strongly reduced in E. coli when PsiB was expressed from pBAD (6 fold induction instead of 11.6 fold, Figure 2A) whereas SOS induction was insensitive to PsiB expression in V. cholerae (∼60 fold induction with and without PsiB over-expression, Figure 2B). These results show that the psiBR64drd (and presumably the psiBR100-1 which presents 85% identity to psiBR64) is expressed during conjugation in E. coli and inhibits the SOS response, whereas in V. cholerae, psiBR64drd/R100-1 has no or very little anti-SOS activity, allowing R64drd and R100-1 transfer to induce SOS. The fact that R64 and R100-1 are narrow host range enterobacterial plasmids [40], [42] and do not replicate in V. cholerae, can explain the continuous induction we observe. Entering plasmid DNA is not replicated and new rounds of conjugation can carry on, resulting in continuous re-induction of SOS.
The psiB anti-SOS function is found in several narrow host range plasmids belonging to the IncFI, IncFII, IncI1, IncK and IncN incompatibility groups [9] (and results obtained by blasting PsiB on GenBank plasmid sequences). These plasmids replicate in bacteria from the genera Enterobacter, Escherichia, Salmonella and Klebsiella. Bagdasarian and colleagues have suggested that PsiB could interact with RecA to inhibit its ability to induce SOS [16]. It was recently shown that PsiB binds to RecA in solution [43]. PsiB would then inhibit SOS by preventing RecA nucleofilament formation on ssDNA. Since PsiB from R64, R100-1 and RIP113 plasmids does not inhibit the SOS response in V. cholerae when over-expressed, we hypothesized that PsiB would be deficient in interacting with RecAVch. Our β-gal tests show that when expressed in V. cholerae together with RecAEco, PsiB reduces the SOS response from 60 fold to 24 fold induction (Figure 2B). Consistently, when co-expressed with RecAVch in an E. coli ΔrecA strain, PsiB does not alleviate SOS (Figure 2A, note that RecAVch is active in E. coli). Finally, expression of RecAEco in the E. coli ΔrecA strain complements SOS induction alleviation by PsiB. Altogether, these data suggest that PsiB is functional only in bacterial species where its carrier plasmids normally reside (here E. coli), thus antagonising RecA in a species-specific manner.
On the other hand, we showed that the RIP113 plasmid isolated from Salmonella, an enterobacterium, also carries the psiB gene and behaves like R64drd and R100-1 in inhibiting SOS in E. coli but not in V. cholerae. Unlike these two plasmids, RIP113 replicates in V. cholerae but since it was isolated in Salmonella, and to our knowledge IncN plasmids have not been described in V. cholerae so far, we considered that V. cholerae is not one of its usual hosts. To our knowledge PsiB is present only in narrow host range plasmids. We conclude that PsiB functions in a species-specific manner.
It was recently shown that the integron integrase is regulated by the SOS response [31]. We showed above that conjugational DNA transfer induces SOS. We then addressed whether conjugation affects V. cholerae IntIA SI integrase expression levels. To do this, we constructed a V. cholerae reporter strain containing a translational fusion between intIA and gfp (7093::p4640), and used flow cytometry to determine the fraction of cells where the integrase-GFP fusion was induced. As expected, no induction was observed in the ΔrecA control strain (Figure 3). In the recA+ strain we observe no induction after conjugation with RP4 and R6Kdrd (Figure 3). Alternatively, the integrase expression increased 2.8 fold when the strain is conjugated with R388, and 5.3 and 6.2 fold with R64drd and R100-1, respectively. In β-gal SOS induction tests shown earlier, RP4 and R6Kdrd also yielded a lower induction in total population graphs (Figure 1B). Note that β-gal induction reflects the recN promoter, which is more strongly expressed than the intI promoter. Our results imply that the SOS induction during RP4/R6Kdrd conjugation may not reach sufficiently high levels to induce the integrase reporter used in flow cytometry experiments.
Finally we tested mating of E. coli carrying an SXT element with V. cholerae. SXT transfer is induced through induction of SOS when the donor is treated with MMC [37]. Transfer of the SXT element into V. cholerae increased intIA promoter activity 12 fold compared to a plasmidless control and was 2 fold higher than uninduced cells (i.e. without MMC treatment of donor).
We have shown that conjugation induces SOS in the recipient bacteria and flow cytometry analysis clearly shows that the integron integrase is induced during conjugation in V. cholerae. In a first set of experiments, we wanted to test if the SOS induction leads to a higher activity of the integrase promoter in E. coli, using the class 1 integrase IntI1. We developed an experimental strategy in an E. coli strain that contains an insertion in the dapA gene (7949). This strain is unable to synthesize DAP (2,6-diaminopimelic acid), and as a result is not viable without DAP supplemented in the medium. The insertion in dapA is flanked by two specific recombination sites, attI and attC. Integrase expression causes site-specific recombination and excision of the synthetic cassette, restoring a functional dapA gene and allowing the strain to grow on DAP-free medium (Figure 4A). We transformed in this dapA- strain a multi-copy plasmid (p7755) carrying the intI1 gene under the control of its natural SOS regulated promoter. The recombination rate due to integrase expression is calculated as the ratio of the number of cells growing in the absence of DAP over the total number of cells. Figure 4B shows the cassette excision rate in E. coli 7949 p7755 after conjugation with different conjugative plasmids. In the absence of a conjugative plasmid in the donor cell, the spontaneous excision rate is about 10−5, which reflects the stringency of the intI promoter. Conjugation with R6Kdrd and R388 increases excision rate to 10−3 and 10−2 respectively, whereas conjugation with R64drd does not increase significantly beyond the basal recombination level. RP4 yields an intermediate level of DAP+ cells, which is compatible with its intermediate SOS induction level in E. coli. These results are consistent with SOS induction results in E. coli, and as expected, there is a correlation between SOS induction and integrase induced cassette recombination. To confirm that cassette recombination is due to integrase expression, we performed the same experiment in strain 7949 lacking the integrase carrying plasmid p7755, and no cassette excision was observed (<10−8). We conclude that conjugation with psiB deficient plasmids in E. coli induces the expression of the integrase from the intI1 promoter, and thus triggers cassette recombination.
In the cassette excision experiment described above, we used a multicopy plasmid expressing the intI1 integrase in E. coli. Since conjugation induces the SOS response and in turn expression of the integron integrase in V. cholerae, we addressed in a second set of experiments whether conjugation in wild type V. cholerae can trigger recombination events in the superintegron. The V. cholerae SI carries a promoterless catB cassette that is not expressed in V. cholerae laboratory strain N16961 because it is located 7 cassettes (approximately 5000 bp) downstream of the Pc promoter [44]. When expressed, the catB gene confers resistance to chloramphenicol (Cm). We tested if conjugation can spontaneously yield Cm-resistant (Cm-R) V. cholerae cells, i.e. if IntIA is induced and recombines the catB cassette to a location allowing its expression (Figure 5A).
Our results show that when the donor strain does not carry any conjugative plasmid, the rate of CmR cells is about 7.10−11 (Figure 5B). Consistent with the intIA induction results, conjugation with RP4 and R6Kdrd did not increase this frequency (6.10−11). Conjugation with R388, R64drd and R100-1 increased the CmR cfu appearance rate 28 fold, 280 fold and 140 fold, respectively. To verify that this increase was dependent on SOS, we deleted recA in the recipient strain and found that the conjugative plasmids yielded a rate of CmR lower than 10−11 for all plasmids (no colony observed).
To determine if these events corresponded to IntIA mediated cassette rearrangement, we performed a PCR analysis with primers in the Pc promoter and at the beginning of the catB cassette. In the wild type strain, this PCR amplifies a band of about 5000 bp. In the CmR colonies, the PCR amplified a band of 1432 bp (Figure S2A). Sequencing confirmed that the catB cassette had been relocated closer to the Pc promoter. catB was now present in second position, after the first cassette, compatible with an excision and integration in the first attC site downstream of attI. We know that IntIA can promote recombination between two attC sites [45]. The first cassette coding for a hypothetical protein downstream of attI may be important for viability under laboratory conditions. Alternatively, there may be a strong promoter in this first cassette, allowing a better expression of the catB cassette that could be insufficiently expressed from any other location under our selective conditions (involving high Cm concentrations).
In order to determine if cassettes between the Pc promoter and catB gene were deleted after rearrangement – i.e. if catB moved because cassettes were deleted or because it was re-integrated – we performed PCR analysis with several oligonucleotides amplifying cassettes located between attI and catB in V. cholerae N16961. We found that these cassettes were still present in the genome of the Cm-resistant clones (data not shown), showing that they were not deleted, and indicating that catB was relocated by recombination events.
To further address if other cassette rearrangements had occurred in those cells after conjugation, we isolated the genomic DNA (gDNA) from three CmR colonies obtained after conjugation. We digested gDNA with AccI, which has 35 restriction sites evenly distributed within the SI. We then hybridized with a mix of 10 probes complementary to 17 cassettes in the SI. The southern blot (Figure S2B) clearly shows that 2 of the 3 colonies have different hybridization profiles compared to the original N16961 gDNA, confirming that several cassettes (other than catB) have moved within the SI after conjugation induced SOS.
We conclude that conjugation with strong SOS inducing plasmids, R388, R64drd and R100-1, increases IntIA expression levels and promotes cassette rearrangements at a 100 fold to 1000 fold higher rate than under non stressful conditions. Although conjugation of R6Kdrd strongly induced the SOS response, it did not have any effect on intIA expression and cassette recombination in our experimental setup. Even though it is possible that R6Kdrd encodes an uncharacterized function able to specifically prevent the IntI expression during the SOS induction, we think this observation is most likely due to an insufficient sensitivity of our setup.
We showed that conjugation of RP4, R6Kdrd, R388 and Rs-a plasmids, that do not carry any anti-SOS function, induces the SOS response in recipient E. coli and V. cholerae cells. Alternatively, plasmids R64drd, R100-1 and RIP113 that do carry the anti-SOS psiB gene do not induce SOS when the recipient cell is E. coli, while the SOS response is induced in V. cholerae. Finally, the SXT element (here integrated in the E. coli chromosome) is also able to induce SOS when it transfers to recipient V.cholerae.
It has been shown that during inter-species Hfr conjugation, SOS is induced in the host cell [46], [47]. It was proposed that the low level of homology prevents rapid recombination of incoming DNA into the chromosome and thus dramatically enhances the SOS induction. This suggests that SOS induction levels may reflect the ability of RecA to find homologous DNA and initiate strand exchange [48]. In the case of plasmid conjugation, there is no homology with the bacterial chromosome, explaining the very high SOS induction levels we observe. Moreover, SOS induction proceeds as a wave. At the early stages of SOS induction, the concentration of LexA decreases as it is self-cleaved, then LexA synthesis is induced at later stages, and if ssDNA does not persist, the SOS induction level gradually decreases. This is what happens when RP4, R388 and R6Kdrd conjugate into E. coli cells (and RP4 in V.cholerae), and explains the bell-shaped induction curves we obtained. After 1h, nearly all the recipient cells are conjugants and no new conjugation is initiated because of plasmidic entry exclusion systems [49]. As conjugation or establishment rates are very low for R388 and R6Kdrd in V. cholerae, plasmidless host cells are always present in the total population so we observe a plateau reflecting new rounds of conjugation during the course of the experiment. The fact that R64drd and R100-1 are the strongest inducers in V. cholerae could thus be due to the inability of these plasmids to synthesize the complementary DNA strand and establish themselves in V. cholerae, increasing the prevalence of ssDNA accessible to RecA binding. Moreover, this strongly suggests that abortive conjugation induces SOS, which explains the fact that we are able to detect SOS induction in the total population despite the lack of transconjugants in V. cholerae. This is consistent with data showing very high induction values when we calculate the SOS induction in conjugants only. The plateau of induction in the whole population points to a “permanent conjugation state” where ssDNA enters the host cell, induces SOS, but does not replicated, and a new round of conjugation begins.
Another interesting observation is that SOS induction does not seem to prevent conjugation. Indeed, the conjugation rate for all plasmids (in E. coli for instance) is approximately the same after 2h of mating even though they induce SOS differently at the beginning of mating. To test this point, we used recipient cells already induced for SOS by pre-incubation with MMC, and these cells yielded the same conjugation rates in E. coli for a given plasmid regardless of the SOS induction level (Figure S1).
SOS induction is due to RecA binding to ssDNA. We have shown above that plasmids R64drd, R100-1 and RIP113 that carry the anti-SOS psiB gene do not induce SOS when the recipient cell is E. coli. PsiB has been shown to interact with RecAEco in vitro [43] and in vivo (Figure 2), preventing it from binding to ssDNA and inducing the SOS response. Even though RecAVch and RecAEco show 79% protein identity, our data suggests that PsiB is impaired in its interaction with RecAVch in vivo (Figure 2), explaining why PsiB does not strongly reduce the SOS induction in V. cholerae as it does in E.coli. The presence of the psiB anti-SOS function in narrow host range plasmids such as R64 and R100 suggests that the dissemination strategies of narrow and broad host range plasmids could be distinct. Induction of the SOS response can be potentially detrimental to the host cell because of the induction of mutagenic polymerases or cell division arrest (like E.coli sfiA) [50]. Thus, it is tempting to speculate that narrow host range plasmids use their anti-SOS gene as a furtive strategy to hide from their customary host and thus prevent the host cell from being stressed and change its own or the incoming plasmid DNA. Note that by narrow host range plasmids, we mean plasmids that only replicate in a restricted number of bacteria (such as R64) but also plasmids that are found only in a few kinds of hosts in nature, even though they are able to replicate in others, such as the RIP113 originally in Salmonella.
One consequence of SOS induction during conjugative DNA transfer is the triggering of integron cassette recombination. Conjugation with strong SOS inducer plasmids R388 and R6Kdrd in E. coli increases expression of IntI1 from its SOS regulated natural promoter leading to an increased RecA-dependent cassette excision rate, whereas plasmids R64drd and R100-1 that do not induce SOS in E. coli do not trigger cassette recombination in our E. coli cassette recombination assay. These results highlight the existence of a link between conjugation and site-specific recombination, leading to genome evolution. We also showed that conjugation triggers cassette recombination in the natural context of the SI carried in wild type V. cholerae. Plasmids R388, R64drd and R100-1 strongly induce SOS (and intIA) in V. cholerae and significantly increases the cassette recombination rate.
Our results highlight the link between conjugative HGT and genome evolvability in V. cholerae. Since conjugation induces integrase activity, one can consider conjugative plasmids as both vehicles for cassette dissemination and cassette shuffling for those already present in the SI. Indeed, some plasmids such as R388 [51] and R64 [52] carry an RI platform that can acquire new cassettes and transmit them to a new host by conjugation. It was shown that R388 can incorporate the catB cassette from the V. cholerae SI and transfer it to other bacteria [44]. Here we observed the displacement of a cat cassette catalyzed by the V. cholerae IntIA in its natural context. Conjugation can thus bring new cassettes but also favour their integration into the host chromosomal integron by inducing SOS.
Cassette recombination upon conjugation could be a widespread mechanism, since conjugation is a naturally occurring phenomenon in highly concentrated bacterial environments, such as the host intestinal tract (see for example [17]–[19], [21]), biofilms [53]–[55] forming in the aquatic environment where V. cholerae grows, or even on medical equipment in hospitals [56]. Moreover, no mutation has been found in the bacterial chromosome that can prevent the uptake of conjugative DNA, meaning that bacteria cannot avoid being used as recipient cells [57]. By inducing the SOS response, incoming DNA triggers its own recombination not only through integrase induction but also homologous recombination, promoting genomic rearrangements.
Another important effect of SOS induction is the derepression of genes implicated in the transfer of integrating conjugative elements (ICEs), such as SXT from V.cholerae, which is a ∼100 kb ICE that transfers and integrates into the recipient bacterial genome, conferring resistance to several antibiotics [37]. Moreover, different ICEs are able to combine and create their own diversity in a RecA-dependent manner via homologous recombination [35], [36], and also, as observed here for SXT transfer in V. cholerae, by inducing SOS following transfer. Thus, SOS induction leads to genetic diversification of these mobile elements and to their transfer to surrounding bacteria, spreading antibiotic resistance genes, among others.
Conjugation induced SOS is thus one of the mechanisms allowing bacteria to evolve in their natural niches, creating the diversity that allows them to adapt to new environments and survive. Under conditions where SOS is prevented (by bacterial means such as the PsiB system or exogenously), cassette recombination is decreased to experimentally undetectable levels, showing that SOS induction plays an important role in adaptation, and can be used by broad host range plasmids to adapt to a new host. Consistently, narrow host range plasmids that do not need to adapt to a new host, express an SOS inhibitor to maintain the integrity of the plasmid DNA and host genome. This connection between host range and SOS induction needs to be expanded to a larger range of plasmids to determine its general character. A significant association between laterally transferred genes and gene rearrangements was already suggested in [58], which is consistent with our data, when we consider that SOS induction plays a major role in gene rearrangements. Remarkably, induction of SOS considerably enhances genome duplications and mutagenesis [59]. Further work is needed to test whether other HGT mechanisms, such as transformation, also induces SOS; considering that many bacterial species, like V. cholerae for instance, are naturally competent [60]. It would be interesting to investigate if SOS is induced in the gut of the host animal. If this were the case, inhibiting the bacterial SOS response would become an ideal target to prevent the acquisition of antibiotic resistance genes, and could be used in combination with antibiotics for the treatment of infections.
For strain and plasmid constructions, and oligonucleotide list, see Text S1 and Tables S1 and S2.
Overnight cultures of donor and recipient cells were diluted 100× in LB and grown until OD∼0.5. Donor and recipient cells were then mixed in 1∶1 ratio on 0.045µm conjugation filters on LB plates preheated at 37°C. At each time point, a filter was resuspended in 5ml LB and dilutions were plated on selective plates to count (i) conjugants and (ii) total number of recipients. For details, see Text S1.
β-gal tests were performed on these cultures as described ([61] and Text S1). According to the Miller formula:and we calculated:andwhere is the basal expression per cell when SOS is not induced. For details, see Text S1.
SOS induction tests using MMC were performed as published [31].
The same conjugation assay was performed overnight for the flow cytometry experiments. For each experiment, 100000 events were counted on the FACS-Calibur device. For details, see Text S1 and Figure S3.
Described in the Results section. For details, see Text S1.
The recipient strain was V. cholerae N16961 (Cm sensitive 5µg/ml Cm). The donor strain was DH5α or a dap- derivative (Π1) for counter selection of Cm-R plasmids. Conjugations were performed as described for 4h. Filters were resuspended, centrifuged and the pellet was plated on LB medium containing 25µg/ml Cm. PCR screenings were performed using oligonucleotides cat2/i4 and the GoTaq polymerase. Oligonucleotides 896 to 905 were used to verify the presence of other cassettes.
|
10.1371/journal.pgen.1005009 | An Essential Role of the Arginine Vasotocin System in Mate-Guarding Behaviors in Triadic Relationships of Medaka Fish (Oryzias latipes) | To increase individual male fitness, males of various species remain near a (potential) mating partner and repel their rivals (mate-guarding). Mate-guarding is assumed to be mediated by two different types of motivation: sexual motivation toward the opposite sex and competitive motivation toward the same sex. The genetic/molecular mechanisms underlying how mate presence affects male competitive motivation in a triadic relationship has remained largely unknown. Here we showed that male medaka fish prominently exhibit mate-guarding behavior. The presence of a female robustly triggers male-male competition for the female in a triadic relationship (2 males and 1 female). The male-male competition resulted in one male occupying a dominant position near the female while interfering with the other male's approach of the female. Paternity testing revealed that the dominant male had a significantly higher mating success rate than the other male in a triadic relationship. We next generated medaka mutants of arginine-vasotocin (avt) and its receptors (V1a1, V1a2) and revealed that two genes, avt and V1a2, are required for normal mate-guarding behavior. In addition, behavioral analysis of courtship behaviors in a dyadic relationship and aggressive behaviors within a male group revealed that avt mutant males displayed decreased sexual motivation but showed normal aggression. In contrast, heterozygote V1a2 mutant males displayed decreased aggression, but normal mate-guarding and courtship behavior. Thus, impaired mate-guarding in avt and V1a2 homozygote mutants may be due to the loss of sexual motivation toward the opposite sex, and not to the loss of competitive motivation toward rival males. The different behavioral phenotypes between avt, V1a2 heterozygote, and V1a2 homozygote mutants suggest that there are redundant systems to activate V1a2 and that endogenous ligands activating the receptor may differ according to the social context.
| Males of various species, including humans, remain near a (potential) mating partner and repel their rival males (mate-guarding). Mate-guarding is mediated by two different types of motivation: sexual motivation toward the opposite sex and competitive motivation toward the same sex. Here we show that the arginine-vasotocin (AVT), a non-mammalian homolog of arginine-vasopressin (AVP), system mediates the molecular mechanisms underlying how mate presence (intersexual relationship) affects male competitive motivation (intrasexual relationship) in mate-guarding (a triadic relationship). We first established a novel behavioral paradigm in which medaka robustly exhibit mate-guarding behavior, that allowed us to study the genetic/molecular mechanisms underlying mate-guarding. Behavioral analysis of courtship behaviors, aggressive behaviors and mate-guarding using medaka mutants in this paradigm suggested that the AVT system is involved in the process in which mate (female) presence drives sexual motivation of males, which may facilitate mate guarding in the triadic relationship. Our study provides genetic evidence that the AVT/AVP system regulates mate-guarding behaviors, suggesting that the role of this system in mate-guarding is evolutionarily conserved from teleosts to mammals.
| Male mating strategies are considered to take two major forms: intersexual interaction (female-male interaction) and intrasexual competition (male-male competition) and there is extensive literature focusing on the neural/molecular mechanisms underlying these strategies [1]. In addition to these mating strategies, males of various species, including insects [2, 3], birds [4, 5], mammals [6], primates [7], and humans [8], exhibit mate-guarding behaviors in which they remain near a (potential) mating partner and repel their rival males, which involves both intersexual and intrasexual interactions. Lack of attention to either a mating partner or rival males in mate-guarding would allow the rivals to approach and mate with the partner, known as sneaking [9] and extra-pair copulations [10–12]. In fact, ecological studies indicate that mate-guarding is required to increase individual male fitness in some vertebrates [6, 12].
As mate-guarding in a triadic relationship comprises both inter- and intra-sexual interaction [2–8], mate-guarding is assumed to require two different types of motivation: sexual motivation toward the opposite sex and competitive motivation toward the same sex. Little attention has been paid to the genetic/molecular mechanisms underlying how mate presence (intersexual interaction) affects male competitive motivation (intrasexual interaction) in a triadic relationship, mainly because of the lack of an established behavioral system that robustly elicits this type of complex behavior under laboratory conditions for any genetic model organism. To explore this issue, we focused on medaka fish (Oryzias latipes), which is a commonly used model animal in molecular genetics. The medaka mating system involves socially-regulated female preference (intersexual interaction) [13–16] and male-male competition (intrasexual interaction) [17, 18]. Medaka fish exhibit mating behavior every morning, because sexually mature females have a 24-h reproductive cycle. The medaka mating behavior comprises sequential steps, such as male courtship display and synchronized mating [13–16]. Although there are a number of reports on medaka males aggressive behaviors toward other males in multi-male groups [17, 18], there is only one previous report describing the emergence of mate-guarding behavior in a triadic relationship of medaka fish (1 female and 2 males) [19]. This behavior in medaka fish has not been quantitatively investigated, however, because there has been no applicable assay to reliably assess this behavior.
In the present study, to investigate the genetic/molecular mechanisms underlying mate-guarding behavior, we developed a behavioral assay that robustly elicits mate-guarding behavior in medaka. We then examined the possible involvement of arginine-vasotocin (AVT), a non-mammalian homolog of arginine-vasopressin (AVP). In monogamous male prairie voles, the AVP system is involved in mating-induced selective aggression toward a non-mate in a dyad, which may represent mate-guarding [20, 21]. In teleost fish, AVT is implicated in various kinds of social behaviors, such as territorial behavior in a tropical damselfish [22, 23] and pair bonding in a monogamous cichlid fish [24]. In particular, AVT increases both courtship and territorial behaviors of non-territorial males in the bluehead wrasse in the field [25], implying that the AVT system has a key role in the motivation of mating-related behaviors. The possible involvement of the AVT/AVP system in actual mate-guarding in a triadic relationship, however, has not been investigated in these species under laboratory conditions. To evaluate the requirement of the molecular components of the AVT pathways in the regulation of mate-guarding, loss of function analysis using knockout (KO) animals is a valid and feasible method. Here we generated medaka avt and V1a type AVT receptor 1 (V1a1) and V1a type AVT receptor 2 (V1a2) mutants using advanced molecular genetics, such as the TILLING (Targeting Induced Local Lesions IN Genomes) [26, 27] and TALEN (Transcription Activator-Like Effector Nucleases) methods [28, 29], and examined how the AVT pathway is involved in mate-guarding behaviors.
We quantitatively analyzed the mate-guarding behavior by calculating the relative position of three fish in a behavioral assay (S1 Fig.). When two males and one female were allowed to swim together in the morning (Fig. 1A), the male-male competition led to the tendency of one male to maintain its position near the female and prevent the other male from approaching the female (S1–S2 Movies). We defined this behavior of the nearest male as “mate-guarding behavior” (Fig. 1B). To quantify the degree of mate-guarding, we generated a novel index to represent the degree of mate-guarding of the focal fish. First, we measured time-series coordinate data of the three individual medaka fish (2 males and 1 female) for 100 s and calculated the relative locations of the three fish. The relative positions of the focal male were calculated when the fixed positions of the female and the other rival male were defined as (0, 0) and (1, 0), respectively (Fig. 1C). We spotted the relative positions of the focal male during the 100 s (once every 5 s; S1 Fig.). We then calculated the probability of the focal male being within the “guarding circle”, defined as a circle with center (1/2, 0) and radius 1/2 (Fig. 1C). The presence of the focal male in the guarding circle indicated that the focal male occupied a dominant position, allowing him to both remain near the female and interfere with the rival (the other male). Thus, the probability of being within the guarding circle was considered to represent the degree of mate-guarding of the focal fish. Hereafter, we defined this probability as the “guarding index”.
To evaluate whether mate-guarding emerges in a triadic relationship, we examined whether the “guarding index” of the nearest of the two males significantly increases based on the interactions of the three fish (guarding test). The “near male” was defined as the male whose mean distance from the female during the 100-s recording period was shorter than that of the other male. The “far male” was defined as the other male. We then generated a “merged group” as a negative control, in which the three fish freely swam in individual aquaria without any social interaction. We separated three fish into three tanks of the same size, recorded the independent movement of each fish, and superimposed the data, which were used to calculate the virtual guarding index of the near male as a negative control. The guarding test using a wild-type medaka strain (drR) revealed a guarding index of 62.6% ± 4.6% for the near male, which was significantly higher than that of the merged control (30.4% ± 4.7%; Fig. 1D: before, merge). There were also significant difference of the guarding index between experimental and merged using different size (small, large) and shape (rectangular vs round) tanks (S2 Fig.), suggesting that mate-guarding robustly emerges irrespective of the geometric constraints of the apparatus. In most cases, medaka males remained near the female without performing an apparent quick-circle (i.e., the male’s courtship display) and interrupted the rival male without expressing aggressive behavior such as attack or bite [17, 18]. Thus, a unique behavioral repertoire in the male-male competition emerges in the triadic relationships.
Mate-guarding in most species is considered to be a male-specific behavior in conspecific social groups [2–8]. Extended periods of mate-guarding (pre- or post-copulation) differ among species [2–8]. Here we investigated whether medaka fish exhibit the behavioral properties of mate-guarding. First, we examined whether medaka fish exhibit pre- or post-spawning mate-guarding. Sexually mature females have a 24-h reproductive cycle and spawn eggs once each morning [13–16]. We compared the guarding indices of the near male (guarding test) just before and after spawning in the morning (S3A and S3B Fig.). In addition, we performed the same test in the evening (S3E Fig.) to examine whether males exhibit mate-guarding in a time of day-dependent manner. All three indices (62.6% ± 4.6%, 62.3% ± 3.7%, and 57.9% ± 6.1%, respectively) were significantly higher than that for the negative control (merged data: 30.4% ± 4.7%, 37.0% ± 3.4%, and 29.3% ± 2.1%, respectively) (Fig. 1E), indicating that mate-guarding occurred irrespective of spawning.
We then examined whether mate-guarding emerged in the presence of females of other fish species. When the medaka female was replaced with a female zebrafish (S3C Fig.), the guarding index of the near male (31.9% ± 3.5%) was almost same as that of the merged control (27.8% ± 1.9%), suggesting that mate-guarding behavior is mediated by conspecific social cognition (Fig. 1F). Next we examined whether two medaka females exhibit mate-guarding toward a male (S3D Fig.). When two females and one male were placed in a single tank, the guarding index of the near female between the two females (21.6% ± 2.9%) was almost same as that of the merged control (26.4% ± 2.0%) (Fig. 1F). Furthermore, medaka males did not exhibit mate-guarding toward a male, and medaka females did not exhibit mate-guarding toward a female (S4 Fig.). In addition, medaka males did not exhibit significant mate-guarding toward sexually-immature females (S4 Fig.). Taken together, these results suggested that mate-guarding in medaka is a male-specific behavior toward sexually mature females.
Finally, we examined whether visual information is required for mate-guarding behavior. In medaka fish, social recognition in mating behaviors is mainly mediated by visual information [13, 15]. There was no significant difference between the guarding index of near males with a single eye removed and that of merged control (S5A Fig.). Single eye-removed males exhibit normal courtship behaviors [15], and thus the eye-ablation surgeries are assumed to not affect overall activity. In addition, medaka males exhibit significant mate-guarding toward females kept within a transparent cylinder tank that allows the males to only see the female without water intercirculating between the female’s enclosure and the male’s enclosure (S5B and S5C Fig.). Taken together, these findings suggest that visual sensory information is necessary and sufficient to elicit male mate-guarding, although the possibility that other sensory information (pheromones, touch) modulates this behavior as well could not be excluded.
In various species, mate-guarding increases male reproductive success [6, 30, 31]. We examined whether dominance in mate-guarding positively correlates with male reproductive success. In the present study, we designed a “dominance test” to determine which male was dominant based on the guarding index. Fig. 2A shows a schema of the procedure used to determine the dominance of two males with different genotypes (A and B) in a triadic relationship. First, we measured the guarding index of a male with genotype A by calculating its relative position as described above. Then we measured the guarding index of a male with genotype B as the focal fish, and compared the “guarding indices” of two males with genotypes A and B (Figs. 2A and S6).
To examine whether a dominant male had high reproductive success, we performed a paternity test using two males with different genotypes (one wild-type and the other a transgenic [Tg] expressing green fluorescent protein [GFP] in the primordial germ cells [homozygote olvas:gfp]). GFP detection in the medaka embryo allowed us to genotype the progeny. The day before mating, we performed a dominance test using wild-type and Tg males, and the next morning we performed a paternity test on the fertilized eggs from the females (Fig. 2B). Medaka females have a single brood of 5 to 20 eggs each morning and, in most cases, the eggs are fertilized by the first male that exhibits ejaculation. Thus, we could determine which male won the competition for mating based on the paternity test [15]. We performed a dominance test using 17 groups for 6 days (1 test/day; S7 Fig.). When Tg males were judged to be dominant (5/17 groups), the percentage of Tg progeny was approximately 93.6%. When wild-type males were judged to be dominant (7/17 groups), the percentage of Tg progeny was approximately 5.7%. In the remaining 5 of the 17 groups, we could not determine dominance because there was no significant difference in guarding index between the two males (Fig. 2C). The combined results of the dominance test and paternity test revealed that the dominant male had a significantly higher reproductive success rate than the subordinate male.
In prairie voles, the AVP system is suggested to be involved in mate-guarding, because AVP system mediates mate-induced selective aggression in a dyadic situation [20, 21]. Here, we found that injection of an AVT antagonist (Manning compound) impaired medaka mate-guarding in the triadic condition. The Manning compound is a commonly used antagonist of V1a receptors, including both subtypes V1a1 and V1a2 [32, 33]. We performed the guarding test using two males and one female, and then intraperitoneally injected an AVT antagonist or saline into the near male. At 5 min after the injection, we performed a second test using the same trio of fish. The guarding index of the injected fish in the second test was significantly lower than that in the first test (Fig. 3A-B and S3–S4 Movies). The AVT antagonist did not affect the overall activity of the injected males (S8A Fig. and S3–S4 Movies). In addition, the guarding index of the uninjected males significantly increased 5 min in the second test (S8B Fig.), while there was no effect of saline injection (S8C Fig.). This effect disappeared within 1 day after AVT antagonist administration. These findings suggested that AVT positively affected mate-guarding. We then examined the possible involvement of individual molecular components in the AVT pathway in mate-guarding by generating medaka mutants for genes encoding AVT [34, 35] and its receptors (V1a1 and V1a2) [36]. In fish species, two V1a-type receptors (V1a1 and V1a2) are expressed mainly in the brain [36–39], while V2-type receptors are prominently expressed in tissues other than the brain, such as gills, heart, and kidney [40]. Using the TILLING method, we identified avt mutants (avtM1R/M1R) in which the first methionine residue was changed to arginine (S9A Fig.), V1a1 mutants (V1a1F93Y/F93Y) in which a conserved phenylalanine residue was changed to tyrosine (S11 Fig.), and V1a2 mutants (V1a2N68I/N68I) in which a conserved asparagine residue was changed to isoleucine (S12 Fig.). We confirmed that the 5’ untranslated region of the annotated avt transcripts was identical with that determined by 5’Race method (S9B Fig.). In addition, we performed mass spectrometry of AVT peptide based on matrix-assisted laser desorption/ionization–time of flight mass spectrometry (MALDI-TOF MS) (S10A and S10B Fig.) and selected reaction monitoring (SRM) (S10C and S10D Fig.), and demonstrated that there was no detectable AVT peptide in the brains of the avt mutants. To examine whether mate-guarding emerges between two mutant males with the same genotype, we performed the guarding test. The guarding test using the wild-type males (Cab strain) that were used for generating the mutants indicated a guarding index of 49.2% ± 3.1% for the near male, which was significantly higher than that of the merged control (35.3% ± 1.5% merge). The guarding index of the wild-type males was relatively lower than that in previous experiments (Fig. 1E), implying that external factors such as seasonal changes may affect our measurement of the guarding index of wild-type males. Interestingly, the guarding indices of the near male of the avtM1R/M1R and V1a1F93Y/F93Y mutants (43.7% ± 2.1% and 54.0% ± 3.0%, respectively) were significantly higher than those of the merged controls (31.3% ± 2.4% and 31.0% ± 2.9%, respectively), indicating that mate-guarding emerges between these mutant males (Fig. 3C). Thus, these two genes (avt and V1a1) were not required to elicit mate-guarding, suggesting that other ligands possibly activating the V1a2 receptor could compensate for the avt deficit. In contrast, there was no significant difference between the guarding indices of the near male among the V1a2N68I/N68I mutants (38.1% ± 2.3%) and that of the merged control (36.1% ± 3.1%) (Fig. 3C), indicating that mate-guarding did not occur between two V1a2 mutant males (S5 Movie). The V1a2N68I/N68I mutation did not affect overall activity (S13 Fig.) and visual locomotion of the males (S14 Fig.). Thus, the V1a2 gene was required to elicit mate-guarding behavior. To confirm the behavioral phenotype of the loss-of-function mutations for the V1a1 and V1a2 genes, we generated V1a1 and V1a2 knockout mutant males using the TALEN method (S15 and S16A Figs.). The V1a1 and V1a2 knockouts have 4 and 7-bp deletions in the first exon, respectively. Both of the mutated transcripts encode C-terminal deleted proteins, lacking at least six of the seven transmembrane domains encoded in the first exon. Considering that the V1a1 and V1a2 receptors are seven-transmembrane receptors, the lack of six of the transmembrane domains should lead to a loss of function. The guarding index of the near male of the V1a1 KO mutant (49.6% ± 2.7%) was significantly higher than that of the merged control (29.4% ± 4.5%). In contrast, there was no significant difference between the guarding index of the near male of the V1a2 KO mutant (37.3% ± 3.8%) and that of the merged control (34.1% ± 1.9%) (S16B Fig.). Thus, these findings further supported that the loss of function of the V1a2 gene, but not the V1a1 gene, attenuated mate guarding.
We then examined whether these mutations affected dominance of male-male competition in mate-guarding behavior. We performed dominance tests using two male siblings with a different genotype: one a homozygote and the other a heterozygote mutant (S6 Fig.). The guarding index of avtM1R/M1R (23.0% ± 3.5%) was significantly lower than that of heterozygote mutant (38.9% ± 3.6%), and that of V1a2N68I/N68I (18.6% ± 3.9%) was significantly lower than that of heterozygote mutant (55.6% ± 6.4%; Fig. 3D), indicating that avtM1R/M1R and V1a2N68I/N68I mutant males tended to be subordinate in male-male competition against their heterozygote mutant siblings. We also confirmed that the probability of greater proximity to the female of avtM1R/M1R and V1a2N68I/N68I males was significantly lower than that of heterozygote mutants (S17 Fig.). In contrast, the guarding index of V1a1F93Y/F93Y (34.9 ± 4.0%) did not differ significantly from that of V1a1+/F93Y (32.5 ± 4.1%; Fig. 3D), indicating that V1a1F93Y/F93Y mutant males show equivalent mate-guarding as the heterozygote mutants. Our results demonstrated that avt homozygote mutants exhibited decreased dominance against avt heterozygote mutants in mate-guarding, although male-male competition for mate-guarding occurred between two avt homozygote mutant males (Fig. 3C). Taken together, these findings suggest that AVT ligands enhance dominance of male-male competition in mate-guarding.
We examined whether these mutants have defects in sexual motivation toward the opposite sex and/or male competitive motivation toward the same sex that could cause mate-guarding abnormalities. To test this issue, we tested aggressive behavior elicited in groups comprising only males and male courtship behaviors in a male/female pair, respectively. The homozygote avt mutant males exhibited normal aggression in a non-mate guarding situation (Fig. 4A), whereas the mutant males exhibited fewer courtship displays than wild-type males (Fig. 4B), showing that avt mutants normally have competitive motivation to the same sex (rival males), but not to the opposite sex (a potential mating partner) (Fig. 5B). In contrast, the frequencies of aggressive behaviors and courtship display of homozygote V1a2 mutant (V1a2N68I /N68I) males were significantly lower than those of the wild-type control (Figs. 4A and B), indicating that the homozygote V1a2 mutant did not normally have social motivation to either the same sex or opposite sex (Fig. 5D). Interestingly, the frequencies of aggressive behaviors of heterozygote V1a2 mutant (V1a2+/N68I) males were significantly lower than those of the wild-type control (Fig. 4A), while the heterozygote mutant males normally exhibited courtship displays (Fig. 4B), revealing that the heterozygote mutant males normally have sexual motivation, but not competitive motivation (Fig. 5C). The single functional V1a2 allele might not produce enough of a gene product, leading to an attenuated aggression in a non-mate guarding situation. In conclusion, avt mutant males displayed defects in courtship display and did not normally exhibit mate-guarding (Fig. 5B). In contrast, mate-guarding behavior exhibited by heterozygote V1a2 mutant males appeared normal despite defects in aggression (Fig. 5C). Based on these findings, the mate-guarding deficits of avt mutants were due to impaired sexual motivation, but not to impaired competitive motivation toward the same sex. In addition, behavioral analysis of V1a1 mutant males suggested that V1a1 is not required for either courtship or aggressive behavior (S18 Fig.), further implying a functional difference between V1a1 and V1a2 in social behaviors (S1 Table). In addition, AVT administration to wild-type males and V1a2 heterozygote mutant increased the frequency of aggressive behaviors in a male group, but had no effect in V1a2 homozygote mutant (S19 Fig.). Considering that avt mutants exhibit normal aggressive behaviors in a non-mate guarding situation, administration of exogenous AVT might artificially activate V1a2 receptors, which enhances male aggression.
Here we showed that mate presence drives male competitive motivation, leading to mate-guarding behaviors in a triadic relationship. The behavioral repertoire of the triadic relationship, however, differs between courtship behavior in a dyadic setup and aggressive behavior elicited in a male group. Mate presence in several species facilitates male aggressive behaviors [41–46]. For example, intrasexual competition in jacanas [45] and syngnathid fishes [46] requires mechanisms of selective attention to females. Apparent aggressive behaviors such as attacking and biting, however, were not enhanced in the triadic relationship of medaka fish. Thus, the unique behavioral repertoires (approaching behavior to a mating partner and interrupting behaviors to a rival male) emerge in the triadic relationships of medaka. Furthermore, the behavioral phenotypes in mate-guarding and aggressive behavior differed between the avt and V1a2 mutants (Figs. 5B and D). V1a2 was required for the emergence of mate-guarding among mutant males, whereas avt was not required (Fig. 3C). avt was required just for the dominance of mate-guarding (Fig. 3D). The double functional V1a2 allele was required to elicit aggressive behavior to the same degree as the wild-type, while the avt mutation did not significantly affect this behavior (Fig. 4A). These findings implied that V1a2 receptors are better conserved and more central in these signaling systems than the ligands and that some compensatory systems activate the V1a2 receptors in avt mutants. AVT may not be the only ligand that activates the V1a2 receptors (Fig. 5). The involvement of AVT/AVP ligands in aggressive behaviors has been investigated based on pharmacologic manipulations in various vertebrates from fish to rodents [22, 23, 25, 47–49]. Administration of exogenous ligands (AVT) into the brain enhances aggression, while AVT/AVP receptor V1a antagonist suppresses aggression. In medaka, AVT administration also enhanced aggression (S19 Fig.). These findings based on pharmacological analysis, however, do not preclude the possibility that other endogenous ligands activate the V1a receptors. Our findings are consistent with a recent rodent study using a rat natural mutant with vasopressin (AVP) deficiency [49], which revealed that a loss of function of the AVP gene does not affect aggressiveness, especially in reproductively experienced males. Thus, the differences in behavioral phenotypes between avt and V1a2 mutants may be due to the existence of redundant systems that activate the V1a2 receptors in medaka fish brain. Isotocin, a non-mammalian homolog of oxytocin, is a candidate endogenous ligand that activates V1a2 receptors. Isotocin has affinity for the teleost vasotocin receptor [50] and there is significant cross-talk between oxytocin, AVP and their receptors in mammals [51]. Further analysis of isotocin and its mutants is required to understand the signaling pathways that activate V1a2 receptors.
Our findings suggest that the AVT system is involved in the process in which mate presence drives sexual motivation towards the opposite sex, which facilitates male competitive motivation for mate-guarding in the triadic relationship (Fig. 5A). Our finding is consistent with previous studies in the bluehead wrasse (Thalassoma bifasciatum) and African cichlid (Astatotilapia burtoni) [25, 52]. Pharmacological manipulations of the AVT system in both of the two fish species facilitate male courtship and territorial aggression behaviors that are associated with social dominance [24, 39]. In non-fish species such as bird and frog, involvement of AVT system in courtship and territorial aggression also has been suggested [53–55]. In mammalian species, the socially monogamous prairie vole (Microtus ochrogaster) has been used as a model organism for investigating the neurobiology of this type of complex social behavior, such as pair-bonding, which involves both intersexual and intrasexual interactions [56–58]. Prairie voles exhibit pair-bonding behavior involving affiliation toward a mate and agonistic behavior toward non-mates. A series of studies using prairie voles revealed that AVP and the V1a receptor subtype have essential roles in pair bonding and other behaviors associated with monogamy [20, 21, 56–58]. In the monogamous vole, however, each of the behavioral components (affiliation toward mates and agonistic behavior toward non-mates) was analyzed individually in the dyadic setup [56–58], and possible involvement of the AVT/AVP system in the actual mate-guarding behavior in the triadic relationship has not yet been investigated in prairie voles. Establishing a quantitative assay system for this behavior under laboratory conditions allowed us to genetically study the molecular mechanisms underlying mate-guarding behavior in a triadic experimental setup, but not dissect the sub-behaviors, which have been studied for several decades.
Furthermore, it should be noted that the AVT receptor V1a2 has a central role in regulating fish social behaviors such as mate-guarding, courtship behaviors, and aggressive behaviors. The AVT receptor function in fish species has been investigated based on pharmacologic manipulations alone, and the selectivity of mammalian V1a receptor antagonists like the Manning compound for the two V1a-type receptors (V1a1 and V1a2), is unknown [22–25, 33, 50]. Thus, functional differences between the two receptors cannot be determined based on pharmacologic studies alone. Some recent studies reported differential gene expression between V1a1 and V1a2. In the larval brain of zebrafish, the two genes are expressed in the same brain regions, but few neurons coexpress V1a1 and V1a2 [37]. The differential regulation of gene expression levels in the bluehead wrasse Thalassoma bifasciatum implies the importance of V1a2 over V1a1 in fish social behavior [36]. Our findings suggested differences in the behavioral function between V1a1 and V1a2 in mate-guarding.
In adult fish brains, V1a2-expressing neurons are broadly located in various brain regions, such as the dorsal and ventral telencephalon, and the preoptic area [36, 37, 39], that are thought to be important for social decision-making [59]. It remains unknown, however, how V1a2-expressing neurons mediate individual social components in different social contexts. More importantly, it should be noted that the gene knockout method eliminates AVT or its receptors in the relevant social behavior-related neural areas, but also more widely in the brain and other tissues such as the gonads (which also express AVT receptors) throughout development. Further studies are needed that selectively manipulate subpopulations of V1a2-expressing neurons in the adult brain. Genetic mosaic techniques are available in medaka fish to visualize and/or genetically modify a neuronal subpopulation within complex neural circuits [60, 61]. Genetic dissection of the AVT system using such advanced molecular genetic methods will allow us to identify the microcircuits that regulate social behaviors. The present study using medaka mutants is an important first step toward unveiling this complex neuromodulatory pathway, for which our current understanding is very poor.
The work in this paper was conducted using protocols approved by the Animal Care and Use Committee of the University of Tokyo (permit number: 12–07). All surgery was performed under anesthesia using MS-222, and all efforts were made to minimize suffering, following the NIH Guide for the Care and Use of Laboratory Animals.
Medaka fish (Oryzias latipes; drR strain, Cab strain, mutants, Tg(homozygote olvas:gfp), were maintained in groups in plastic aquariums (13 cm x 19 cm x 12 cm (height)). TILLING mutants were provided by the Medaka National BioResource Project (http://www.shigen.nig.ac.jp/medaka/). All fish were hatched and bred in our laboratory. Sexually mature male (2.5 cm~3.2 cm, 240 mg~330 mg) and female (2.7 cm~3.1 cm, 370 mg~400 mg) adult medaka 3~5 months of age producing fertilized eggs every morning were used at least once a week for the behavior assay. The water temperature was ~28°C and light was provided by standard fluorescent lamps for 14 h per day (08:00–22:00).
A detailed procedure is provided in S1 Fig. One female and two males were placed in an aquarium (water depth was about 3–4cm, the light intensity was about 400–500 lx), and their behavior was recorded from the bottom of the aquarium, in the morning (10:00 to 12:00) and in the evening (20:00 to 21:00). Light was provided for 14 h per day (08:00 to 22:00). As a negative control group (merged group), we performed the same experiment using virtually merged trios, recording one female and two males, each placed in a separate aquarium (“Merge”). We converted video files into 21 image sequences per 5 s, and manually spotted the head and tail positions of the three medaka fish using ImageJ (NIH, Bethesda, MD, USA) to calculate the center positions as the body positions. The male whose mean distance from the female was shorter than that of the other male was “the near male” and the other was “the far male”. Based on the positions of the female (xF, yF), the far male (xMf, yMf), and the near male (xMn, yMn), the relative positions of the near male (X, Y) were calculated by the following formula when the female and far male positions were defined as (0, 0) and (1, 0), respectively (See S1 Fig.).
We spotted the relative positions of the near male and defined a circle with center (1/2, 0) and radius 1/2 as the “guarding circle”. When the near male was within the guarding circle, the angle between the vectors from the near male to the female and from the near male to the far male was obtuse. The probability of being in the guarding circle was defined as the “guarding index” (See S1 Fig.).
A detailed procedure is provided in Fig. 2A. We used one genotype A male and one genotype B male and evaluated their mate-guarding behavior in the presence of a female. We measured the relative locations of the three fish and calculated the probability of the genotype A male being in the guarding circle when the female and genotype B male positions were defined as (0, 0) and (1, 0), respectively (Left). We defined this probability as the “guarding index of genotype A”. In contrast, we also calculated the probability of the genotype B male being in the guarding circle when the female and genotype A male positions were defined as (0, 0) and (1, 0), respectively (Right). We defined this probability as the “guarding index of genotype B” and compared with that of genotype A. A higher guarding index indicates higher dominance in the mate-guarding behavior compared with the other male (Fig. 2A).
The paternity test was performed as described previously [15]. A detailed procedure is provided in Figs. 2B and S7. One female and two males, one of which was the drR wild-type strain and the other of which was Tg (homozygote olvas:gfp), were used for this test. To determine which male was dominant, we performed a dominance test over 6 days using same three fish (6 trials). The male whose mean guarding index for the 6 trials was significantly higher was considered dominant and the other was considered subordinate (S7 Fig.). Tg(olvas:gfp) were distinguished from drR by the GFP fluorescence of primordial germ cells, visible at 3 days post-fertilization, and even in the ventral area of the adult, by fluorescent microscopy (MZ16F, Leica, Tokyo, Japan). Therefore, we genotyped the progeny based on GFP detection.
This procedure was performed as previously described [25] with minor modifications. After the guarding test, we anesthetized the near males using MS-222 and intraperitoneally injected the Manning compound. We injected 3.2 μg Manning compound/g body weight.
This procedure was performed as previously described [26]. To amplify the avt, V1a1, and V1a2 locus that includes the final gene product, we performed polymerase chain reaction (PCR) with avt-specific primers: 5′- AGACGTCCACACCGACA-3′ and 5′- GCCAAAAGCATCTCACCT-3′, and V1a1-specific primers: 5′- GGACAGCCTTTGCAACTT-3′ and 5′- GTTTGTGGAGGAGAGGGTA-3′, and V1a2-specific primers: 5′- CAGCGTGCTGCTCTTGA-3′ and 5′- CGATGTAACGGTCCAAAGT-3′. The PCR conditions were as follows: 1 cycle of 94°C for 2 min, followed by 45 cycles of 94°C for 15 s; annealing at 63°C (avt) or 60°C (V1a1, V1a2) for 30 s, and then at 68°C for 30 s (avt, V1a2) or 60 s (V1a1); and a final denaturing and re-annealing step (1 cycle of 94°C for 30 s, followed by rapid cooling to 28°C). Each of the 5771 PCR products derived from genomic DNAs was subjected to the high-resolution melting assay. Based on differences in the melting curves, mutant candidates were selected. Melting curves were analyzed using the LightScanner (Idaho Technology, Salt Lake City, UT, USA), as previously described [26]. The mutations were then identified by sequencing the PCR product of the second positive genomic DNAs using BigDye Terminator version 3.1 (Applied Biosystems, Foster City, CA, USA) and the ABI 3730XL sequencing platform. We backcrossed the TILLING mutants with Cab fish three times and crossed those fish to generate the homozygote mutants.
Total RNA was extracted from the male medaka brain (drR and Cab strains) using TRIZOL Reagent. 5’-Rapid amplification of cDNA ends (5’-RACE) was performed using a SMARTer RACE cDNA Amplification Kit (Clontech) following the manufacturer’s instructions. The amplification was performed using the Universal Primer A mix and 5’- TGATCCCAGCCTCCGGCAAT-3’ (avt gene specific primer). The amplified PCR products were cloned into a pGEMT Easy Vector (Promega) and sequenced. We sequenced 10 cDNA clones derived from the drR strain and 9 cDNA clones derived from the Cab strain and confirmed that the sequences of all 19 cDNA clones started from the transcription initiation site, which was predicted based on the annotated avt sequence.
Mass spectrometry (MS) was performed as described previously with minor modifications [62, 63]. Peptides in the pituitary of wild-type (Cab) and avt mutant strains were extracted using 0.1% (v/v) trifluoroacetic acid (TFA). After concentration and purification with a self-made C18 STAGE tip [62, 64], 5 μl of sample/one pituitary in 0.1% (v/v) TFA solution was analyzed using MS. To create peptide profiles of the pituitary of wild-type (Cab) and avt mutant strains, we performed MALDI-TOF MS analysis using an AXIMA TOF2 mass spectrometer (Shimadzu biotech, Kyoto, Japan) [34, 63]. The peptide solution (0.5 uL) was mixed with 0.5 uL matrix solution [2% (w/v) α-cyano-4-hydroxycinnamic acid in 50% (v/v) acetonitrile/0.1% (v/v) TFA] and spotted on a stainless-steel MS sample plate. To confirm the expression level of the avs peptide, SRM analysis was performed on a QTRAP5500 mass spectrometer coupled to an Eksigent nanoLC-Ultra system via a cHiPLC-nanoflex module (AB SCIEX, Framingham, MA, USA). Peptides were separated on a nano cHiPLC C18-reversed phase column (Chrome XP C18CL, 75 μm ID × 15 cm) and eluted at a constant flow rate of 300 nL/min. A linear gradient (2%−50% mobile phase B) was applied for 15 min, followed by a 6-min wash with 90% mobile phase B; the column was then equilibrated for 20 min with 2% mobile phase B.
TALEN experiments were performed as described previously [28, 29]. Potential TALEN target sites in the locus were searched using the TALEN Targeter program (https://tale-nt.cac.cornell.edu/node/add/talen). TAL repeats were assembled using the Golden Gate assembly method with slight modifications. Expression vectors for the TALENs were linearized by digestion with NotI. Capped RNAs were synthesized using the mMessage mMachine SP6 Kit (Life Technologies, Gaithersburg, MD, USA) and purified using the RNeasy Mini Kit (Qiagen, Valencia, CA, USA). Pairs of RNA for the TALENs (150 ng/μl) were injected into fertilized eggs of the drR strain by a microinjection method.
We obtained the genomic sequence information of V1a receptors from the Ensemble medaka genome browser (http://www.ensembl.org/Oryzias_latipes/Info/Index) and predicted their secondary structure using “SOSUI”, which is a program that predicts transmembrane regions from amino acid sequences (http://bp.nuap.nagoya-u.ac.jp/sosui/).
We assessed the optomotor response of the avtM1R/M1R, V1a2+/N68I, and V1a2N68I/N68I fish using our previously described apparatus [65] (see S14 Fig.). The medaka were placed in a fixed 15-cm-diameter circular tank, which was placed within a striped 20-cm-diameter cylinder. The depth of water in the tank was 2 cm. The striped cylinder was positioned on a rotatable metal disk that was driven by a motor, IHT6P3 (SERVO, Kiryu, Japan), that could move in either direction and at various speeds using the C-30PN (SERVO) motor driver. We recorded the optomotor response of medaka using a CCD camera (XC-ST70; SONY, Tokyo, Japan) and extracted the position of the medaka and stripes. A series of frames was analyzed using the software Move-tr/2D 7.0 (Library, Tokyo, Japan).
This procedure was performed as previously described [15]. Males and females were separated in the evening (18:00–19:00) the day before the assay. The mating pair was then placed together in a single tank (the light intensity was about 600–700 lx under breeding condition) the next morning, and mating behavior was recorded for 5 min. We counted the number of courtship displays. We performed a quality check on female reproductive states following our previously described procedure [15]. We determined whether the females had spawned fertilized eggs 30 min after recording the movie. If the females had not produced fertilized eggs at that time, we judged that the females were not in a reproductive state, which might be due to stress or lack of food. The percentage of females in a reproductive state was 65%~75% and we did not analyze the data of the females not in the reproductive state.
This behavioral assay was performed as previously described [17, 18] with minor modifications. We placed three males of the same strain into a single tank (13 cm x 19 cm x 12 cm (height)). Water depth was about 7–8cm, the light intensity was about 600–700 lx under breeding condition and we allowed them to adapt to the apparatus for 60~70min. Movement of each fish was recorded for 5 min. We defined three behavioral components, “bite” and “attack” [17, 18] as “aggressive behaviors”. The difference between “aggressive behaviors” in the previous work [17, 18] was that in the present study we did not consider approaching and threatening behaviors that did not include touching each other (chase, replace, and frontal–lateral display) as “aggressive behaviors”, because it is very difficult to discriminate these behaviors from shoaling-like behavior that V1a2 mutants frequently exhibit in the male group. We counted aggressive behaviors of three fish (Figs. 4A, S18) or the focal fish (S19 Fig.).
|
10.1371/journal.pntd.0007138 | Knowledge, attitudes and practices with regard to schistosomiasis prevention and control: Two cross-sectional household surveys before and after a Community Dialogue intervention in Nampula province, Mozambique | The Community Dialogue Approach is a promising social and behaviour change intervention, which has shown potential for improving health seeking behaviour. To test if this approach can strengthen prevention and control of schistosomiasis at community level, Malaria Consortium implemented a Community Dialogue intervention in four districts of Nampula province, Mozambique, between August 2014 and September 2015.
Cross-sectional household surveys were conducted before (N = 791) and after (N = 792) implementation of the intervention to assess its impact on knowledge, attitudes and practices at population level. At both baseline and endline, awareness of schistosomiasis was high at over 90%. After the intervention, respondents were almost twice as likely to correctly name a risk behaviour associated with schistosomiasis (baseline: 18.02%; endline: 30.11%; adjusted odds ratio: 1.91; 95% confidence interval: 1.14–2.58). Increases were also seen in the proportion of people who knew that schistosomiasis can be spread by infected persons and who could name at least one correct transmission route (baseline: 25.74%; endline: 32.20%; adjusted odds ratio: 1.36; 95% confidence interval: 1.01–1.84), those who knew that there is a drug that treats the disease (baseline: 29.20%, endline: 47.55%; adjusted odds ratio: 2.19; 95% confidence interval: 1.67–2.87) and those who stated that they actively protect themselves from the disease and cited an effective behaviour (baseline: 40.09%, endline: 59.30%; adjusted odds ratio: 2.14; 95% confidence interval: 1.40–3.28). The intervention did not appear to lead to a reduction in misconceptions. In particular, the belief that the disease is sexually transmitted continued to be widespread.
Given its overall positive impact on knowledge and behaviour at population level, Community Dialogue can play an important role in schistosomiasis prevention and control. The intervention could be further strengthened by better enabling communities to take suitable action and linking more closely with community governance structures and health system programmes.
| Schistosomiasis is a parasitic neglected tropical disease that affects around 190 million people worldwide, causing chronic ill health and disability. Central to its prevention and control are the acceptance of health interventions such as the distribution of drugs on a mass scale and the adoption of good hygiene and sanitation practices in communities where the disease thrives. One promising method for promoting such behaviours is the Community Dialogue Approach, which involves training volunteers to host regular community meetings, where local health concerns are discussed and culturally appropriate solutions are agreed upon. In 2014/15, Malaria Consortium implemented a Community Dialogue intervention in four districts of Nampula province, Mozambique, to improve knowledge, attitudes and practices with regard to schistosomiasis prevention and control. To assess the effectiveness of the approach, two household surveys were conducted. Results show that before the intervention, knowledge of how schistosomiasis is acquired, transmitted, prevented and treated was low. After the intervention, knowledge and self-reported adoption of positive behaviours had improved substantially, demonstrating that Community Dialogue can play a central role in strengthening disease prevention and control. The approach could be strengthened by further empowering communities to take action and reducing deeply-held misconceptions.
| Schistosomiasis is one of the most common parasitic diseases, with approximately 190 million people infected worldwide [1], causing over 100,000 deaths [2] and the loss of over 1.8 million disability-adjusted life years every year [3]. The disease is caused by worms of the Schistosoma (S.) genus, which use certain types of freshwater snails as intermediary hosts. The infected snails release cercariae, which infect humans by penetrating the skin when they come into contact with freshwater that contains the parasite. Inside the human body, the cercariae go through several stages and eventually grow into adult worms, which live in infected people’s blood vessels. Here they undergo sexual reproduction, producing and releasing eggs into the lumen of the gut, bladder or urinary tract. The parasite’s life cycle is completed when infected people excrete its eggs while urinating or defecating near freshwater, where the eggs hatch and the resultant miracidia infect the snails [4,5]. One of the countries most affected by schistosomiasis is Mozambique, where the main parasite species present is S. haematobium. Countrywide, the prevalence of S. haematobium infection among school-age children is 47% [6] and 18 million of the total population of 23.5 million require preventive chemotherapy [7]. S. haematobium causes urogenital schistosomiasis, of which the defining symptom is blood in the urine. Pathology can also include scarring, calcification, bladder cancer, and occasional ectopic egg granulomas in brain or spinal cord [4,5].
In areas with moderate to high transmission of schistosomiasis, a key disease control strategy is preventive chemotherapy using praziquantel, typically delivered to at-risk populations in the form of periodic, large-scale mass drug administration (MDA) [8]. Other prevention and control strategies recommended by the World Health Organization include increasing access to and use of safe water, improving sanitation and hygiene, and implementing snail control [9]. Individual and community perceptions of schistosomiasis are likely to have a significant impact on schistosomiasis transmission and uptake of disease prevention and control interventions. For example, preventive chemotherapy programmes are more likely to be successful if they are adapted to local circumstances to build trust, address fear of side effects and correct misconceptions about the treatment among target populations [10]. Social and behaviour change and community engagement are therefore crucial elements of successful MDA campaigns [11] and will also have an important role to play in improving adoption of positive health seeking behaviours more generally [12].
The Community Dialogue Approach [13] is a promising social and behaviour change and community engagement intervention, which has shown potential for improving uptake of health services and promoting recommended behaviours in the context of Integrated Community Case Management of childhood illness [14]. In this approach, volunteers from within the community receive a brief training on a health issue and group facilitation skills. Equipped with a set of visual tools, the volunteers then host regular community meetings to discuss the health issue. Each meeting comprises three stages:
The Community Dialogue Approach targets both individual and social behaviour determinants [15]: It seeks to increase individuals’ knowledge and awareness of a health issue by sharing key messages through the volunteers and visual tools (during the “explore” and “identify” stages), but it also aims to influence social norms through public dialogue and collective decision making (during the “identify” and “decide” stages). Although only a small proportion of the population will actively participate in community meetings, knowledge is expected to spread through word of mouth, while changing social norms are expected to affect the wider community to which the norms apply.
To assess if the Community Dialogue Approach can strengthen prevention and control of schistosomiasis at community level, Malaria Consortium conducted an implementation research study in partnership with the Republic of Mozambique’s Ministério de Saúde (Ministry of Health) and the Direcção Provincial de Saúde (Provincial Health Authority) in Nampula province. Between August 2014 and September 2015, 157 volunteers facilitated regular community dialogue meetings in four districts of Nampula province, discussing the causes and symptoms of schistosomiasis and how the disease can be prevented and controlled, including MDA and adoption of improved water, hygiene and sanitation practices. The main tool developed for this intervention was a flipchart containing images illustrating key intervention messages. Approximately 1,500 community meetings were conducted, each typically comprising between 25 and 45 participants.
To evaluate the intervention under programmatic conditions, the implementation research study drew on a range of sources, including qualitative and process evaluation data, mainly focussing on the feasibility and acceptability of the Community Dialogue Approach. Another component of the study explored the intervention’s impact on knowledge, attitudes and practices (KAP) at population level, testing the assumption that the intervention will have impact beyond those who actively participate in community meetings. To this end, two cross-sectional household surveys were conducted in the four study districts, immediately before (baseline) and after (endline) implementation of the Community Dialogue intervention. Baseline survey results have been published [16] and showed that while awareness of the disease was high, correct knowledge of how it is acquired, transmitted and prevented was low. Few respondents reported that their children had received MDA and the misconception that schistosomiasis is sexually transmitted was widespread. Only a minority of respondents reported practicing protective behaviours. This paper presents unpublished endline survey data, demonstrating changes in population-level KAP over the course of the Community Dialogue intervention. A paper describing the intervention in more detail, as well as presenting qualitative and process evaluation data exploring its feasibility and acceptability will be published elsewhere.
The study was conducted in Nampula province (Fig 1A), which records the highest provincial schistosomiasis prevalence figures in Mozambique, with around 78% of school-age children infected with S. haematobium [6]. The four study districts of Eráti, Mecubúri, Mogovolas and Murrupula (Fig 1B) were selected purposively in consultation with the Direcção Provincial de Saúde. Prevalence of schistosomiasis in the districts was known to be high and they were considered to be comparable with regard to the population’s exposure to common risk factors. Specifically, large proportions of the population depend on subsistence agriculture in or near freshwater sources, and there is poor access to clean water for bathing and washing. According to the 2007 Mozambique census, the total population in the four study districts is approximately 839,000 [17].
An integrated neglected tropical disease (NTD) control programme is implemented in all study districts and regular MDA has been conducted since 2009. According to an independent MDA coverage survey carried out in 2015, coverage was comparatively high across northern Mozambique, but lower in Nampula than in any of the other provinces [18]. Only two of the four districts (Mecubúri and Murrupula) received MDA for schistosomiasis while the community dialogue intervention was implemented. These campaigns targeted the entire population over five years of age, whereas MDA conducted before the start of the intervention had targeted school-age children between five and 14 years. Reported programme coverage was 64% and 81% respectively. According to the Direcção Provincial de Saúde, no other programmes targeting schistosomiasis prevention and control were implemented in the study area during the study period.
Baseline and endline KAP surveys were designed as cross-sectional household surveys, with the four study districts considered as one sampling domain. All households were eligible for selection. It was calculated that, when comparing proportions between baseline and endline, a sample size of 389 households was needed to give 80% power to detect a change of at least 10%, conservatively assuming a percentage of 50% at baseline. In order to adjust for confounders, non-response and design effect, the sample size was more than doubled to an intended sample size of 800 households for each survey.
Using proportional-to-size methods, 40 out of the 68 enumeration areas in the four study districts used in the 2007 census were randomly selected. Enumeration areas are based on localidades, the lowest level of the central state administration. Some larger localidades are further subdivided into bairros, which loosely correspond to villages and communities. In a second step, 20 households were selected in each of the 40 enumeration areas, using a random sampling approach adapted from the one recommended for malaria indicator surveys [19]. Household sampling was based on lists of households obtained from community leaders. Taking into account available time and budget, it was not possible to perform household mappings, but field teams were instructed to discuss the reliability of the household lists with community leaders and correct inaccuracies before commencing the sampling process. In each selected household, the household member best placed to answer questions about the household’s health as nominated by the acting household head was interviewed. Only household members over 18 years were eligible. Participation in a Community Dialogue meeting was not a selection criterion. If selected households could not be located, no eligible respondent was available despite at least three repeated visits to the household or respondents declined to be interviewed, the household was dropped from the sample without replacement. To avoid bias and for operational reasons, different samples were selected for the two surveys. It is therefore unlikely that the households sampled or individuals interviewed at baseline and endline were identical.
Baseline data were collected in July 2014, just before the start of the Community Dialogue intervention. The endline survey was conducted in December 2015, shortly after completion of the intervention. On both occasions, data were collected by five field teams, each comprising four researchers and one supervisor. Researchers were typically educated to college level, while supervisors were required to have completed a university degree. All field team members were native speakers of Macua, the language commonly spoken in the study districts. Before the surveys, they were trained on the survey procedures, data collection tools and interview techniques. Consent procedures and issues relating to ethical data collection were also discussed. Researchers were trained for three days and supervisors for four days, including one day in the field to conduct a mock survey.
A structured face-to-face questionnaire (S2 Appendix) was developed in English and subsequently translated into Portuguese. Survey questions were further translated into Macua. Accuracy of the Macua translation was checked with researchers and supervisors during the baseline training. The mock survey conducted as part of the baseline training was used to pre-test the questionnaire. Field teams were instructed to establish the most appropriate local term for schistosomiasis in each community where the survey was conducted in consultation with district health staff and community leaders and to use this term throughout the interview. Otherwise, researchers read out questions exactly as provided. Responses were recorded on paper, assigning them to pre-defined answer categories, which were provided in Portuguese at the request of the field researchers as there is no tradition of reading and writing in Macua. After asking questions about the drug that treats schistosomiasis, field researchers informed all respondents that the name of the drug is praziquantel and showed sample tablets before proceeding to ask whether any of the children living in the household had ever received the drug. Only respondents with children living in their households were asked this question, as prior to the baseline survey, MDA had exclusively targeted school-age children. Table 1 illustrates how the topics covered by the questionnaire relate to the three stages of each Community Dialogue meeting.
Data were independently double-entered into EpiData 3.1 (EpiData Association) by data entry officers who had attended a one-day training. Where differences between first and second entry were detected, records were verified against the paper questionnaire. Data were further checked for consistency and prepared for analysis using STATA Version 12 (StataCorp LP). Responses recorded under ‘other’ were reviewed by senior members of the study team and either re-assigned to a pre-defined answer category, assigned to a newly created category or left in the ‘other’ category. The survey procedures in STATA were used to account for the study design. All percentages reported are population average estimates. Only results considered programmatically relevant are reported in this paper, but more detailed survey responses, including 95% confidence intervals (CI), at baseline and endline can be found in S3 Appendix.
To examine the association between baseline and endline KAP results, a multivariate logistic regression analysis was conducted for eleven key indicators. See S4 Appendix for an overview of how the key indicators were operationalised. Odds ratios (ORs) were calculated to provide a quantifiable measure of the increased likelihood of respondents having correct KAP after, compared with before the intervention. ORs were adjusted for sex, education and district, the three socio-demographic characteristics consistently associated with significant KAP differences at baseline [16]. Unadjusted ORs for key indicators are reported in S5 Appendix. Statistical significance of unadjusted and adjusted ORs was determined by a Wald p-value of <0.05.
Ethical approval for the study, including the consent procedures used for the two KAP surveys, was granted by the University of Leeds School of Medicine Research Ethics Committee (SoMREC/13/071) and the Comité Nacional de Bioética para Saúde in Mozambique (42/CNBS/2014). Participation in the surveys was voluntary and informed written consent was taken from all respondents. All data were kept confidential and have been anonymised.
At baseline, five of the 800 randomly selected households could not be located or a suitable respondent was not available despite repeated visits to the household. A further four households were retrospectively excluded from the analysis because respondents’ ages were recorded as under 18. A total of 791 respondents were therefore included in the analysis. At endline, three of the 800 selected households could not be located or a suitable respondent was not available. A further five households were excluded because respondents’ ages were recorded as under 18, resulting in a total of 792 respondents included in the analysis. None of the selected households declined to be interviewed. Table 2 shows survey respondents’ socio-demographic characteristics at baseline and endline.
Adjusted (aOR) and unadjusted ORs for all key indicators were generally found to be similar. There was therefore no confounding by sex, education or district and, with the exception of children who have taken praziquantel, results by socio-demographic respondent characteristics will not be reported in this paper. However, key indicators analysed by sex, education and district can be found in S6 Appendix.
Previous studies showed that Community Dialogue can bring about changes in social norms and improve health seeking among participants [20]. This study provides good evidence that the approach can also have a positive impact on knowledge at population level. General awareness of schistosomiasis remained universally high at over 90%, possibly as a result of regular MDA campaigns over a number of years. High levels of awareness have been found consistently in other surveys across sub-Saharan Africa [21–25], including one in neighbouring Cabo Delgado province, Mozambique [26]. A recently published systematic review on schistosomiasis KAP in sub-Saharan Africa concluded, however, that high awareness is typically coupled with poor knowledge of causes, prevention and control [27]. This pattern was also found in the four study districts at baseline [16] and it was therefore encouraging to see generally improved knowledge at endline. Specifically, significant improvements were seen for three key components addressed by the Community Dialogue intervention:
It was also noticeable that for many questions exploring respondents’ knowledge, the proportion of those who indicated that they did not know how to answer the question decreased. However, some decreases in specific areas of knowledge were also observed. For example, fewer respondents knew that infected people can spread schistosomiasis. From the responses recorded by field researchers in the ‘other’ category, it was evident that many respondents were not clear about the difference between infection and transmission, which has also been cited by several other studies examining community-level knowledge of schistosomiasis [23,28]. This is a complex distinction requiring fairly sophisticated understanding of the parasite’s transmission cycle, which, moreover, may not influence people’s motivations and resulting behaviour. A more concerning finding was that knowledge of specific prevention and treatment mechanisms decreased, most alarmingly a reduction in knowledge of defecation by freshwater as a transmission route and awareness of the use of latrines and good hygiene as prevention mechanisms. This suggests that training materials and tools developed for the Community Dialogue intervention will need to be reviewed to ensure less well-understood aspects of disease prevention and control are adequately communicated. It also suggests that the “identify” stage of this Community Dialogue intervention, in which participants are encouraged to discuss locally appropriate actions to prevent and control schistosomiasis, needs to be strengthened, for example by providing more detailed and specific guidance with regard to how communities can actively contribute to improved disease prevention and control.
It was also noticeable that only around 5% of respondents cited MDA as a prevention and control mechanism. However, “treatment for all infected people” was frequently cited, which may reflect a misinterpretation of the rationale behind the approach, especially in a context where infection rates are very high. The surveys also found a decrease in the number of respondents who stated that they would want their children to receive praziquantel through MDA, which may be an unintended side effect of strengthening community members’ self-efficacy. It is worth noting, however, that the vast majority of respondents remained supportive of MDA. Nevertheless, the intervention may need to put more emphasis on the safety and community-level benefits of MDA during the “identify” stage.
The overall increase in correct knowledge of schistosomiasis did not appear to lead to a significant reduction in misconceptions. In particular, the belief that the disease is sexually transmitted continued to be widespread. This misconception has been reported by a number of studies in a range of countries [21,22,24] and is perhaps not surprising given that some of the most common symptoms of urogenital schistosomiasis affect the reproductive organs. The persistence of traditional beliefs despite adoption of biomedically correct disease causality models has been well documented, for example with regard to malaria and HIV/AIDS [29,30]. It may be necessary to more directly address known misconceptions in the intervention’s key messages communicated during the “explore” stage, which involves participants’ exchanging experiences and challenging each other’s ideas and concepts with the help of the Community Dialogue volunteers. Training materials and intervention tools should also be reviewed in light of survey findings to ensure they adequately address and do not entrench misconceptions, especially those that might obstruct the adoption of positive behaviours.
Results with regard to the adoption of protective behaviours were mixed. While those who reported actively doing something were significantly more likely to cite an effective behaviour at endline, a majority of around 60% of respondents did not report practicing any protective behaviours. The proportion of those who stated that they do not practice a protective behaviour because they do not know what they can do also remained high at around 80%. It was also concerning that while there was a significant increase in respondents who reported that their children had received praziquantel, this proportion remained very low at 15%, despite MDA campaigns in two of the four study districts while the Community Dialogue intervention was implemented. Two factors may have played a role in limiting the intervention’s population-level impact on behaviour. First, it is well known that improved knowledge does not automatically translate into increased uptake of recommended practices [31], especially where interventions seek to address complex sets of behaviours that are entrenched and structured by poverty, livelihood patterns and social environments [32]. While the Community Dialogue Approach encourages communities to collectively reach decisions about tackling health issues, it is likely that, in many communities, there was an initial focus on increasing knowledge. Second, behaviour change is a non-linear process that unfolds over time [33]. Approximately 14 months of implementing a Community Dialogue intervention may not be sufficient to affect widespread behaviour change.
The limited impact on adoption of positive health seeking behaviours therefore points to a need to strengthen the “identify” and “decide” stages of the Community Dialogue intervention, in which communities were asked to publicly debate how the community could improve schistosomiasis prevention and control and to collectively commit to a plan of action. The intervention could, for example, give more concrete examples of the kind of action people could take, assist planning for and monitoring of actions taken by the community and better embed this process in the existing health system and community governance mechanisms to strengthen social accountability. The intervention should also aim to create stronger links between communities and health care providers or other relevant programmes, such as those addressing water, sanitation and hygiene (WASH), who could assist communities in making decisions and taking action. This could, for example, be informed by the principles of the Enhanced Development Governance model, which has recently been described for a pilot project in Tanzania that sought to strengthen community participation for WASH programmes [34]. Crucially, the Community Dialogue intervention should also link more closely with the national and provincial MDA programme, possibly by giving the volunteers who facilitate the community dialogues a role in planning and delivering campaigns within their community.
The KAP surveys aimed to detect changes in knowledge, attitudes and practices with regard to schistosomiasis prevention and control at population level over the course of a Community Dialogue intervention. They were not designed to:
Another limitation relates to the languages used to develop survey tools and to conduct the survey. While the quality of translations of data collection tools and researcher training materials was checked repeatedly, it is possible that subtleties got lost in the translation chain from English to Portuguese to Macua. Similarly, though care was taken to use appropriate local terminology and formative research carried out before the baseline survey concluded that local terms for schistosomiasis broadly concur with the biomedical definition of the disease, it cannot be ruled out that some respondents may have referred to disparate disease concepts.
Finally, responses regarding adoption of practices need to be interpreted bearing in mind that the survey relied on self-reporting. It was not possible to objectively verify survey participants’ responses. In general, social desirability bias may have led respondents to give answers they considered more socially acceptable. While this would have applied at both baseline and endline, the pressure to give socially desirable responses may have been stronger following an intervention designed to shape social norms.
The survey results are thought to be representative of the population in the four study districts. As the study area shares many characteristics of predominantly rural, resource-poor areas in sub-Saharan Africa and our findings reflect those of other studies in similar settings, the results reported in this paper are thought to have wider applicability. We therefore believe that, given its overall positive impact on knowledge, attitudes and practices, the Community Dialogue Approach can play an important role in affecting positive social and behaviour change, a key requirement for improved disease prevention and control identified by a recent systematic review [27]. The approach goes beyond more established community engagement approaches for NTD control, such as community-directed delivery of MDA [35] or involving community health workers in the detection of suspected NTD cases [36]. Rather than being community-directed or community-based, it is community-owned and has the potential to serve as a platform for community participation, conceived as a process rather than an outcome [37], beyond a single disease or control intervention. More research is needed, however, to investigate the intervention’s mechanisms of impact, trace the spread of information, as well as uncover the conditions under which Community Dialogue can correct misconceptions and trigger sustainable behaviour change. Further research should also include testing of the required intervention dose and reach to determine the number of facilitators required per population unit, the percentage of the population that needs to actively participate in community meetings, or the required number and frequency of community meetings.
|
10.1371/journal.pgen.1003577 | ABI4 Regulates Primary Seed Dormancy by Regulating the Biogenesis of Abscisic Acid and Gibberellins in Arabidopsis | Seed dormancy is an important economic trait for agricultural production. Abscisic acid (ABA) and Gibberellins (GA) are the primary factors that regulate the transition from dormancy to germination, and they regulate this process antagonistically. The detailed regulatory mechanism involving crosstalk between ABA and GA, which underlies seed dormancy, requires further elucidation. Here, we report that ABI4 positively regulates primary seed dormancy, while negatively regulating cotyledon greening, by mediating the biogenesis of ABA and GA. Seeds of the Arabidopsis abi4 mutant that were subjected to short-term storage (one or two weeks) germinated significantly more quickly than Wild-Type (WT), and abi4 cotyledons greened markedly more quickly than WT, while the rates of germination and greening were comparable when the seeds were subjected to longer-term storage (six months). The ABA content of dry abi4 seeds was remarkably lower than that of WT, but the amounts were comparable after stratification. Consistently, the GA level of abi4 seeds was increased compared to WT. Further analysis showed that abi4 was resistant to treatment with paclobutrazol (PAC), a GA biosynthesis inhibitor, during germination, while OE-ABI4 was sensitive to PAC, and exogenous GA rescued the delayed germination phenotype of OE-ABI4. Analysis by qRT-PCR showed that the expression of genes involved in ABA and GA metabolism in dry and germinating seeds corresponded to hormonal measurements. Moreover, chromatin immunoprecipitation qPCR (ChIP-qPCR) and transient expression analysis showed that ABI4 repressed CYP707A1 and CYP707A2 expression by directly binding to those promoters, and the ABI4 binding elements are essential for this repression. Accordingly, further genetic analysis showed that abi4 recovered the delayed germination phenotype of cyp707a1 and cyp707a2 and further, rescued the non-germinating phenotype of ga1-t. Taken together, this study suggests that ABI4 is a key factor that regulates primary seed dormancy by mediating the balance between ABA and GA biogenesis.
| Seed dormancy prevents or delays germination in maturated seeds. The optimal level of seed dormancy is a valuable trait for agricultural production and post-harvest management. High ABA and low GA content in seeds promote seed dormancy. However, the precise molecular mechanisms controlling seed dormancy and germination remain unclear. We found that ABI4, the key transcription factor in the ABA signaling pathway, indeed controls primary seed dormancy. This result contradicts the previous conclusion that ABI4 is not involved in the control of seed dormancy. Several lines of evidence support our conclusion. For example, detailed physiological analysis of the germination of abi4 seeds that were harvested immediately and stored for various periods of time and subjected to various treatments allowed us to conclude that ABI4 negatively regulates primary seed dormancy. The molecular mechanism responsible for this control is as follows: ABI4 directly or indirectly regulates the key genes of the ABA and GA biogenesis pathways, which then regulates the ABA and GA contents in seeds. Importantly, further genetic interactions between CYP707A1, CYP707A2, GA1, and ABI4 also support our conclusion.
| Seed dormancy prevents or delays the germination of maturated seeds even when conditions are favorable for germination [1]–[3]. Seed dormancy is an important trait for diverse, important crop species including rapeseed, wheat, corn and rice, because seed dormancy inhibits pre-harvest spouting or vivipary [4]. Vivipary usually causes great economic loss to cereal production, including losses in seed quantity and quality, especially in humid regions worldwide [5], [6]. On the other hand, deep seed dormancy is problematic, especially in the horticultural and forest industries, and chemical treatments may be required to promote germination [7]. Thus, the optimal level of seed dormancy is a valuable trait for agricultural production. Therefore, it is essential to understand the precise molecular mechanisms that control seed dormancy and germination.
Diverse endogenous and environmental factors including phytohormones, nutrients, temperature and light affect seed dormancy through different pathways [3], [8]. Extensive studies have shown that abscisic acid (ABA) and gibberellin acid (GA) are the primary endogenous factors that regulate the transition from dormancy to germination, and they regulate this process antagonistically [1], [2], [9]–[11]. ABA is essential for the induction and maintenance of seed dormancy, while GA is required for the release of dormancy and for the initiation of seed germination [9], [10]. In line with this conclusion, some ABA-deficient mutants such as nced6, nced3, nced5, nced9, aba2 and aao3 are better able to germinate than WT seeds [9], [12]. In support of these observations, the overexpression of the ABA biosynthesis gene ABA2 enhances ABA accumulation and maintains deep seed dormancy [13]. Further, overexpression of other ABA biosynthesis genes, NCED6 and NCED9, even inhibits precocious germination of developing seeds due to increased ABA biogenesis [14]. By contrast, some ABA metabolic pathway mutants, such as cyp707a1, cyp707a2 and cyp707a3, accumulate higher ABA levels than WT and subsequently exhibit hyperdormancy in seeds [15]–[17]. In addition to ABA content, ABA signaling also positively regulates seed dormancy [2], [3]. Although ABI1 and ABI2 are negative regulators in the ABA signaling pathway, the abi1-1 and abi2-1 mutants show the reduced dormancy levels [18]. This phenotype results from dominant-negative mutations and therefore, these two PP2Cs (Protein Phosphatases type 2C) are unable to bind to ABA receptors [19], [20]. In addition, the abi3 mutant also shows reduced seed dormancy levels [21]. Furthermore, the allelic mutant abi3-3 even rescues the non-germinating phenotype of ga1 in the absence of exogenous GA treatment, indicating that ABI3 is a negative regulator of GA biosynthesis [22]. Although a previous study has concluded that abi5 does not reduce seed dormancy [23], other studies have shown that this gene negatively regulates seed germination [24], [25].
In contrast to ABA, GA negatively regulates seed dormancy [2], [3]. Mutants severely defective in GA biosynthesis such as ga1 show deep seed dormancy and fail to germinate in the absence of exogenous GA [26]. On the other hand, mutants defective in GA 2-oxidases (GA2ox), which deactivate bioactive GA, exhibit reduced seed dormancy and germinate normally, even in the dark [27]. Mutations in two negative regulators in the GA signal transduction pathway, rgl2 (RGA-LIKE2) and spy (SPINDLY), rescue the non-germination phenotype of ga1-3 in the absence of exogenous GA [28], [29]. Combined with the conclusion that ABA and GA regulate seed dormancy antagonistically [3], the ability to synthesize GA is enhanced in the aba2 mutant, indicating that ABA is involved in the suppression of GA biogenesis in both developing and imbibed seeds [9]. These pioneering studies demonstrated that ABA and GA biogenesis and signaling play key roles in the control of seed dormancy and germination. However, the detailed molecular mechanism by which the crosstalk between ABA and GA at the hormone biogenesis level regulates seed dormancy is largely unknown.
ABI4 encodes an AP2/ERF transcription factor, which is an enhancer in the ABA signal transduction pathway that functions especially during seed development and germination [23], [30], [31]. Furthermore, ABI4 is also involved in other aspects of plant development including lipid mobilization from the embryo [32], glucose responses [33], [34], salt responses [35] and the mitochondrial and chloroplast-nucleus retrograde signaling pathways [36]–[38]. Most recently, ABI4 was found to regulate ABA and cytokinin inhibition of lateral root development by reducing polar auxin transport [39], as well as ABA- and jasmonate-dependent signaling pathway crosstalk [40] and the nitrogen deficiency stress response [41]. Except in young seedlings, the ABI4 transcript level is relatively low through most stages of vegetative growth but high in both developing and imbibed seeds [31]. The abundance of ABI4 protein is partially regulated by the 26S-proteasomal pathway [42]. These excellent studies demonstrate that ABI4 is a versatile factor, which functions in diverse signaling pathways and is tightly regulated at the post-transcriptional level. However, the role of ABI4 in crosstalk between ABA and GA has not yet been elucidated.
As described above, many mutants in which the ABA signal is attenuated, such as abi1-1, abi2-1 and abi3, exhibit the reduced seed dormancy phenotype [18], [21]. Although this protein is a positive regulator of the ABA signaling pathway, however, previous studies have concluded that ABI4 has no effect on seed dormancy [23], and this opinion has been accepted in the field [3], [43]. Recently, two studies have demonstrated that a mutation in a double-repeat AP2 domain transcription factor, CHOTTO1, results in reduced primary seed dormancy, and, interestingly, ABI4 likely acts upstream of CHOTTO1 in the genetic pathway [10], [44]. On the other hand, the ABI4 transcript level is relatively low at almost all growth stages except during seed maturity and germination [31]. These studies inspired us to reconfirm the effects of ABI4 on primary seed dormancy as well as postgerminative growth.
Here, we show that abi4 mutant seeds indeed exhibited reduced primary seed dormancy and increased cotyledon greening. The differences in germination rates and cotyledon greening between abi4 and WT decreased moderately after stratification. After-ripening treatment caused the rates of germination and cotyledon greening to be comparable between abi4 and WT. In line with these results, the ABA content in abi4 dry seeds was significantly lower than that in WT, but the ABA levels were comparative after stratification treatment. Consistently, the GA level in abi4 seeds was upregulated compared to WT. Further analysis showed that abi4 was resistant to exogenous paclobutrazol (PAC), a GA biosynthesis inhibitor, while OE-ABI4 was sensitive to PAC during germination, and exogenous GA rescued the delayed germination phenotype of OE-ABI4. The qRT-PCR assay also showed that the transcript levels of some GA biosynthesis and ABA inactivation genes were upregulated in germinating abi4 seeds, while some GA inactivation and ABA biosynthesis genes were downregulated. ChIP-qPCR and transient expression assays showed that ABI4 indeed inhibits CYP707A1 and CYP707A2 transcription by directly binding to these promoters, and the CCAC cis-elements are essential for this repression. Further genetic analysis showed that abi4 restored the delayed germination phenotype of cyp707a1 and cyp707a2, and importantly, mutation in ABI4 also rescued the non-germinating phenotype of ga1-t even in the absence of exogenous GA treatment, reconfirming that ABI4 is a negative regulator of GA biogenesis and a positive regulator of ABA biosynthesis during seed germination. Taken together, this study demonstrates that ABI4 plays pivotal and complex roles in fine-tuning the ABA/GA balance to control primary seed dormancy.
Occasionally we found that the abi4-1 seeds germinated more quickly than WT seeds when the siliques fell onto the surface of the soil. Thus, we decided to investigate the effect of the ABI4 gene on seed dormancy. The abi4-1 (hereafter referred to as abi4) mutant was obtained from ABRC (Arabidopsis Biological Resource Center at the Ohio State University; stock number CS8104). This mutant contains a point mutation in the open reading frame in which the 469th base G is deleted, resulting in a frame shift at codon 157 and producing a protein containing the predicted DNA binding and dimerization domains but lacking the presumed activation domain [31].
To investigate the effect of abi4 on seed dormancy, the germination of the abi4 mutant and WT seeds was scored. Using seeds subjected to one week of dry storage, the germination rate of abi4 seeds was clearly shown to be significantly higher than that of WT without 4°C stratification treatment (Figure 1A). At 1.5 days after sowing, the endosperms of most abi4 seeds were ruptured, and radicles emerged from some seeds, while the testas of WT seeds had not even ruptured at this time (Figure 1A). The germination rate was nearly 80% for abi4 seeds at day 2; however, the germination rate for WT seeds was less than 40% at this time point (Figure 1A). Consistent with the reduced dormancy level, the abi4 mutant also exhibited markedly faster cotyledon greening than the WT (Figure S1A). In addition, it is noteworthy that at 4.5 days after sowing, the germination rates of abi4 and WT reached 100% (Figure 1A). Taken together, the results of this time-course experiment show that the abi4 mutant indeed exhibits reduced seed dormancy.
Previous studies have shown that stratification treatment reduces primary seed dormancy and thus promotes seed germination [10], [45]. Therefore, we also investigated the effect of stratification on primary seed dormancy in abi4. When seeds subjected to one week of dry storage were stratified for 3 days, the differences in the rates of germination and cotyledon greening between abi4 and WT were moderately reduced (Figures 1B, S1B), compared with the larger difference shown in Figures 1A and S1A. However, the percentage of germination and cotyledon greening of abi4 was still higher than that of WT, and the growth rate of the radicle of abi4 was significantly faster than that of WT (Figures 1B, S1B). Next, when we examined seeds subjected to two weeks of dry-storage, the similar trends were detected (Figures 1C, S1C), and the differences between abi4 and WT decreased moderately with stratification treatment (Figures 1D, S1D). Indeed, the abi4 seeds subjected to short-term storage germinated more quickly than WT, especially without stratification treatment (Figure 1A, 1B, 1C and 1D). Subsequently, the effect of after-ripening on primary seed dormancy was investigated. The faster germination phenotype of abi4 seeds was abolished when the seeds were fully after-ripened, either with or without stratification treatment (6-month dry-storage; Figure 1E, 1F). Consistent with these results, we also did not detect differences in cotyledon greening rates between abi4 and WT when we employed fully after-ripened seeds (Figure S1E, S1F). Altogether, these results suggest that abi4 reduces primary seed dormancy.
A reduced primary seed dormancy level usually results in preharvest sprouting or vivipary in cereals, especially if moist conditions are encountered [46].Therefore, we tested whether the abi4 mutant exhibits vivipary in developing seeds using a protocol employed in a previous study [14]. The results show that abi4 seeds in developing siliques indeed germinated more quickly than WT both on 1/2 MS medium and on soil (Figure 1G). In particular, at 8 days after sowing, only a few seeds in WT siliques germinated, while young abi4 seedlings were already established (Figure 1Ga, 1Gb). At 14 days after sowing, most of the abi4 and WT siliques produced seeds that germinated, and the cotyledons greened (Figure 1Gc, 1Gd), indicating that the seed vigor in these developing siliques was normal. Therefore, we reasoned that the difference in germination rate between abi4 and WT siliques resulted from different seed dormancy levels.
To further confirm that the reduced primary dormancy level phenotype of abi4 resulted from a mutation in the ABI4 locus, we obtained another T-DNA insertion mutant in this locus from ABRC (the stock name is SALK_080095, hereafter referred to as abi4-t). A previous study showed that this line is a knockout mutant [47]. Similar to abi4, the decreased seed dormancy level, early germination phenotype of abi4-t was also observed when we analyzed seeds subjected to one week of dry storage (Figure S2A), and the percentage of cotyledon greening of abi4-t was also significantly higher than that of WT seeds (Figure S2B). On the other hand, the faster germination phenotype of abi4-t seeds (compared with WT) was abolished when fully after-ripened seeds were employed (data not shown). The similar phenotype of the two allele mutants further proved that a mutation in the ABI4 locus is indeed responsible for the reduced primary seed dormancy phenotype.
To demonstrate the reproductivity of our experiment, we employed several mutants with seed dormancy phenotype in ABA pathway as controls. Previous study demonstrated that snrk2.2/snrk2.3 obviously reduced seed dormancy level compared to WT [48]. This phenotype is resulted from the impaired ABA signaling in this double mutant [48]. In our growth condition, the reduced seed dormancy phenotype of snrk2.2/snrk2.3 was repeated perfectly (Figure S3A). Notably, the decreased seed dormancy phenotype of snrk2.2/snrk2.3 was stronger than that of abi4 (Figure S3A). Furthermore, the reduced seed dormancy phenotype of abi1-1 and abi2-1 compared to Ler seeds was also detected in the same condition (Figure S3B), which was consistent with our current knowledge [18]. These results demonstrated that the present experimental condition is eligible and reliable.
As described above, abi4 seeds subjected to short-term storage exhibited significantly reduced dormancy compared with WT seeds, but the difference in germination rate between abi4 and WT was decreased moderately or even abolished after stratification or longer period of storage (Figure 1). Since ABA positively regulates seed dormancy [1], and stratification and after-ripening treatment reduce the ABA content [10], we next examined endogenous ABA levels in dry and stratified abi4 seeds using a liquid chromatography–tandem mass spectrometry system. We chose seeds subjected to two weeks of dry storage for this experiment. As expected, the result showed that the ABA content in abi4 was significantly lower than that in WT (Figure 2) when no stratified seeds were analyzed, suggesting that ABI4 positively regulates ABA biogenesis. On the other hand, stratification treatment impairs ABA biosynthesis [49]. Accordingly, after a 3-day stratification treatment, both WT and abi4 mutant seeds contained lower ABA levels, and importantly, the ABA levels were comparable between WT and abi4 (Figure 2). The trend of ABA level in dry or stratified abi4 seeds is similar to that of CHOTTO1, a positive regulator of primary seed dormancy that may acts downstream of ABI4 in a genetic pathway [10], [44]. The hormonal measurements described above revealed that the decreased ABA level in abi4 seeds is at least partially responsible for the reduced primary seed dormancy phenotype of this mutant (Figure 2).
To further investigate the precise mechanism by which ABI4 regulates primary seed dormancy, ABI4-overexpressing plants were generated. The coding region of ABI4 was introduced into the vector pCanG-HA-GFP under the control of the CaMV (Cauliflower mosaic virus) 35S promoter and transformed into WT Arabidopsis. Several independent T3 homozygous lines were identified through qRT-PCR and western blot assays, and two of them were shown (Figure 3A, 3B). Because ABI4 directly promotes ABI5 transcription [50], we examined the ABI5 expression levels in those transgenic lines. The qRT-PCR assay showed that ABI5 transcription was indeed upregulated in these ABI4 overexpressing lines (Figure 3C). Thus, we reasoned that these two overexpressing lines are functional and they were employed in further analysis.
Because both the abi4 and abi4-t mutants showed the reduced primary seed dormancy phenotype (Figures 1, S2), we first tested the seed dormancy level of OE-ABI4 seeds subjected to two-week dry-storage on normal 1/2 MS medium. The results showed that the two independent lines germinated slowly than WT (Figure 3D), and accordingly, the cotyledon greening rates of these two lines were also moderately lower than that of WT (Figure S4A). These results indicate that the seed dormancy level in OE-ABI4 seeds was higher than that of WT, which is in contrast to the phenotype of the both of abi4 mutants.
Our results show that ABI4 positively regulates ABA biogenesis (Figure 2), and a previous study demonstrated that ABA is involved in the suppression of GA biosynthesis in imbibed seeds [9]. Thus, we speculated that the GA level in abi4 was higher than that in WT. To confirm this speculation, we analyzed the responsiveness of abi4 mutant and OE-ABI4 seeds to GA and PAC treatment. Our results showed that OE-ABI4 seeds were sensitive to PAC during germination (Figure 3E) and cotyledon greening (Figure S4B), while abi4 was resistant (Figure 3E, S4B). However, the rates of germination and cotyledon greening among abi4, WT and OE-ABI4 were comparable when we used medium supplemented with exogenous GA (Figure 3F, S4C). The increased resistance of abi4 to the GA biosynthesis inhibitor suggests that this mutant contains higher levels of active GA or possesses stronger GA signaling than the WT [51]. Combined with the fact that exogenous GA can rescue the delayed germination and cotyledon greening phenotypes of OE-ABI4, we proposed that ABI4 attenuates GA biosynthesis to positively regulate seed dormancy.
The responsiveness analysis of abi4 and OE-ABI4 seeds to GA and PAC treatments suggested that ABI4 negatively regulates GA biogenesis (Figure 3E, 3F). Furthermore, because that ABA is involved in the suppression of GA biosynthesis during seed germination [9], and the ABA measurements between abi4 and WT seeds also supported this speculation (Figure 2). Then, we examined the endogenous GA content in abi4 and WT seeds. The result showed that the active GA4 level in abi4 dry seeds was significantly higher than that in WT (Figure 3G), suggesting that ABI4 indeed regulates GA biosynthesis negatively. Combined with the ABA quantification result, the endogenous hormone measurements demonstrated that the decreased ABA level and the increased GA content in abi4 seeds are responsible for the reduced primary seed dormancy of abi4.
To further confirm that ABI4 functions as an attenuator of GA biogenesis during seed germination, we analyzed the effect of the ABI4 mutation on the expression of GA biosynthesis genes and GA inactivation genes in dry and imbibed seeds. The results of qRT-PCR analysis showed that the transcript levels of GA biosynthesis genes, including GA3, GA3ox1, GA20ox1, GA20ox3, KAO1, KAO2 and GA20ox2, were upregulated to varying degrees in abi4 seeds after imbibition (Figure 4A). The expression levels of GA3 and KAO2 in dry abi4 seeds were 2-fold higher than that in WT, and this trend was maintained throughout the imbibition treatment process (Figure 4A). Higher levels of GA3ox1 mRNA were detected in the abi4 mutant after 6 hours of imbibition (Figure 4A). The transcripts of KAO1, GA20ox1, GA20ox2 and GA20ox3 were higher in abi4 than in WT during the entire imbibition process, although the differences were not significant (Figure 4A). By contrast, the transcript level of GA2ox8, a key GA inactivation gene, was lower in the abi4 mutant than in the WT (Figure 4B). The increased expression of GA biosynthesis genes and the decreased expression of GA inactivation genes in imbibed seeds are accordance with the GA measurements in abi4 mutant seeds which contains higher levels of active GA than the WT (Figure 3G). Consistent with this, the RGL3 gene, which encodes a DELLA transcription regulator that represses testa rupture during seed germination [52], was downregulated in both dry and imbibed abi4 seeds (Figure 4C).
Since ABA and GA regulate seed germination antagonistically [2], the expression levels of ABA biosynthesis and inactivation genes in dry and imbibed seeds were also analyzed. Analysis by qRT-PCR showed that the mRNA levels of ABA biosynthesis genes, including NCED2 and NCED3, were downregulated in abi4 (Figure 4D), while the ABA inactivation genes such as CYP707A1, CYP707A2 and CYP707A3 were upregulated (Figure 4E). The higher transcription levels of these three inactivation genes in abi4 were maintained throughout the entire imbibition process. Notably, the expression level of CYP707A2 in abi4 was almost 4-fold higher than that in WT at 6 hours after imbibition (Figure 4E). The high transcription levels of ABA inactivation genes, and the low level of ABA synthesis in abi4, explain the results of ABA measurement (Figure 2). Together, the transcript levels of GA biosynthesis and ABA inactivation genes were upregulated in germinating abi4 seeds, while GA inactivation and ABA biosynthesis genes were downregulated. These results are consistent with the notion that ABI4 negatively regulates GA biosynthesis while positively regulating ABA biogenesis (Figures 2, 3).
Previous studies have demonstrated that ABI4 is a versatile transcription factor that binds to the CACCG motif to promote the expression of some genes; this factor also binds to the CCAC element to directly inhibit the transcription of some genes [38], [50].
To investigate whether ABI4 directly regulates the expression some GA and ABA metabolism genes, we first examined the promoters of the genes described in Figure 4 because the expression levels of these genes were altered in abi4 during germination. CYP707A1, CYP707A2 and CYP707A3 were most interesting because 6, 5 and 7 CCAC motifs were detected in these three promoters, respectively (Figure 5A, 5B and 5C). This inspired us to examine whether ABI4 directly binds to these promoters in vivo. We then conducted a ChIP (chromatin immunoprecipitation)-qPCR assay with the ABI4 transgenic lines to examine whether ABI4 binds to these promoters directly. Because ABI4 binds directly to the promoter of ABI5 [50], a DNA element of the ABI5 promoter was used as positive control. Two independent OE-ABI4 transgenic lines (OE1 and OE2) were subjected to ChIP-qPCR analysis, which produced similar results. We determined that the promoters of CYP707A1 and CYP707A2 were enriched in the chromatin immunoprecipitated DNA using the anti-GFP antibody (Figure 5D), especially the P2 and P3 regions in CYP707A1 and the P5 region in CYP707A2. This result indicates that ABI4 directly binds to the promoters of CYP707A1 and CYP707A2 in vivo. However, we did not detected significant enrichment of all of the elements tested from promoter of CYP707A3 (Figure 5D), although this promoter possesses 7 CCAC motifs (Figure 5C). These results indicate that ABI4 may repress CYP707A1 and CYP707A2 expression by directly binding to the promoters of these genes.
Physiological and molecular evidence support the notion that the biogenesis of ABA and GA during seed germination is affected by ABI4, and ABI4 positively regulates primary seed dormancy. To further confirm this conclusion, we subsequently dissected the genetic relationship between ABI4 and hormone metabolism genes.
GA1 encodes ent- ent-copalyl diphosphate synthase synthase, a key enzyme that catalyzes a relatively early biochemical reaction in the biosynthesis of GA [53], [54]. The ga1 loss-of-function alleles cause GA deficiency and abolish seed germination in the absence of exogenous GA [26], [54]. OE-ABI4 seeds were sensitive to PAC during germination, while abi4 seeds were resistant (Figure 3E), and further, the GA biogenesis was attenuated in abi4 seeds compared to WT (Figure 3G), indicating that GA biosynthesis is indeed negatively regulated by ABI4. Therefore, we examined whether abi4 could rescue the non-germination phenotype conferred by ga1-t. First, we created a double mutant between the abi4 and ga1-t homozygous mutants through genetic crossing. Subsequently, seed germination was analyzed in the abi4, ga1-t and abi4/ga1-t double mutants using seeds subjected to two weeks of dry storage. The results showed that the abi4/ga1-t double mutants germinated normally, and the cotyledons greened normally, even in the absence of exogenous GA, while ga1-t did not germinate under these condition (Figures 6A, S7A). As expected, the exogenous application of GA restored the germination of ga1-t. Furthermore, the abi4/ga1-t double mutant also germinated, and the cotyledons greened slightly more quickly than those of ga1-t in the presence of exogenous GA (Figures 6B, S7B). These results demonstrate that ABI4 indeed negatively regulates GA biogenesis from the view of genetics.
On the other hand, the cyp707a1 and cyp707a2 mutants accumulate higher levels of ABA than the WT and thus exhibit the delayed germination phenotype [15]. Since the ABA level was downregulated in the abi4 mutant (Figure 2) and ABI4 directly inhibits CYP707A1 and CYP707A2 expression by binding to those promoters (Figure 5D to 5H), we tested whether abi4 could rescue the germination defect phenotype of cyp707a1 and cyp707a2. Therefore, abi4/cyp707a1 and abi4/cyp707a2 double mutants were created between abi4 and the homozygous mutant cyp707a1 (SALK_069127) and cyp707a2 (SALK_083966C), respectively. Our results showed that the seeds of these double mutants showed higher germination frequencies than the corresponding cyp single mutants, cyp707a1 and cyp707a2, but lower than abi4 when the seeds were subjected to two weeks of dry storage (Figure 6C, 6D, right panel). Given that ABI4 positively regulates ABA biogenesis, we speculated that the reason responsible for the recovery of abi4/cyp707a1 and abi4/cyp707a2 regarding the delayed germination phenotype of cyp707a2 and cyp707a2 is that the ABA biogenesis is impaired in these double mutants. For this end, we further detected the ABA content in the cyp707a2 single mutant and abi4/cyp707a2 double mutant respectively. Indeed, our results revealed that the ABA level in cyp707a2 is significantly increased compared to WT (Figure S8), which is consistent with the previous study [15]. Importantly, we detected that the ABA content in abi4/cyp707a2 is decreased compared to cyp707a2 single mutant (Figure S8). These results indicate that a mutation in the ABI4 locus recovers the reduced germination potential of cyp707a1 and cyp707a2 through attenuating the ABA biogenesis. Together, these genetic analyses between CYP707A1, CYP707A2, GA1 and ABI4 further confirmed the notion that ABI4 indeed positively regulates ABA biosynthesis and negatively regulates GA biogenesis.
Physiological analysis of germination, hormone measurements, gene expression analysis and biochemical and genetic analysis have demonstrated that a mutation in the ABI4 locus indeed reduces primary seed dormancy, and the molecular mechanism responsible for this phenotype is as follows: ABA biogenesis is downregulated, and GA biosynthesis is upregulated in the abi4 mutants. The present study clarifies that like ABI3, ABI4 also positively regulates primary seed dormancy. Further, this study also strongly suggests and opens up the possibility that ABI4 plays pivotal and complex roles in the crosstalk between ABA and GA in the regulation of primary seed dormancy and early plant development.
Pre-harvest sprouting of diverse cereal seeds usually occurs under humid conditions during harvest time and results in the germination of grains that are still on the mother plant. Sprouting, which results from the reduced dormancy level of crop seeds, lowers the value of crop seeds in terms of both quantity and quality [2], [6]. Therefore, pre-harvest sprouting has attracted increasing amounts of attention from researchers, especially in agronomic regions; the precise molecular mechanism underlying seed dormancy and pre-harvest sprouting is worth exploring.
In the present study, the abi4 seeds obviously germinated significantly more quickly than WT when the seeds were subjected to short-term storage; this mutant even exhibited the vivipary phenotype (Figure 1). On the other hand, it is noteworthy that the percentages of germination of abi4 and WT seeds were comparable at 4.5 days after sowing (all reached nearly 100%; Figure 1A to 1D), which is in accordance with previous result [23]. In a previous study, the germination rate was scored at 5 days after sowing, and the abi4 mutant showed the same degree of dormancy as WT seeds (both genotypes reached 100% germination) [23]. Therefore, ABI4 was thought to have no effect on seed dormancy. Subsequent studies and reviews cited this conclusion [3], [31], [43]. Recently, two studies showed that CHOTTO1 regulates primary seed dormancy positively, and, more interestingly, ABI4 likely acts in the same genetic pathway as CHOTTO1 [10], [44]. Both studies, along with our own occasionally observation that about abi4 germinated more quickly than WT when the siliques fell onto the soil, inspired us to reconfirm the effect of ABI4 on primary seed dormancy. We speculate that the reason for the previous conclusion (that ABI4 has no effect on seed dormancy) is that the germination rates were not scored using detailed time-course analysis [23].
Seed dormancy can be classified as primary or secondary seed dormancy [55]. Freshly harvested seeds, or dormant seeds subjected to short-term storage, are deemed to have primary dormancy, which is induced by ABA during seed maturation on the mother plant and is abolished by longer period of dry-storage treatment (after-ripening) [10], [56], [57]. By contrast, secondary dormancy can be induced in seeds with non-deep physiological dormancy after seed dispersal, and it is often associated with annual dormancy cycles in seed banks [56]. In the present study, abi4 seeds subjected to shorter period of dry-storage showed reduced seed dormancy levels and even the vivipary phenotype (Figure 1A to 1D, 1G). By contrast, the germination frequencies of abi4 and WT were comparable when the seeds subjected to longer period of storage (Figure 1E, 1F). On the other hand, further investigation revealed that OE-ABI4 seeds subjected to two weeks of storage germinated more slowly than WT seeds (Figure 3D), and the cotyledon greening rates of different genotypes were consistent with the dormancy levels (Figure S4A). Taken together, we conclude that ABI4 indeed positively regulates primary seed dormancy.
After confirming the effect of ABI4 on primary seed dormancy, we dissected the molecular mechanism underlying this phenotype. The reduced primary seed dormancy of abi4 was moderately decreased by stratification and was even abolished by longer period of after-ripening treatment (Figure 1B, 1D, 1E, 1F). Furthermore, stratification and after-ripening treatments reduce ABA content [10], [49]. Therefore, we tested the ABA levels in dry and imbibed seeds. As expected, the ABA content in dry abi4 seeds was lower than that in WT and became comparable after stratification (Figure 2). This result is similar to previously reported results about CHOTTO1, which also positively regulates primary seed dormancy [10], [44]. In these studies, the ABA level was downregulated in the cho1 mutant, which was responsible for the reduced primary seed dormancy phenotype of cho1 [10]. Therefore, we conclude that the decreased ABA level in the abi4 mutant is at least partially responsible for the reduced primary seed dormancy phenotype, and further, ABI4 positively regulates ABA biogenesis. On the other hand, GA biosynthesis is enhanced in the ABA deficient aba2 mutant, indicating that ABA is involved in the suppression of GA biosynthesis in both developing and imbibed seeds [9]. Because the ABA content in abi4 seeds was markedly downregulated (Figure 2), we tested the responses of abi4 and OE-ABI4 to PAC and GA during seed germination. A previous report showed that the increased resistance to PAC suggests that the mutant contains higher levels of active GA or stronger GA signaling than the WT [51]. We found that OE-ABI4 was sensitive to PAC and abi4 was resistant, while exogenous GA rescued the delayed germination phenotype of OE-ABI4 (Figures 3D to 3F, S4), and further, the GA measurements result showed that abi4 seeds indeed contain higher levels of active GA4 than the WT (Figure 3G). These results are consistent with the ABA measurements (Figure 2). Therefore, we propose that ABI4 attenuates GA biosynthesis and promotes ABA biosynthesis to precisely regulate seed germination.
To further confirm the changes in ABA and GA content during seed germination, we also investigated the expression levels of ABA and GA biosynthetic and inactivation genes in dry and imbibed seeds. The results showed that the expression of most genes involved in ABA and GA metabolism was altered in dry and imbibed abi4 seeds (Figure 4), which is consistent with the results of ABA and GA quantification, and the analysis of the responsiveness of OE-ABI4 and abi4 to GA and PAC treatments (Figures 2, 3, S4). These results were similar to results obtained from the analysis of sorghum grains, i.e., changes in the expression level of GA metabolism genes affects the seed dormancy and germination potential of sorghum grains [58]. In particular, the expression levels of CYP707A1 and CYP707A2 were remarkably decreased in the abi4 seeds (Figure 4E). Furthermore, ChIP-qPCR analysis and the tobacco transient expression assays revealed that ABI4 inhibits both of the two ABA inactivation genes (CYP707A1 and CYP707A2) expression by directly binds to the promoters (Figure 5D). In addition, the CCAC motifs in these promoters are important and the inhibition effect of ABI4 on its transcription was depended on the CCAC cis-element (Figure S6).
Further evidence confirming the regulation of ABA biogenesis by ABI4 was obtained by genetic analysis; the abi4 mutant rescued the delayed germination phenotype of cyp707a1 and cyp707a2 (Figure 6C, 6D). Accordingly, our further experimental evidences demonstrated that ABI4 directly repress CYP707A1 and CYP707A2 expression to promote ABA biosynthesis (Figure 5E to 5H), and the higher expression level of CYP707A1 and CYP707A2 in the absence of ABI4 result in reduced ABA content and, subsequently, the decreased seed dormancy level (Figures 2, 5). Notably, except for cyp707a1 and cyp707a2, abi4 also rescued the non-germination phenotype of ga1-t without exogenous GA treatment (Figure 6A), suggesting that ABI4 is indeed involved in regulation of GA biogenesis. Mutation at early stage of GA synthesis gene does not totally abolish GA in plant, and the ga1-3, an allele mutant of ga1-t, contains very low level of GA [59]. In abi4 and abi4/ga1-t double mutants, reduced ABA contents and activated downstream GA synthesis and down regulated GA metabolic gene transcription might increase GA/ABA ratio in seeds, thus promotes the germination of abi4/ga1-t double mutant (Figures 4, 6A). abi4 has the similar effects of spy, rgl2 and abi3 on the ga1 mutant [22], [28], [29]; these genes also are negative regulators of the GA biogenesis or signaling pathway. Taken together, we conclude that ABI4 regulates ABA biogenesis positively, and GA biosynthesis negatively, during seed germination.
Previous elegant studies demonstrated that ABI4 is a key ABA signaling component per se [31], and in this study, we further showed that ABI4 is also involved in ABA and GA biogenesis (Figures 2, 3). High GA level could induce the transcription of α-amylase gene, whose product in turn hydrolyzes the seed coat which is essential for normal germination process. In opposite, ABA inhibits seed germination through suppressing the α-amylase gene expression [3]. Furthermore, previous study revealed that ABA is involved in the suppression of GA biogenesis [9]. Therefore, the decreased ABA level in abi4 seeds could further activates the GA biogenesis, and subsequently, the increased GA content further promotes the α-amylase gene transcription. Accordingly, the seed dormancy level of abi4 is decreased.
Although the decreased ABA level and increased GA content in abi4 seeds are responsible for the reduced primary seed dormancy in this mutant (Figures 2, 3, S4), it is noteworthy that reduced seed dormancy was also detected when the short-term stored abi4 seeds were stratified (Figure 1B, 1D), even the corresponding ABA levels were comparable between abi4 and WT after stratification treatment (Figure 2). These results suggest that ABA signaling plays an important role in the control of primary seed dormancy. Indeed, previous studies have demonstrated that ABI4 positively regulates ABA signaling during seed germination [31], [60], and our results are consistent with this conclusion (Figure 1D). The other evidence about the key regulators in ABA signaling involved in seed dormancy control was from the analysis of the mutation in ABI3 locus. Similar to abi4, abi3 also was found to show the decreased seed dormancy [18]. ABI3, ABI4 and ABI5 were demonstrated to work in the same pathway in ABA signaling. Whether ABI5 is also involved in seed dormancy still need to be addressed in the future. Therefore, ABA signaling might also play a positive role during the control of seed dormancy.
Taken together, the present study demonstrates that ABI4 positively regulates primary seed dormancy by mediating the biogenesis of ABA and GA. Further, this study also strongly suggests that ABI4 plays a pivotal role in these two signaling pathways. Further functional dissection of ABI4 during the biosynthesis and signaling of ABA and GA is necessary to obtain a deeper understanding of the crosstalk between these two hormones.
Arabidopsis ecotype Columbia-0 was used as the wild type in this study. The point mutant abi4-1 (CS8104) and the T-DNA insertion mutants abi4-t (SALK_080095), cyp707a1 (SALK_069127) and cyp707a2 (SALK_083966C) were obtained from the ABRC (The Ohio State University, Columbus, OH, USA). It is noted that the T-DNA insertion mutant SALK_080095 was named as abi4-2 [47]. But the name of abi4-2 has been given much earlier to the other mutant harboring a point mutant in ABI4 gene [35].Thus the T-DNA insertion line SALK_080095 was named as abi4-t in this work. The ga1-t mutant (SALK_023192) in the Columbia-0 background was a gift from Dr. Xiangdong Fu (Institute of Genetics and Developmental Biology, Chinese Academy of Sciences, Beijing). The abi1-1, abi2-1, snrk2.2/snrk2.3 mutants seeds were supplied by Dr. Zhizhong Gong (College of Biological Sciences, China Agricultural University, Beijing). Arabidopsis seeds were surface-sterilized with 10% bleach and washed at least four times with sterile water. Sterile seeds were suspended in 0.2% agarose and sown on 1/2 MS medium plus 1% sucrose. The seeds were stratified on plates in the dark at 4°C for 0 or 3 days, depending on the experiment, and then transferred to a tissue culture room at 22°C under a 16-h-light/8-h-dark photoperiod. For ga1-t, the seeds were soaked in 100 µM GA solution for 3 days at 4°C, as the ga1-t mutant cannot germinate in the absence of exogenous GA. Normal 1/2 MS medium was supplemented with 1% sucrose and, unless otherwise noted, GA (product number G7645, Sigma-Aldrich Company ltd, USA) or PAC (product number 46046, Sigma-Aldrich Company ltd, USA) was added as needed.
Transgenic plants carrying constitutively expressing ABI4 were generated. To produce 35S-ABI4 plants, the 987-bp CDS (coding sequence) fragment was amplified by PCR and then cloned into the vector pCanG-HA-GFP, in which ABI4 was expressed under the control of the CaMV 35S promoter. Transformation of Arabidopsis was performed by the vacuum infiltration method using the Agrobacterium tumefaciens strain EHA105 [61]. T2 seeds were germinated on MS plates containing 50 mg/mL kanamycin for vector pCanG-HA-GFP, and the resistant seedlings were transferred to soil to obtain homozygous T3 seeds. For more detailed phenotypic analysis, two independent T3 homozygous lines containing a single insertion were employed.
To test germination rates, seeds were collected at the same time. Seeds subjected to various periods of dry storage were sown on normal 1/2 MS medium or 1/2 MS medium supplemented with various concentrations of GA or PAC. Radicle emergence was scored at the indicated time points, and at the same time, the percentages of cotyledon greening were also scored. For each germination test, approximately ≥45 seeds per genotype were used, and three experimental replications were performed. The average germination percentage ± SE (standard error) of triplicate experiments was calculated. For photography, a Leica MZ16 FA stereomicroscope was employed (Leica Company, Germany). Photographs were taken using the same settings at the indicated time points. The vivipary assay was performed according to a previously described protocol [51]. Developing siliques at the long-green stage were collected from the same sites of plants with various genotypes, sterilized with 70% ethanol for 1 minute and 25% bleach for 10 minutes and plated on 1/2 MS medium or damp soil.
Total RNA preparation (from dry or imbibed seeds at various times), first-strand cDNA synthesis and qRT-PCR were performed as previously described [62]. DNase I-treated total RNA (2 µg) was denatured and subjected to reverse transcription using Moloneymurine leukemia virus reverse transcriptase (200 units per reaction; Promega Corporation). Quantitative PCR was performed using the SsoFast EvaGreen Supermix (Bio-Rad) and a CFX96 Touch Real-Time PCR Detection System (Bio-Rad). Gene expression was quantified at the logarithmic phase using the expression of the housekeeping 18S RNA as an internal control. Three biological replicates were performed for each experiment. Primer sequences for qRT-PCR are shown in Table S1.
To test the ABI4 protein levels in transgenic plants (35S-ABI4-GFP), western blotting was performed according to previously described protocols [62], [63]. Approximately two-week-old seedlings grown on 1/2 MS medium were ground in liquid nitrogen and extracted with 4 M urea buffer. Crude extracts were separated by SDS-PAGE and transferred onto nitrocellulose membranes. The membranes were stained with 0.2% Ponceau S, with Rubisco serving as an internal control. The anti-GFP antibody was purchased from Santa Cruz Biotechnology, Inc.
ChIP was performed as previously described [64], with minor modifications. Transgenic seeds containing 35S-ABI4-GFP were grown on 1/2 MS medium for approximately 2 weeks. The seedlings were then harvested (1.5 g) and crosslinked with 1% formaldehyde for 30 minutes under a vacuum; the crosslinking was stopped with 0.125 M glycine. The seedlings were ground in liquid nitrogen, and the nuclei were isolated. Immunoprecipitations were performed with the anti-GFP antibody and protein G beads. Immunoprecipitation in the absence of anti-GFP served as the control (CK). DNA was precipitated by isopropanol, washed with 70% ethanol and dissolved in 10 µl water containing 20 µg/mL RNase. The qRT-PCR analysis was performed using specific primers corresponding to different promoter regions of CYP707A1, CYP707A2 and CYP707A3. TUB4 was used as an internal control. Since ABI4 directly binds to the promoter of ABI5 [50], this promoter was employed as a positive control. Primers used in the ChIP-qPCR assay are shown in Table S1.
This transient expression assay was performed in N. benthamiana leaves as previously described [65]. The 2329 bp for native CYP707A1 promoter (Pro-CYP707A1) and 2015 bp for native CYP707A2 (Pro-CYP707A2) were amplified separately from genomic DNA. In addition, the several mutated CYP707A1 promoter fragments (including Pro-CYP707A1 (m1), Pro-CYP707A1 (m2), Pro-CYP707A1 (m1+m2)) were generated by PCR amplification. All these five promoter fragments were cloned into pENTR using the pENTR Directional TOPO cloning kit (Invitrogen). Then, these promoter versions were fused with the luciferase reporter gene LUC through the Gateway reactions into the plant binary vector pGWB35 [66]to generate the several reporters constructs. The effector construct was the pCanG-ABI4-GFP.
For analysis of ABA content in dry or imbibed seeds, the seeds were ground in liquid nitrogen, and 150 mg of seed powder was homogenized and extracted for 24 h in methanol containing D6-ABA (purchased from OIChemIm Co. Ltd.) as an internal standard. Purification was performed with an Oasis Max solid phase extract cartridge (150 mg/6 cc; Waters) and eluted with 5% formic acid in methanol. The elution was dried and reconstituted, and it was then injected into a liquid chromatography–tandem mass spectrometry system consisting of an Acquity ultra performance liquid chromatograph (Acquity UPLC; Waters) and a triple quadruple tandem mass spectrometer (Quattro Premier XE; Waters). Three biological replications were performed.
The endogenous gibberellins were determined by the method described [67]. Arabidopsis seeds (200 mg) were frozen in liquid nitrogen, ground to fine powder, and extracted with 80% (v/v) methanol. GA isotope standards were added to plant samples before grinding. The crude extracts were purified by reversed-phase solid-phase extraction, ethyl ether extraction and derivatization. The resulting mixture was injected into capillary electrophoresis-mass spectrometry (CE-MS) for quantitative analysis.
|
10.1371/journal.pntd.0005440 | The effect of current Schistosoma mansoni infection on the immunogenicity of a candidate TB vaccine, MVA85A, in BCG-vaccinated adolescents: An open-label trial | Helminth infection may affect vaccine immunogenicity and efficacy. Adolescents, a target population for tuberculosis booster vaccines, often have a high helminth burden. We investigated effects of Schistosoma mansoni (Sm) on the immunogenicity and safety of MVA85A, a model candidate tuberculosis vaccine, in BCG-vaccinated Ugandan adolescents.
In this phase II open label trial we enrolled 36 healthy, previously BCG-vaccinated adolescents, 18 with no helminth infection detected, 18 with Sm only. The primary outcome was immunogenicity measured by Ag85A-specific interferon gamma ELISpot assay. Tuberculosis and schistosome-specific responses were also assessed by whole-blood stimulation and multiplex cytokine assay, and by antibody ELISAs.
Ag85A-specific cellular responses increased significantly following immunisation but with no differences between the two groups. Sm infection was associated with higher pre-immunisation Ag85A-specific IgG4 but with no change in antibody levels following immunisation. There were no serious adverse events. Most reactogenicity events were of mild or moderate severity and resolved quickly.
The significant Ag85A-specific T cell responses and lack of difference between Sm-infected and uninfected participants is encouraging for tuberculosis vaccine development. The implications of pre-existing Ag85A-specific IgG4 antibodies for protective immunity against tuberculosis among those infected with Sm are not known. MVA85A was safe in this population.
ClinicalTrials.gov NCT02178748
| There is an urgent global need for an improved TB vaccine strategy. Adolescents are an important target population for a TB vaccine. In field settings where the need for a vaccine is greatest, co-infection with other pathogens including malaria, HIV and helminths, may interfere with the impact of such strategies. In this study, we used a model TB vaccine candidate based on a genetically engineered viral vector, and with an excellent safety profile, to determine whether there was any immunological interference when this vaccine was used in adolescents who were co-infected with S. mansoni. Our data shows comparable immunogenicity in adolescents infected and uninfected with S. mansoni, suggesting that co-infection with this helminth species may not have an adverse impact on candidate TB vaccines of this type. The vaccine was safe and induced robust cellular responses in both the infected and uninfected adolescents following immunisation. These findings are important and encouraging for tuberculosis vaccine development.
| Helminth infection is widespread with over a billion people infected worldwide [1]. In 2014, at least 258 million people required treatment for schistosomiasis, 90% of whom lived in Africa [2]. In animal models, schistosomiasis impairs responses to immunisation against hepatitis B [3], malaria [4], mycobacterial infection and BCG immunisation [5, 6]. In humans, the effects are less clear. Helminth infections can modulate vaccine-induced responses against tetanus [7], influenza [8], cholera [9], and Salmonella typhi [10]. T-helper (Th)1 responses are important for protective immunity against intracellular bacteria and viruses, while helminths characteristically induce Th2 [11] and immunoregulatory responses, both of which can down-regulate Th1 pathways [8]. If concurrent helminth infection impairs vaccine-induced immunity, policies for helminth treatment before, or at, the time of immunisation may be needed.
Tuberculosis (TB) remains a global health problem with very high mortality and morbidity. In 2014, there were an estimated 9.6 million new cases and 1.5 million deaths from the disease [12]. The only licensed TB vaccine, Bacille Calmette-Guérin (BCG), prevents severe disease in childhood [13–15] but does not protect against pulmonary TB in many parts of the world. Development of an effective vaccine is a key strategy for combating the epidemic.
Adolescents are an important target population for a TB vaccine [16, 17] because TB presents as the pulmonary (transmissible) form, in adolescents and young adults. However, adolescents harbour a high prevalence and intensity of helminth infection in countries where helminths are endemic.
MVA85A, a recombinant Modified Vaccinia Ankara virus expressing the Mycobacterium tuberculosis (Mtb) antigen 85A, is a clinically advanced TB vaccine candidate. It has been administered to over 2400 human subjects and has an excellent safety profile. It is highly immunogenic in BCG-primed UK adults [18], but only modestly immunogenic in BCG-vaccinated South African infants [19], and did not confer protection in either BCG-vaccinated infants or HIV-infected adults [20]. Reasons for the reduced immunogenicity, and methods of enhancing immunogenicity, are currently being evaluated [21]. Co-infection with helminths may contribute to poor immunogenicity in high TB burden countries.
We here use MVA85A, a candidate TB vaccine with an excellent safety profile, as a model TB vaccine candidate with which to evaluate the effect of S. mansoni (Sm) infection on the T cell immunogenicity induced. Findings from this study may be important for all TB vaccine development.
The study was approved by the Research Ethics Committee of the Uganda Virus Research Institute, the Uganda National Council for Science and Technology, the Uganda National Drug Authority, Oxford Tropical Research Ethics Committee, and London School of Hygiene & Tropical Medicine Ethics Committee. Participants provided written informed assent and their parents or guardians gave written informed consent.
This was a phase II open-label, non-randomised trial with two trial groups: Group 1 without any current helminth infection detected and Group 2 with current Sm infection only. Laboratory staff performing immunological assays were blinded to helminth infection status.
The primary objective was to compare the T cell immunogenicity of MVA85A in adolescents with and without Sm infection. Secondary objectives were to explore effects of Sm infection on other aspects of the immune response following MVA85A immunisation and to assess the safety of MVA85A in BCG-vaccinated Ugandan adolescents.
The study took place in five primary schools within schistosomiasis-endemic areas in Wakiso District, Uganda, on the shores of Lake Victoria, between June 2014 and January 2015. Community leaders, district Ministry of Education and Health officials and school staff were consulted and parent-teacher meetings held before the trial started to describe the trial, explain procedures and answer questions.
Pre-screening, by Kato Katz microscopy, was undertaken by the Vector Control Division (VCD) of the Ministry of Health who hold the mandate to screen for helminths in Ugandan schools. Children who had no helminths (candidates for Group 1) or Sm only (candidates for Group 2), who were within the eligible age group and who had a BCG scar, were invited for screening visits.
Assenting adolescents, with consent from a parent or guardian, were screened on the school premises. A maximum of 60 days was allowed between screening and vaccination. Participants provided blood for haematology and biochemistry, urine and three stool samples at screening, and blood samples for evaluation of immune responses at screening, enrolment and 7, 28 and 56 days post-vaccination. Stool samples for Kato Katz for all participants were performed at D28 and D56. Group 2 participants had an extra visit at D42 for anthelminthic treatment.
Participants were eligible for vaccination if aged 12 to 17 years, resident in the study area, BCG-vaccinated (based on BCG scar or written documentation), healthy by history and physical examination. Exclusion criteria were clinical, radiological, or laboratory evidence of, or previous treatment for, active or latent TB (including a positive ELISpot for ESAT6 or CFP10 at screening); sharing a residence with an individual on anti-TB treatment or with culture or smear-positive pulmonary TB within the last year; positive serology for HIV (Murex Diasorin, Italy and Vironostika, Biomerieux, France), hepatitis B or C (Innotest HCV Ab IV, Innogenetics, Belgium); positive rapid diagnostic test for malaria (SD-Bioline Inc, Korea); Mansonella perstans infection (modified Knott’s method [22]); intestinal parasites other than Sm; pregnancy; history of anaphylaxis or allergy likely to be exacerbated by vaccine; haematological or biochemical findings deemed clinically significant at screening; Sm infection intensity>2000 eggs per gram of stool (these individuals were treated immediately).
Three baseline stool samples were examined by the Kato Katz method [23, 24]. Urine was examined for schistosome circulating cathodic antigen (CCA) by rapid test (ICT International, Cape Town, South Africa). Participants were recruited to Group 1 (no helminths) if none of these assays were positive. Participants were recruited to Group 2 (Sm infected) if at least one stool sample was microscopy positive on Kato Katz examination for Sm, and none of the investigations were positive for any other helminths.
One stool sample per participant was also examined by PCR [25–28] for Sm, Strongyloides stercoralis and Necator americanus (helminths prevalent in the area [29, 30]). Participants were excluded from Group 1 if Sm positive and from both groups if S. strongyloides positive. The protocol initially required exclusion from both groups if PCR positive for N. amercianus, however, this was amended when a high proportion of Sm positive individuals were found to be N. americanus PCR positive (despite having three samples negative by microscopy).
MVA85A was produced under Good Manufacturing Practice conditions by IDT Biologika GmbH, Dessau-Rosslau, Germany. It was stored at -80°C in a locked, temperature‐monitored freezer at the Uganda Virus Research Institute-International AIDS Vaccine Initiative HIV Vaccine Program pharmacy, Entebbe. On vaccination days, vaccine was transported from pharmacy to field on dry ice.
All participants received 1x108 plaque forming units (PFU) MVA85A as a single intramuscular injection in the deltoid region.
All participants were treated for helminths with praziquantel (40mg/kg) and albendazole (400 mg) under observation by trial staff. Group 1 received a single dose on D56 (in accord with annual mass treatment performed by VCD). Group 2 participants received two doses of praziquantel, one on D28 (after samples were obtained) and one on D42, to optimise clearance of infection. In this community, free diagnosis and treatment for Schistosomiasis is largely provided by the VCD, with whom we worked, so any treatment was done with the study team’s knowledge.
The primary endpoint was T cell immunogenicity assessed by ex vivo Ag85A-specific interferon gamma (IFN-γ) ELISpot response. Secondary endpoints were the profile of cytokine responses assessed by multiplex Luminex assay of supernatant following six-day whole blood stimulation, plasma antibody concentrations, and solicited and unsolicited local and systemic adverse events.
Whole blood assays were conducted as previously described [18, 31, 33]. Briefly, whole blood diluted to a final concentration of 1:4 in RPMI supplemented with 10% Fetal Calf Serum (R10) was stimulated with Ag85A (2μg/ml), PPD-T (10μg/ml), schistosome worm antigen (SWA;10μg/ml, provided by Professor M Doenhoff, University of Nottingham, UK) and phytohaemaggluttin (Sigma-Aldrich; 10μg/ml) or left unstimulated and incubated for 6 days in 5% carbon dioxide at 37°C. Cytokine responses (IFN-γ, interleukin (IL)10, tumour necrosis factor (TNF)α, granulocyte-monocyte stimulating factor (GMCSF), IL12P40, IL13, IL17A, IL1A, IL2, IL5, inducible protein (IP)10, macrophage inflammatory protein-1-alpha (MIP1ɑ), IL6, IL1270, IL4, monocyte chemoattractant protein-3 (MCP3) and macrophage-derived chemokine (MDC)) in culture supernatants were assessed using a MILLIPLEX MAP Kit, (EMD Millipore Corporation, Billerica, MA, USA). Detection antibody cocktail, streptavidin-phycoerythrin working concentrations, and the bead cocktail were diluted to allow three plates to be run from each kit.
ELISA assays were used to measure levels of total IgG, IgG1, IgG4 and IgE to Ag85A, SWA and SEA using an established protocol modified to include responses to mycobacterial antigens [34], detailed in the supplementary methods (S1 Methods).
A portion of a stool sample from each participant was suspended in ethanol and stored at -80°C. Subsequently, DNA was extracted and samples were examined for Sm, S. stercoralis and N. americanus DNA, as previously described [25].
Aliquots of EDTA-anticoagulated whole blood were stored at -80°C. To investigate for low-intensity Plasmodium falciparum infection (the common malaria species in this setting [35] which could have been missed by the rapid diagnostic test) DNA was extracted from 500μl whole blood then assessed using Real Time PCR as previously described [36] and detailed in the supplementary methods (S1 Methods). A positive malaria PCR was not an exclusion criterion for eligibility.
Solicited and unsolicited adverse events were collected using a diary card completed at home by the participant from D0 to D6 and were assessed by the clinician at every clinic visit.
The initial plan was to enrol 30 subjects in group 1 and 40–60 subjects in group 2. The sample size was amended due to challenges with enrolment, to 12–24 subjects in each group. With 12–18 participants in each group, the study had 56–74% power to detect a 40% reduction in geometric mean SFUs, and 82–94% power to detect a 50% reduction. The sample size calculation was based on previous ELISpot data from an adolescent trial [37] and assumed a mean response on the log-scale for SFUs of 6.40 with a standard deviation of 0.59 corresponding to a geometric mean of 600 SFUs on the back-transformed original scale. The initial plan was to match participants in the two groups by age and gender, however, this was not possible and these variables were therefore considered a priori as confounders.
The primary immunogenicity analysis was comparison of Ag85A ELISpot responses between Sm infected (Group 2) and uninfected (Group 1) adolescents at peak response (D7). Non-parametric analyses using the Mann Whitney U-test and multiple linear regression with bootstrapped confidence intervals adjusting for age, gender, school and hookworm PCR results were used. Area under the curve (AUC) analysis was done from D0 to D28.
Multiplex Luminex data were analysed using R (version 3.2.2). Unstimulated cytokine responses were subtracted from antigen-stimulated results and negative values were assigned 0. Baseline responses were compared between the groups and change following vaccination was evaluated. Cytokine responses were compared between Sm infected and Sm uninfected individuals by fitting regression models which also contained terms for appropriate confounders [38], age, gender and school. Mean (geometric mean) fold differences were calculated. Principal component analysis was conducted on the standardised log AUC cytokine data. Bonferroni corrections were used to account for multiple testing.
For antibody concentrations, comparisons were made between baseline levels and change following vaccination.
Solicited and unsolicited adverse events overall and for each group were summarised by frequency and severity.
1068 pupils were pre-screened, and 174, who were within the target age group of 12–17 years, were approached for screening. Sixty-seven were excluded before consenting, 35 because parents/guardians declined to consent; 12 had no BCG scar; 10 left school during the initial processes; five had no adult or responsible legal guardian accessible for consenting; three had recently been treated for schistosomiasis; one volunteered known HIV-positive status and one was found to be under age at screening (11 years) despite having reported being 12 years at pre-screening. 107 subjects consented, assented and were screened for eligibility of whom 36 were enrolled (Fig 1).
Eighteen subjects had Sm infection only and were enrolled into Group 2, 18 had no helminth infection detected and were enrolled into group 1 (3 of these subsequently had Sm detected on PCR and were excluded from all further analysis). Participants in Group 1 and 2 differed in age, school and N. americanus PCR status (Table 1).
The peak D7 Ag85A IFNγ ELISpot response was compared between groups. Two participants from group 1 and four from group 2 with missing data were excluded from the analysis. Median Ag85A SFC/ 1x106 PBMC were not significantly different (Mann-Whitney p = 0.65). Regression and AUC analysis of the Ag85A ELISpot response (D0 to D28) adjusting for age, gender, school and stool PCR for N. americanus showed no evidence of an association between Sm infection and MVA85A induced IFNγ response (see Table 2 and Fig 2).
D7 ELISpot and AUC analysis (D0—D28) for purified protein derivative (PPD) responses showed no significant difference between the two groups (S1 Table). ESAT and CFP 10 responses remained negative at all time points (S1 Fig).
Ag85A antibody levels were not significantly different in the two groups with the exception of IgG4, which was significantly higher in the Sm infected group at baseline and at all time points post-vaccination (Fig 3); this difference remained significant after correcting for multiple testing using Bonferroni correction (p = 0.012). Ag85A-specific antibodies were present at baseline and showed no significant change post-vaccination. Our assay did not detect much Ag85A specific IgE in the study participants.
SWA-specific IgG antibodies and sub-classes were significantly higher (p values for IgG = 0.0060, IgG1 = 0.0136 and IgG4 = 0.0004) in the Sm infected group (Fig 4). SWA-specific IgE levels were low in both groups, and no statistically significant differences were seen between the groups (Fig 4). SWA-specific antibody levels tended to increase at day 56 in the Sm infected group (after praziquantel treatment). Exploring the observed elevated levels of Ag85A-specific IgG4, we found no correlation between SWA or SEA-specific IgG4 and Ag85A-specific IgG4 among Sm-infected participants.
Analysis of the multiplex Luminex assays following six-day stimulation and culture for Ag85A and PPD stimulation showed no significant difference in pre-immunisation responses for any cytokine, and all cytokines showed an enhanced response post-vaccination in both groups, with the exception of IL2, for which measured levels were very low. There was no significant difference post-vaccination in levels of any individual cytokine. Regression analysis (supplementary S3 and S4 Tables) of AUC accounting for age, gender and school as confounders suggested a higher response to Ag85A and PPD among Sm infected compared to uninfected participants for all cytokines except IL12P40 (to Ag85A) and IL2 (to both Ag85A and PPD). There was no significant difference in geometric means following stimulation with Ag85A and PPD for any of the cytokines (except for IL1A to PPD, p value = 0.017) between uninfected and infected when adjusted for multiple testing using Bonferroni correction (S3 and S4 Tables).
Safety results were considered for all 36 subjects who received the vaccine despite the fact that three subjects initially allocated to Group 1 were found to have Sm on PCR and therefore excluded from the immunogenicity analysis (Table 3 and S2 Table).
No serious adverse events were reported in this trial. Using data collected from the diary cards and at clinic visits on days 0, 7, 28 and 56, all 36 children reported at least one adverse event. 34 (94%) of 36 children vaccinated experienced at least one local adverse event and 6 (17%) experienced at least one systemic event. There was no difference in adverse events between the two groups. All adverse events had resolved by day 56.
Local warmth (94%) and pain (92%) at the injection site were the most commonly reported of the solicited local adverse events by subjects in both groups, while headache (69%) was the most commonly reported of the solicited systemic adverse events in the diary card. The AE profile was not different in those excluded from the immunogenicity analysis.
259 graded adverse events were recorded (See S2 Table). 192 (72%) of the adverse events were mild. Two severe adverse events occurred and were pain at vaccination site and itching at vaccination site, both observed in participants in the no helminth group.
MVA85A induced a robust cellular response, with no difference in cellular immunogenicity between adolescents with and without current Sm infection. MVA85A did not boost the baseline Ag85-specific antibody response (as expected [18]). Levels of Ag85A specific IgG4 antibody were elevated among the Sm infected participants at baseline and at all subsequent time points. The safety profile was consistent with other published data on this vaccine from African populations [37, 39].
The peak (D7) Ag85A ELISpot response of >500 SFC/million PBMC, was comparable to, or higher than, responses previously observed among adolescents and adults in the Cape region of South Africa [37, 39], where lifetime exposure to schistosomiasis and other tropical parasitic infections is likely to be lower than in tropical Africa [40]. Current exposure to Sm among Ugandan adolescents did not impair the cellular immunogenicity of MVA85A. This is important given the poor efficacy of BCG in tropical latitudes [41]. One potential advantage for booster TB vaccines may be prior exposure to environmental mycobacteria: besides BCG, the Ugandan participants would have been extensively exposed to environmental mycobacteria [42], as evidenced by anti-mycobacterial antibody levels prior to MVA85A vaccination [18]. A point for caution is that selection for prior BCG based largely on BCG scar (as most had no records available) may have biased recruitment to individuals who received a more immunogenic formulation of BCG, or who were predisposed to stronger anti-mycobacterial responses, since scarring varies between BCG strains and only about 60% of infants scar in response to strains commonly used in Uganda [43].
Our initial focus was on Th1 cytokines as measured by ex-vivo ELISpot and Luminex, as there is clear evidence for a role for Th1 cytokines in protective immunity against Mycobacterium tuberculosis. Given the increased interest in a role of humoral immunity and B cells in protective immunity against mycobacteria, we also investigated the humoral response as measured by IgG to the Ag85A insert. Further measures of immunogenicity such as mycobacterial growth inhibition assays and other T cell functions may be considered as part of future work on stored samples from this trial. There is as yet no clearly defined immunological correlate of protection with which to measure candidate vaccine immunogenicity. However, the recent paper by Fletcher et al [44] indicated that pre-MVA85A-immunisation BCG-specific ELIspot responses and post MVA85A antibody (specific for antigen 85) were protective. We did not perform the former but did measure the latter and observed no effect.
Recruitment to this study was challenging and the final sample size had power only to detect a large effect of current Sm infection on vaccine immunogenicity. However, there is no suggestion in the data that Sm was associated with a reduced or biased response to immunisation. If anything, cellular responses to MVA85A were slightly stronger, and increased further following treatment with praziquantel, in the Sm infected group. This was surprising given prior results in animal models of schistosomiasis and BCG immunisation and infection [5, 6]. Our findings in the human population may relate, again, to the impact of prior mycobacterial exposure, to intensity of Sm infection (low in most of our subjects) and to the characteristics of MVA85A (a viral vector vaccine which enters host cells but does not replicate [18]) as opposed to BCG (an attenuated mycobacterial vaccine which replicates in the host).
Lack of bias in the profile of cytokine response following MVA85A was also surprising, given evidence from earlier studies of Th2 bias following tetanus immunisation among individuals with Sm infection [7], and more particularly given evidence of bias in the antibody response to Ag85A (to IgG4) prior to MVA85A administration.
The groups in this study were defined by detection of current Sm infection. It is possible that prior helminth exposure might influence vaccine responses (rendering current status less relevant). Recent helminthic disease acquisition in the Sm negative group during follow up, detectable by Kato Katz, was ruled out using the follow up samples collected. It is plausible that previous exposure to helminths in the Sm negative group (which we could not measure in this study) may be responsible for the lack of difference, but the differences in Sm-specific antibody levels make this unlikely. For the Sm positive group, we expect that infections were chronic–as we have previously reported among adults in related fishing communities [45]. The finding that schistosome-specific IgG levels were significantly higher in the Sm infected group suggests that these individuals differed in prior, as well as current, Sm exposure. Sm-specific IgE increases gradually among exposed populations. Low levels in these adolescents (Sm infected or otherwise) is in keeping with this pattern [46–49].
This was an observational study, susceptible to unmeasured confounding. We attempted to exclude confounding by parasitic co-infections by enrolling participants in whom no other parasites that are common in this setting could be detected. This proved challenging for sub-microscopic hookworm and malaria infection, but adjusting for hookworm PCR status and sub-microscopic malaria (measured by malaria PCR) made no difference to the results.
It may not be possible to extrapolate these findings to other vaccines and other species of helminth. There is substantial evidence of variability between populations in vaccine immunogenicity: notably, impaired responses to yellow fever vaccine were observed in Uganda, compared to Switzerland, in a study that found reduced vaccine virus replication and neutralising antibody production and persistence, in the context of elevated innate and adaptive immune response activation, among Ugandan volunteers [50]. There is evidence, for other vaccines, of differences (such as Th2 bias) in the response between people with and without schistosomiasis [7, 10], and of modulation of vaccine responses by other helminth species [8, 9]. Key factors in helminth-vaccine interactions are likely to include vaccine characteristics (live, replicating, protein or toxoid), prior exposure to environmental organisms or pathogens homologous to the vaccine, and the nature of the desired response (cellular or antibody).
We conclude that current Sm infection did not interfere with the cellular immunogenicity of MVA85A in our study population. These findings are important and encouraging for the development of TB vaccines in general, and support the further development of booster TB vaccines for populations in tropical countries. MVA85A was safe in this population of African adolescents.
|
10.1371/journal.pgen.1003954 | ADP1 Affects Plant Architecture by Regulating Local Auxin Biosynthesis | Plant architecture is one of the key factors that affect plant survival and productivity. Plant body structure is established through the iterative initiation and outgrowth of lateral organs, which are derived from the shoot apical meristem and root apical meristem, after embryogenesis. Here we report that ADP1, a putative MATE (multidrug and toxic compound extrusion) transporter, plays an essential role in regulating lateral organ outgrowth, and thus in maintaining normal architecture of Arabidopsis. Elevated expression levels of ADP1 resulted in accelerated plant growth rate, and increased the numbers of axillary branches and flowers. Our molecular and genetic evidence demonstrated that the phenotypes of plants over-expressing ADP1 were caused by reduction of local auxin levels in the meristematic regions. We further discovered that this reduction was probably due to decreased levels of auxin biosynthesis in the local meristematic regions based on the measured reduction in IAA levels and the gene expression data. Simultaneous inactivation of ADP1 and its three closest homologs led to growth retardation, relative reduction of lateral organ number and slightly elevated auxin level. Our results indicated that ADP1-mediated regulation of the local auxin level in meristematic regions is an essential determinant for plant architecture maintenance by restraining the outgrowth of lateral organs.
| Plant architecture is one of the key factors that affect plant survival and productivity. It is well established that the plant hormone auxin plays an essential role in organ initiation and pattern formation, thus affecting plant architecture. We found that a putative MATE (multidrug and toxic compound extrusion) transporter, ADP1, which was expressed in the meristematic regions, through regulating the level of auxin biosynthesis, controls lateral organ outgrowth so as to maintain normal architecture in Arabidopsis. The more ADP1 was expressed, the less levels of local auxin were detected in the meristematic regions of the plant, resulting in increased growth rate and a greater number of axillary branches and flowers. The reduction of auxin levels is probably due to decreased level of auxin biosynthesis in the local meristematic regions. Down-regulated expression of ADP1 and its three closely related genes caused plants to grow slower and to produce less lateral organs. Our results indicated that ADP1-mediated regulation of the local auxin levels in meristematic regions is an essential determinant for plant architecture by restraining the outgrowth of lateral organs.
| Higher plants have a diverse range of body structures. Phyllotaxis of lateral organs, branching pattern, as well as size, shape and position of lateral organs all contribute to the overall architecture of a plant. Plant architecture is the most obvious morphology of mature plants and has long served as an important criterion for systematic and taxonomic classification of plant species [1], [2]. Plant architecture is largely determined by genetic programs and, to some extent, by environmental cues, such as light, humidity, temperature, nutrition, and plant density. Detailed studies have been focused on genetic factors that are crucial for maintenance of shoot apical meristem (SAM), initiation and outgrowth of axillary meristem (AM), proper growth rate for lateral organ development, and correct timing for reproduction and senescence [2]. Research on plant architecture has important agronomic implications because it has a direct impact on the suitability and productivity of a plant. One of the most successful modifications of plant architecture is the Green Revolution, which is based on the selection of wheat cultivars with shorter and sturdier stems, resulting in plants with enhanced yield via improved resistance to wind and rain [3]. Over the past several years, branching patterns have been intensively investigated in rice, since the formation of tillers and panicle branches will greatly affect the efficiency of light absorption, which will in turn influence the adaptation of plants to the environment [4]–[6]. Understanding the genetic and molecular mechanisms of the regulation of plant architecture would help us to modify agronomically useful traits and thus facilitate the breeding of high-yield crops.
The success of the Green Revolution mainly results from selection of plants with altered biosynthesis and/or signaling of plant hormones, among which auxin is a determinant for plant architecture. Auxin is a critical factor controlling a wide variety of developmental processes, including embryogenesis, maintenance of apical dominance, and formation of lateral organs [7], [8]. Active auxin, mainly indole-3-acetic acid (IAA), is reported to be synthesized de novo by tryptophan (Trp)-dependent and/or independent pathways in the shoot apex, young leaves, and root apex [7], [9]–[14]. After synthesis, auxin is transported by the polar transport machinery [7], so that an appropriate distribution of auxin is established to maintain normal plant architecture. Disruption of auxin gradient, either by changing auxin biosynthesis, transport, or signaling, will lead to alteration of organ growth patterns and changes of plant architecture. For example, over-expression of the auxin biosynthesis genes YUCCA1 (YUC1) and YUCCA6 (YUC6) led to auxin over-production, resulting in increased apical dominance [15], [16], whereas the quadruple knock-out mutant yuc1,2,4,6 showed abnormal flower development and loss of apical dominance, i.e., increased branching [17], [18]. The double mutant pgp1-1 pgp19-1 (mdr1-1), which exhibited a 70%–80% reduction in polar auxin transport, displayed pleiotropic phenotypes such as curly leaves, dwarfism and decreased fertility [19], [20]. Moreover, the auxin response mutant axr1-12 and the tir1 afb1 afb2 afb3 quadruple receptor mutant produced highly branched inflorescences at maturity. The quadruple mutant occasionally had no roots or produced only a single cotyledon, leading to lethality at early stages [21]–[23]. Thus, maintenance of proper auxin response and/or homeostasis is critical for normal plant architecture.
In this paper, we identified a dominant Arabidopsis mutant with an abnormal architecture, which we named adp1-D (altered development program 1- Dominant). The architecture of adp1-D was greatly altered at maturity, with increased number of axillary branches, flowers, and lateral roots. The growth rate of the mutant was accelerated throughout its life cycle. We discovered that the mutant phenotypes were caused by over-expression of ADP1 gene. ADP1 encodes a protein with sequence similarity to the multidrug and toxic compound extrusion (MATE) transporter family, which is found in prokaryotes and eukaryotes. MATE transporters are reported to be involved in a variety of important biological processes, since they function in the exclusion of toxic organic cation and disease resistance and exhibit multi-substrate specificity [24], [25].
Here we provide molecular and genetic evidence to demonstrate that the phenotypes of adp1-D were caused by reduction of the local auxin levels in the meristematic regions. The reduction was probably due to decreased levels of auxin biosynthesis in the local meristematic regions. When expression levels of ADP1 and its three closest homologs were down-regulated in Arabidopsis, the resulting quadruple mutant exhibited growth retardation and a slight reduction of lateral organ number. Our results indicated that ADP1 and its homologous genes play important roles in maintaining normal plant architecture, possibly by regulating local auxin biosynthesis.
We screened an Arabidopsis activation tagging mutant collection for mutants with altered plant architectures. The activation tagging mutant collection was generated using the activation tagging vector pSKI015 as described previously [26]. A dominant mutant with abnormal plant architecture, later designated adp1-D, was identified. The mutant displayed pleiotropic phenotypes, including accelerated growth rate of rosette leaves (Figure 1A, 1B and 1C), early flowering (Figure 1B and 1D), increased number of lateral roots (Figure S1A and Figure S1B). At maturity, the mutant had significantly more axillary branches (Figure 1D), including first-order rosette branches (RI, Figure 1E and 1F), higher-order rosette and cauline branches (RII and CII, Figure S1C). The lengths of first-order branches (RI and CI) were almost the same at different node positions (Figure S1D), suggesting the loss of apical dominance in the mutant. Occasionally, the axillary inflorescences were found in the axil of the cotyledons in the mutant (Figure 1H), which is unique because wild-type plants do not produce axillary inflorescences on the axil of the cotyledons (Figure 1G).
adp1-D mutant was almost sterile, producing very few seeds. Sterile phenotypes have been previously reported to be associated with induction of axillary branch outgrowth [27], [28]. In order to clarify whether the bushy phenotype of this mutant was a secondary effect of the sterile phenotype or not, we analyzed the kinetics of the initiation rate of the first-order rosette branches (RI). Our result showed that the difference between the mutant and wild type appeared as early as 3 days after reproductive transition, and the difference became larger with time (Figure 1I), until sterility appeared. However, exogenous application of GR24 [29], [30], a strigolactone-like inhibitor of shoot branching, revealed that the first-order rosette branches (RI) could be completely inhibited by GR24 application (Figure S2B to S2E), whereas higher-order cauline branches (CII) were not affected (Figure S2B to S2F), suggesting that the higher order branches were probably the secondary effect of sterility. Therefore in this study, we only focused on the first-order rosette branches.
To investigate the origin of the abnormal branches, we compared the early stage of axillary bud outgrowth in the first pair of leaves using scanning electron microscopy (SEM). In the wild type, although the axillary buds initiated from the epidermal cells in the semicircular zone, and then bulged outwards to form the meristems, in most cases, the axillary buds ceased development at this stage (Figure 1J to 1L). However, most of the axillary meristems of the mutant appeared to be larger than those of the wild type, and they continued to develop into inflorescences (Figure 1M to 1O). In many cases, the mutant had increased number of axillary meristems (Figure 1O). Since the developmental program of the mutant had been changed from the beginning to the end of the entire life cycle, and since the plant architecture had been greatly altered, we named the mutant as adp1-D (altered development program 1- Dominant).
To investigate the cause for the pleiotropic phenotypes in adp1-D, we performed thermal asymmetric interlaced PCR (TAIL-PCR), and identified a single T-DNA insertion in the intergenic region between At4g29130 and At4g29140 (Figure 2A). To examine whether the T-DNA insertion co-segregated with adp1-D phenotypes, we genotyped the T3 plants produced by heterozygous T2 mutants. Among 420 T3 plants, 102 were wild type without the T-DNA insertion, 106 were homozygous, and 212 were heterozygous with the T-DNA insertion (Figure 2B). All of the plants that were homozygous and heterozygous with the T-DNA insertion showed accelerated growth and increased number of lateral organs, whereas all of the plants without the T-DNA insertion appeared normal, suggesting that the pleiotropic phenotypes in adp1-D were caused by this single T-DNA insertion. To determine which gene was altered in its expression level, we examined the expression levels of all the genes within 10 kb upstream and downstream of the insertion site by quantitative RT-PCR. Only one gene, At4g29140, was over-expressed (by about 25-fold), while the expression of all the other genes remained mostly unchanged (Figure 2C), suggesting that over-expression of At4g29140 could be responsible for the adp1-D phenotypes. To confirm this, we over-expressed At4g29140 in wild-type Arabidopsis under its own promoter with four copies of the 35S enhancer (Figure 2D). The transgenic plants recapitulated all the phenotypes in adp1-D. About 10% of the transgenic plants showed more severe phenotypes of smaller plant size, highly compact leaves, and many more branches (Figure 2E and 2F). The expression level of At4g29140 correlated with the severity of the phenotypes (Figure 2E to 2G). These data indicate that the pleiotropic phenotypes of adp1-D were indeed caused by over-expression of At4g29140.
ADP1 has no intron and encodes a protein of 532 amino acid residues (Supplemental Figure 3A), sharing sequence similarity with the Arabidopsis MATE proteins. These proteins are characterized by 11 to 13 transmembrane helixes and two typical MATE domains [31]. The Arabidopsis genome contains 58 putative MATE transporters, which are grouped into five groups [32]. ADP1 belongs to a clade, which has eight members sharing high sequence identity in the conserved MATE domain (Figure S3B and S3C).
To characterize the expression pattern of ADP1, a 2 kb promoter fragment upstream of ADP1 start codon was fused with β-glucuronidase (GUS) reporter gene and transformed into wild-type Arabidopsis. GUS staining analysis of the homogenous transgenic lines showed that the promoter activity of ADP1 was mainly detected in tissues where cells were actively dividing, such as leaf primordia and young leaves (Figure 3A), the junction between lateral root and the primary root (Figure 3B), root cap (Figure 3C), hydathodes (Figure 3D), the junction between secondary inflorescence and the main inflorescence (Figure 3D), young stamen and young siliques (Figure 3E).
Since the adp1-D phenotypes were apparently associated with shoot apical meristem (SAM) activity, we performed in situ hybridization for both wild-type and adp1-D seedlings. The transcript signals of ADP1 were analyzed in shoot apical tissues, using a 300-bp ADP1 cDNA fragment as the antisense probe. ADP1 transcripts were detected primarily in the meristematic regions (e.g., SAM), young leaves and flowers (Figure 3F and 3G). Comparison of adp1-D and the wild type showed that the overall distribution of mRNA was similar, but the signal in the adp1-D mutant was much stronger (Figure 3I and 3J compared with Figure 3F and 3G). No hybridization signals were detected for the control sense probe in wild type (Figure 3H), or the antisense probe in the ADP1 loss-of-function mutant CS123534 (Figure 3K). Taken together these expression results correlated well with adp1-D phenotypes and indicated that ADP1 transcript levels were up-regulated in adp1-D only within the regions where ADP1 transcript was originally detected.
Antisera were raised against ADP1 and were used to immunolocalize ADP1 protein in the root apical meristem region. ADP1 was localized to small intracellular structures in the root cap and the junction between lateral root and primary root (Figure 4A to 4D). To further characterize these structures, transgenic lines were generated to express ADP1 in fusion with GFP or RFP on its N-terminus, under the CaMV 35S promoter. Approximately 30% of the transgenic lines with these tags recapitulated the mutant phenotypes to varying extents, suggesting that the fusion proteins were functional (Figure S4A to S4C). We found that, in the transgenic lines with recapitulated mutant phenotypes, the fluorescent signals of the fusion proteins were localized to punctuated particles with different sizes and shapes, which were distributed ubiquitously in the cells (Figure 4E), suggesting that ADP1 fusion proteins are retained in the endo-membrane organelles. To investigate the nature of these small particles, we crossed the endo-membrane marker lines with these transgenic lines, and found that ADP1 could co-localize with the endosome marker RabF2a-GFP [33] (Figure 4E to 4G) but not with TGN or ER markers (Figure S4D to S4J). Next, we used the fluorescent dye FM4-64 [34] to trace the endocytic dynamics of the fusion protein. After a half-hour treatment with the sterol dye FM4-64, the green signals (GFP tagged fusion protein) were partially co-localized with FM4-64 particles (Figure 4H to 4J). Next we treated the roots for one hour with brefeldin A (BFA), a fungal toxin that targets a subclass of ARF GEFs and is often used as an inhibitor of vesicle transport [35]. As a result, we found that some of the ADP1-GFP granules aggregated into larger bodies which co-localized with FM4-64, while the rest remained scattering in the cytoplasm (Figure 4K to 4M). Taken together, these results indicate that ADP1 resides in endo-membrane vesicles.
The phenotypes of adp1-D (i.e., accelerated growth rate, highly branched shoots and increased number of lateral organs) resemble those mutants of auxin synthesis, transport, and response. To test whether auxin pathways are defective in adp1-D, we first examined hypocotyl length in a temperature shift experiment. The rationale was that decreased level of auxin, either by alteration of auxin biosynthesis, transport and/or signaling, would prevent hypocotyls from elongating under high temperature condition [36]. As shown in Figure 5A and 5B, after shifting the plants from 22°C to 29°C, the hypocotyl length of the wild-type seedlings increased by three to four folds, while that of adp1-D remained unchanged, suggesting that auxin synthesis, transport, or signaling was affected in the mutant. We then crossed adp1-D to DR5:GUS auxin-responsive reporter lines [37] and observed the GUS signal in different tissues of the F3 homozygous plants. In the wild type, DR5:GUS signals were detected mainly in the actively growing regions, such as the SAM, leaf tips, petiole bases, emerging axillary buds and flower primordia (Figure 5C to 5G). However, in adp1-D, DR5:GUS signals were substantially decreased in almost all the meristematic tissues (Figure 5H to 5L).
The decreased DR5:GUS signals indicated either decreased cellular auxin levels or altered auxin signaling. Root growth and the expression levels of two auxin-responsive genes (IAA1 and IAA5) were examined in adp1-D after auxin treatment; an adp1-D axr1-12 double mutant was also generated to test the genetic interactions of ADP1 with auxin signaling mechanisms [21], [22], [38]. The results of these experiments indicated that ADP1 does not interact directly with auxin signaling mechanisms (Figure S5).
The reduced auxin levels in adp1-D were further confirmed by direct quantitation of free IAA levels. Free IAA levels were measured in seedling shoot apices and axillary buds after bolting with quantification methods described previously [39]. The results showed that the free IAA content was indeed reduced in active dividing tissues in adp1-D (Figure 5M and 5N). Furthermore, sensitive mass spectrometry-based method of auxin metabolome profiling was conducted [40] to show that levels of several precursors [indole-3-acetonitrile (IAN), indole-3-acetamide (IAM), indole-3-pyruvic acid (IPyA) and indole-3-acetaldehyde (IAAld)] of auxin biosynthesis in adp1-D were decreased (Table S1). The same trend was also found in adp1-D, in terms of the free IAA level. These results are consistent with the hypothesis that enhanced outgrowth of axillary meristem might be caused by reduction of local auxin levels in adp1-D.
Although the decreased DR5:GUS signals might be caused by down-regulated auxin biosynthesis, the bushy phenotype of adp1-D is somewhat reminiscent of that of the pgp1 pgp19 (abcb1 abcb19) mutants with impaired polar auxin transport. Furthermore, in abcb19 mutant, that is defective in rootward polar auxin transport, levels of IAA at the shoot apex were decreased while levels of oxIAA and oxIAA-Glc were highly increased, as a result of long-term IAA pooling in this region. Accordingly, the polar auxin transport capacity in inflorescences and seedlings of adp1-D and wild type [41] were examined, revealing no differences for 3H-IAA (Figure S6). Due to the fact that this type of transport assay might not reveal differences in the transport capacity in the shoot apical meristem region, microscale transport assays [42] were used to determine if auxin transport capacity out of the meristem/cotyledonary node is affected in adp1-D. No difference between wild type and the mutant was detected by this method (data not shown), indicating that auxin transport is not impaired in adp1-D.
Analyses of auxin biosynthetic mutants suggested a connection of the observed phenotypes to the YUCCA family which belongs to flavin monooxygenase enzyme proteins functioning in auxin biosynthesis [10], [13]–[15], [17]. YUCCA6 has been shown to be associated with an endomembrane compartment [16]. The yuc1 yuc4 double mutants and yuc1,2,4,6 quadruple mutants, which are defective in auxin biosynthesis [17], [18], also exhibited reduced DR5:GUS signal in the SAM and in leaf petioles where axillary meristems were initiated (Figure S7A to S7H). These higher order yuc mutants also exhibited enhanced shoot branching (Figure S7I). Next, adp1-D was crossed with ProYUCCA1:GUS transgenic marker lines. The GUS signal was decreased in almost all the meristematic regions in adp1-D, compared to that in the wild type (Figure 6A to 6H). This result indicated that ADP1 over-expression affected auxin biosynthesis through the down-regulation of YUCCA expression, at least by YUCCA1. Furthermore, we examined the expression levels of all the YUCCA gene family members by qRT-PCR in the axillary buds after bolting, where ADP1 was highly expressed. Results showed that the expression levels of all the YUCCA genes were decreased by three to ten folds in the axillary buds of adp1-D (Figure 6I to 6S). These results confirmed that the auxin biosynthesis pathway might be down-regulated in adp1-D.
Crosses of adp1-D with a Pro35S:YUCCA1 transgenic line that overproduces auxin [15] largely restored wild type growth (i.e., both the rosette branch number and leaf initiation rate) in double homozygous F2 plants (Figure 7A to 7C). This result indicated that global increases of auxin levels through YUCCA1 overexpression could rescue the pleiotropic phenotypes of the adp1-D mutant. In an effort to restrict the over-production of auxin to the expression domains of ADP1, the bacterial indoleacetic acid-tryptophan monooxygenase (iaaM) gene was expressed under the control of the ADP1 promoter in wild type. iaaM catalyzes the conversion of Trp into indole-3-acetamide (the IAA biosynthetic precursor) [42], [43] and has been used to recapitulate Pro35S:YUCCA1 phenotypes in Arabidopsis [15]. Domain-specific iaaM overexpression resulted in more epinastic leaves and aerial rosettes as well as reduced first-order rosette branch number (Figure S8A to S8I). qRT-PCR analysis showed that the severity of the phenotypes was correlated with iaaM expression levels (Figure S8H).
However, flower number and fertility were also reduced in ProADP1:iaaM lines, and, in some lines with severe phenotypes, the inflorescences were pin-formed or produced 3–4 undifferentiated flower meristems before termination (Figure S8B to S8D). These results suggest that overproduction of auxin in the ADP1 expression domain can disrupt the sequential formation of auxin gradients which are required for normal floral development and phyllotactic growth [44]–[46]. Furthermore, these results indicate that auxin synthesis in the ADP1 expression domains does not fully compensate for ADP1 transporter function in the endo-membrane system.
Based on phenotypes and relative gene expression levels, a stronger and a weaker ProADP1:iaaM transgenic lines were selected and crossed with adp1-D. Domain-specific expression of iaaM rescued the bushy phenotype in a manner corresponding to the expression level of iaaM in the parental lines (Figure 7D to 7K), but did not restore wild-type rosette leaf emergence rates (Figure 7L). These results suggest that increasing the auxin content in ADP1 expression domains is sufficient to restore apical dominance, but is insufficient to overcome developmental defects resulting from reduced auxin production in adp1-D that produce accelerated first-order rosette branches.
The experiments described above indicated that ADP1 functions primarily in regulating auxin biosynthesis in the shoot apex and does not impact long-distance auxin transport capacity or auxin transport out of the SAM/cotyledonary node region. The pinformed1 (pin1) mutant exhibits pin-formed inflorescences due to loss of the local auxin gradients that are mediated by the PIN1 auxin transporter at the shoot apex [47], [48]. Developmental and cell biology studies indicate that polar orientation of PIN1 in the SAM region follows and “canalizes” auxin gradients, which was generated by localized auxin biosynthesis and was associated with initiating floral primordia [49], [50]. Modeling analysis with PIN1-fluorescent fusion proteins and fluorescent auxin reporter indicates that these local gradients are formed successively to maintain phyllotactic growth [51], [52]. Since the PINOID (PID) kinase regulates PIN1 trafficking, polar localization of PIN1 is perturbed in pid, making pid form partial pin-formed inflorescences similar to pin1 [53]–[55].
We hypothesize that the domain-specific decreases in auxin biosynthesis observed in adp1-D would weaken these auxin gradients, and thus would enhance the severity of the pin-formed phenotypes. Unexpectedly, the adp1-D pid double mutant displayed an overall appearance similar to adp1-D (Figure 8A, 8B and 8D), except that the mutant produced many flowers with fused petals (Figure 8F and 8H). Statistical analysis showed that the number of the flowers produced on the main inflorescence was at least three-fold more in the adp1-D pid double mutant than that in the pid mutant (Figure 8J), and the first-order rosette branch number was also increased by about two fold in the double mutant (Figure 8L). ADP1 overexpression also largely rescued the pin1 shoot phenotype, with greatly increased flower generation frequency in the inflorescence (Figure 8E, 8I and 8K). However, the first-order rosette branch number of the double mutants was not much increased (Figure 8M), probably due to the fact that pin1 itself already has increased first-order rosette branch number, resulting from 30% decrease in auxin transport out of the shoot apex [20]. Taken together, these data indicate that increased ADP1 activity at the shoot apex is sufficient to overcome a loss of PIN1-mediated canalization required for phyllotactic growth.
Seven homologous proteins were clustered with ADP1 in the same clade in the phylogenetic tree (Figure S3B and S3C), suggesting the possibility of functional redundancy. This is supported by an over-expression experiment in which each of these seven genes was driven under the CaMV 35S promoter to over-express each gene of interest in Arabidopsis plants. All of the transgenic plants recapitulated the adp1-D phenotypes in their T1 generation to different extents (Figure S9A to S9G).
To further elucidate the function of ADP1 and its homologous genes, higher-order loss-of-function mutants were generated between the T-DNA insertion lines of ADP1 and its closest homologs, i.e., At5g19700, At2g38510 and At5g52050 (Figure S3B). The double mutants of each combination had no obvious phenotypes. However, some combinations of the triple mutants and the quadruple mutants exhibited developmental defects. In contrast to the gain-of-function mutant adp1-D, the quadruple mutants showed retarded growth from early developmental stages to maturation (Figure 9C to 9I and 9K to 9M). First-order rosette branch number was also slightly reduced in quadruple mutants compared to wild type (Figure 9J), and first-order rosette branches were generated much more slowly than the wild type (Figure 9N).
To clarify the origin of the aberrant growth pattern, we first examined the shoot apical regions of the wild type, adp1-D and the quadruple mutants by SEM. The results showed that the difference of the phenotypes between wild type, the quadruple mutants and adp1-D started as early as three days after germination (DAG). While adp1-D increased the size of the shoot apical region and produced more leaf primordia, the quadruple mutants showed much reduction of shoot apical size and retarded leaf initiation at 3 DAG (Figure 10A to 10C). This difference persisted from 3 DAG to 5 DAG (Figure 10D to 10F), suggesting that the developmental defects in the quadruple mutants and adp1-D reflect ADP1 function rather than a concomitant developmental effect. Free IAA content in young seedlings and axillary buds of the quadruple mutant was measured by quantification methods described previously [39], showing that the free IAA levels of the quadruple mutants was not significantly different from that of wild type (Figure 10G and 10H). Same results (Table S1) were obtained by using the sensitive mass spectrometry-based method of auxin metabolome profiling [40], which might be due to additional redundancy of untested MATE transporters. However, we were able to detect slightly increased levels of several precursors (IAN, IAM and IPyA) of auxin biosynthesis (Table S1) by the mass spectrometry-based method. These results were consist with the phenotypes of the epinastic cotyledon and slightly increased hypocotyl length observed in the quadruple mutants (Figure S10), since the same phenotypes were also observed in Pro35S:YUCCA1 plants which was believed to have increased auxin levels [15].
Branching pattern is one of the main factors contributing to plant architecture. In Arabidopsis, the number of the first-order and higher-order branches determines the light harvesting efficiency. While the first-order rosette branch number is controlled by strigolactones, the mechanism that regulates higher-order branches is largely unknown. Because many bushy mutants also exhibit reduced fertility, it is proposed that the increased number of higher order branches may be the consequence of sterility. The evidence presented here demonstrates that, in the bushy mutant adp1-D, increased first-order rosette branch number is a directly result of ADP1 overproduction.
Reduced auxin levels, transport, and signaling have long been associated with overproduction of rosette branches. Higher order yucca (auxin biosynthesis), abcb/pgp (auxin transport), and tir1/afb (auxin perception) mutants develop more rosette branches [17], [23], [56]. In the present study, adp1-D mutants produced an increased number of first-order rosette branches (Figure 1D and 1F) and ProADP1:iaaM plants showed a reduced number of first-order rosette branches (Figure S8I). Furthermore, lower levels of free IAA in adp1-D and the rescue of the adp1-D rosette branch phenotype by crosses adp1-D with ProADP1:iaaM transformants confirmed that reduction of local auxin levels in the active dividing meristems caused the bushy phenotype of adp1-D.
The growth retardation observed in the quadruple mutants was opposite to adp1-D phenotypes, suggesting that the genes in the same clade are functionally redundant in the regulation of lateral organ outgrowth in Arabidopsis. Although axillary shoot branching was not obviously changed in the quadruple mutants, the generation rate of the first-order rosette branches was much more slower in the mutants, compared to wild type, which is also opposite to that in adp1-D. In terms of free auxin levels, no significant difference was found in quadruple mutants, compared with wild-type (Figure 10G, 10H, and Table S1). This is probably due to gene redundancy, since the other four homologous genes of ADP1 are still functioning in the quadruple mutants, which may be able to maintain a proper auxin homeostasis in the whole plants. However, levels of the several precursors (IAM, IAN, and IPyA) of auxin biosynthesis were indeed increased in the quadruple mutants (Table S1). Taken together with the excess-auxin phenotypes observed in quadruple mutants (slightly increased hypocotyls and epinastic cotyledons), the auxin biosynthesis in quadruple mutants might be slightly increased. The fact that both gain-of-function and loss-of-function mutants change the plant growth pattern indicates that temporally and spatially appropriate expression of these MATE genes are essential for maintaining plant architecture.
The partial rescue of pid and pin1 pin-formed phenotypes by crosses with adp1-D is more difficult to rationalize. It is quite possible that over-expression of ADP1 actually increases auxin levels within the ADP1 expression domain. This increase is probably sufficient to overcome a loss of PIN canalization in some cell types, but leads to increased auxin catabolism in the majority of cells in the ADP1 expression domain. Similar situation was found in the abcb19/pgp19 auxin transport mutant in which decreased auxin levels and increased oxIAA-Hex levels were observed and attributed to an effect of auxin pooling near the SAM [41]. Alternatively, the decrease of IAA levels in adp1-D could alleviate repression of primordia induction by auxin, or, more likely, generate micro-gradients that stimulate development of new floral primordia.
ADP1 is a member of the large MATE family of transporters that have been implicated in mobilization of ions, toxins and secondary metabolites. As with many other transporter families, MATEs have expanded in plants and function in many aspects: the sequestration of a diverse range of secondary metabolites in vacuoles or their excretion out of the cells, and defense against herbivores and microbial pathogens [57]–[61]. One of the best characterized MATE transporters is TRANSPARENT TESTA12 (TT12), which has been shown to transport proanthocyanidin across the tonoplast [57], [61]. However, a number of MATE proteins appear to regulate the transport of organic acids. FRD3/AtDTX43 controls responses of iron deficiency in plant and is thought to mediate citrate secretion into the xylem and the rhizosphere [60]. The EDS5/AtDTX47 MATE protein functions in salicylic acid signaling for disease resistance [59]. ALF5/AtDTX19 was reportedly involved in the regulation of lateral root formation [58], suggesting a potential role in auxin signaling.
The results presented here indicate that ADP1 is localized to a post-Golgi endomembrane compartment and acts upstream of, or co-ordinately with, YUCCAs in auxin biosynthesis. YUCCA6 was localized to a similar endo-membrane compartment [16], suggesting that ADP1 may function in mobilization of IAA precursors to YUCCAs (for conversion to IAA), or, less likely, movement of IAA out of the endosomal compartments. The function of ADP1 may be homeostatic and involve reversible activity, because a prokaryotic MATE, NorM, which has been crystallized from Vibrio cholera, may exhibit conformational change on substrate binding [32]. Alternatively, ADP1 may simply prevent adsorption of hydrophobic indolic compounds into endosomal membranes or export IAA out of the endo-membrane vesicles. ADP1 may also be involved in auxin cellular homeostasis, which is maintained by PIN5 [62] and PILS auxin transporter [63]. Since so far no auxin exporter in ER has been reported except pollen specific PIN8 [64], it would be interesting to investigate in the future whether ADP1 could balance auxin homeostasis in ER. The lack of successful ADP1 protein expression in multiple heterologous expression systems (e.g., different vectors in different E. coli strains, S. cerevisiae, Pichea pastoris, and SF9 insect cells) has prevented more detailed biochemical characterization.
There are multiple ADP1 homologues in rice, maize and sorghum, sharing 50% to 65% identity at the amino acid level. It is reasonable to speculate that over-expression of these ADP1 genes could also change plant architecture in these crops. Crop plant architecture determines planting density in the field which, to a large extent, affects the light harvest, disease resistance, use of nutrients, and lodging [2]. Understanding the molecular mechanisms in the regulation of plant architecture will therefore provide a basis for modification of the plant architecture of crops, ultimately facilitating crop production.
Seeds of Arabidopsis thaliana, ecotype Columbia, were surface sterilized with 15%? NaClO, stratified for 3 days at 4°C before incubation on Murashige and Skoog (MS) medium containing 1% sucrose at 22±2°C under long-day conditions (16 h light/8 h dark) for 1 week. Seeds of adp1-D mutants were sown on MS medium containing DL-phosphinothricin and drug-resistant seedlings were transferred to soil and grown under the same conditions.
For the temperature transferal experiment, plants were germinated and grown on MS medium for the wild type, and on MS medium containing DL-phosphinothricin for the adp1-D mutant, for 3 days under long-day conditions and then transferred to the same MS medium in a test chamber at 20°C and 29°C for another 7 days before measurement.
For assay of root elongation as an auxin sensitivity test, seedlings were grown on vertically placed MS medium for the wild type, and on MS medium containing DL-phosphinothricin for the adp1-D mutant, under long-day conditions for 4 days before transferal to medium containing 0, 20 nM, 40 nM, 60 nM, 80 nM and 100 nM of synthetic auxin 2,4-D. Root length was measured after incubation for 6 days.
For GR24 treatment, adp1-D were germinated and grown on MS medium containing DL-phosphinothricin for 7 days under long-day conditions and then transferred to MS medium containing 5 µM GR24 or 5 µM acetone for 40 days before phenotype analysis.
Genomic DNA was extracted from homozygous and heterozygous adp1-D mutants and the flanking sequence of the T-DNA insertion was determined by thermal asymmetric interlaced PCR [65]. The specific degenerate primers in the T-DNA border and the random primers for three sequential PCRs were used as described previously [26]. Three primers (P1, P2 and P3-1) were designed for co-segregation analysis. P1 and P2 corresponded to the genomic sequence flanking the T-DNA insertion and P3-1 corresponded to the T-DNA vector sequence (Figure 2A). The primer sequences were as follows: P1 (5′-ATC CCA CTA AAG CAC TGT CA-3′); P2 (5′-TTT AAG CTA CTT ACC GTT GA-3′); and P3-1 (5′-TTG GTA ATT ACT CTTTCT TTT CCT CC-3′). For cloning of ADP1, the primer pair ADP1-F (5′-ATG TGT AACCCA TCA ACA ACA-3′) and ADP1-R (5′-TTA ATA AAG CAC CGT GAT GC-3′) were designed according to the cDNA sequence from the National Center for Biotechnology Information database (accession number NM_119058). Two primers, ADP1-RT-F (5′-CGA ACC GGA CTC TTC CTC GA-3′) and ADP1-RT-R (5′-GGT GAG CAC CGAAGG CTT GA-3′), were designed based on the coding region sequences to detect transgene transcripts in overexpression lines. The primers designed for amplification of the auxin-responsive genes IAA1 and IAA5 were IAA1-F (5′-GCG TCA GAA GCA ACAAGC G-3′); IAA1-R (5′-TCC TTT GTA GCC TTC TCT CTC GGA-3′); IAA5-F (5′-AGA TCT TGC TTC CGC TCT GCA A-3′) and IAA5-F (5′-CCC AAG GAA CATCTC CAG CAA GC-3′), respectively. The primers for detecting the transcription level of endogenous auxin transporters were as follows: PIN1-F (5′-TAC GGC GGC GGACTT CTA CC-3′); PIN1-R (5′-CGG CGA GGA AAC GGA GGT TC-3′); PIN2-F (5′-AATGGC CGT GAA CCC CTC CA-3′); PIN2-R (5′-TTG ACG TTC TCG GCG TCA CG-3′); PIN3-F (5′-CGG TAG CCT CGA GTG GAG CA-3′); PIN3-R (5′-CCG CCG GAC CGAAAT TGG AG-3′); PIN7-F (5′-TCT ACA CCG TCC TCA CGG CG-3′); PIN7-R(5′-AAG TTC GAA AGC CGG CCA CC-3′). The TUB2 (β-tubulin) gene was used as an internal control in real-time PCR (qRT-PCR); the primers for TUB2 used were those described previously (Qin et al., 2005). The qRT-PCR procedure comprised 40 cycles as described previously (Qin et al., 2003). PCR reactions were performed for 26–35 cycles (94°C for 30 s, 58°C for 30 s, and 72°C for 30 s to 1.5 min).
The ADP1 cDNA was amplified from wild-type Arabidopsis by RT-PCR and cloned into the EcoR V site of the pBluescript SK+ vector (designated pBADP1). The presence of ADP1 in the recombinant plasmids was confirmed by sequencing in both sense and antisense orientations. The ADP1 promoter was amplified from genomic DNA using the primers 5′-GCT CAC AGG AGC CTT ACT TAT-3′ and 5′-GAC GGT GAT GAT GATGAT GGT-3′ and cloned into the EcoR V site of pBluescript SK+ (designated pBADP1P).
The CaMV 35S enhancer tetrad was amplified from pSKI015 as described previously [66] and was designated pA4Ehancer. The iaaM gene was amplified from the plasmid pBJ36-iaaM with the primers 5′-ATG TCA GCT TCA CCT CTC CT-3′ and 5′-TAA TTT CTA GTG CGG TAG TTA- 3′ and then cloned into the EcoRV site of pBluescript SK+ (designated pBiaaM). For the construction of the plant expression vector, the pQG110 vector [66], pJIM19 vector and pBI101.3 vector were used. 4Enhancer-ADP1 was constructed by ligation of four DNA fragments: the HindIII-XbaI fragment from pQG110, the HindIII-EcoRI enhancer tetrad fragment from pA4Ehancer, the EcoRI-KpnI ADP1 promoter fragment from pBADP1P, and the KpnI-XbaI ADP1 fragment from pBADP1. The Pro35S:ADP1 construct was obtained by ligation of two DNA fragments: the KpnI-SacI fragment from pJIM19, and the KpnI-SacI ADP1 fragment from pBADP1P. Pro:ADP1-iaaM was constructed by ligation of three DNA fragments: the XbaI-SacI fragment from pBI101, the XbaI-KpnI ADP1 promoter fragment from pBADP1P, and the KpnI-SacI iaaM fragment from pBiaaM. Wild-type plants were used for Agrobacterium-mediated transformation by the floral dip method. The seeds of transgenic wild-type plants were screened on MS medium containing 50 mg/mL kanamycin. The resistant seedlings were transferred to soil.
In situ hybridization was performed as described previously [67]. Antisense and sense probes were synthesized with digoxigenin-11-UTP (Roche Diagnostics) using T7 and T3 RNA polymerases, respectively. The primers used to amplify the DNA template for the probe synthesis were as follows: ADP1-INSITU-F (5′-ATG TGT AAC CCA TCA ACA ACA-3′) and ADP1-INSITU-R (5′-CGG TTA TGTTAG CAA AGG CAA T-3′).
GUS staining was performed in the following steps. Samples were first fixed in 90% acetone on ice for 20 min, then washed thoroughly three times with staining buffer (0.1 M Na3PO4, pH 7.0, 10 mM EDTA, 0.5 mM ferricyanide, 0.5 mM K ferrocyanide, and 0.1% Triton X-100) on ice, vacuum-infiltrated briefly, and incubated in staining buffer containing 50–100 mg X-gluc per 100 mL for 3–12 h.
Tissue sectioning was performed as described previously [68]. Inflorescence stems were collected from the most basal 5 cm of stems of wild-type and adp1-D plants and fixed in FAA solution (50% ethanol, 5% acetic acid, and 3.7% formaldehyde). After dehydration with an ethanol gradient series, the samples were embedded in Historesin (Leica). Sectioning was performed using a Leica microtome and 7 µm sections were mounted on slides. The sections were stained with 0.25% (w/v) toluidine blue O (Sigma-Aldrich) and observed under an Olympus BX51 microscope as previously described [69]. Digital images were captured with a SPOT camera (Diagnostic Instruments) and processed using Adobe Photoshop.
DR5:GUS, ProYUC1:GUS, ProPIN1:PIN1:GFP, ProPIN2:PIN2:GFP, ProPIN3:PIN3:GFP, ProPIN7:PIN7:GFP, ProPIN1:GUS and ProPIN1:PIN1:GUS marker lines were crossed to adp1-D and the homozygous lines in the T3 generation were used for analysis. In all analyses, the parental lines were used for comparison with those in the mutant background. Endo-membrane organelle localization and endocytotic dynamic analysis were performed as described previously [70].
The double mutants adp1-D axr1-12, adp1-D pin1 and adp1-D pid were generated by crossing heterozygous adp1-D with axr1-12, pin1 or pid. The double mutants were identified from the F2 progeny grown in soil by comparison with the parental phenotypes and through PCR-based molecular analyses.
Auxin transport in the inflorescence stem was assayed as described previously [71]. Inflorescence stems of 6-week-old plants were cut into 2.5 cm segments, submerged with one end in a 1.5 ml microcentrifuge tube containing 30 µl MES buffer (5 mM MES, 1% [w/v] sucrose, pH5.5) with 100 nM 3H-IAA in 1.45 µM total IAA at room temperature in the dark for 24 h. Basipetal or acropetal auxin transport was measured in accordance with the orientation of the inflorescence segments. After incubation, the segments were removed and the terminal 5 mm of the non-submerged ends were excised and placed into a scintillation vial containing 2.5 ml scintillation fluid for 18 h before counting with a liquid scintillation counter. Microscale auxin transport assays in seedlings were conducted as described previously [70].
For measurement of IAA content in seedlings, 7 days seedlings of adp1-D, wild type and quadruple mutants in long day conditions were sectioned with a sharp blade, and collected 200 mg tissues (including upper hypocotyls, shoot apical meristems and young leaves without cotyledons) for each genotype. For measurement of IAA content in axillary buds, 200 mg tissues of each genotype were collected three days after bolting, including 1 mm basal leave petiole and the attached newly produced axillary meristems. IAA content measurement was conducted as previously reported [39]. Mass spectrometry-based method of profiling the auxin metabolome [40] was conducted also, which requires much less amount of sample (approximately 20 mg). The samples were collected as the same way as for the above-mentioned method, except that only about 20 mg of each genotype was collected, and three biological replicates were conducted.
The quadruple mutants were generated first by crossing CS123534 and CS878754, and crossing SALK_144096 and SALK_128217 to get the F1 heterozygous generation. Next, the two F2 homozygous mutants from the F1 selfed generation were crossed to obtain the F3 heterozygous mutants. Finally, the F3 selfed generation was screened by PCR to obtain the homozygous quadruple mutants in the F4 generation.
Sequence data for ADP1, YUCCA1, PIN1, PIN2, PIN3, PIN7, AXR1 and PINOID can be found in the GenBank/EMBL data libraries under accession numbers NM_119058, NM_119406.2, NM_106017.3, NM_125091.3, NM_105762.2, NM_102156.1, NM_001035893 and NM_129019.
|
10.1371/journal.pgen.1001114 | Chromatin Landscape Dictates HSF Binding to Target DNA Elements | Sequence-specific transcription factors (TFs) are critical for specifying patterns and levels of gene expression, but target DNA elements are not sufficient to specify TF binding in vivo. In eukaryotes, the binding of a TF is in competition with a constellation of other proteins, including histones, which package DNA into nucleosomes. We used the ChIP-seq assay to examine the genome-wide distribution of Drosophila Heat Shock Factor (HSF), a TF whose binding activity is mediated by heat shock-induced trimerization. HSF binds to 464 sites after heat shock, the vast majority of which contain HSF Sequence-binding Elements (HSEs). HSF-bound sequence motifs represent only a small fraction of the total HSEs present in the genome. ModENCODE ChIP-chip datasets, generated during non-heat shock conditions, were used to show that inducibly bound HSE motifs are associated with histone acetylation, H3K4 trimethylation, RNA Polymerase II, and coactivators, compared to HSE motifs that remain HSF-free. Furthermore, directly changing the chromatin landscape, from an inactive to an active state, permits inducible HSF binding. There is a strong correlation of bound HSEs to active chromatin marks present prior to induced HSF binding, indicating that an HSE's residence in “active” chromatin is a primary determinant of whether HSF can bind following heat shock.
| Many Transcription Factors (TFs) have been shown to bind DNA in a sequence-specific manner. However, only a sub-set of possible binding sites are occupied in vivo, and it remains unclear how TFs discriminate between sequences of equal predicted binding affinity. We set out to determine how a specific TF, Heat Shock Factor (HSF), distinguishes between utilized and unused potential binding sites. HSF is uniquely qualified to study this problem, because HSF is inactive and lowly bound to DNA in unstressed cells and upon stress HSF becomes active and strongly binds to DNA. We compared the properties of the unstressed chromatin between the sites that become HSF-bound or remain HSF-free following stress activation. We find that sites that are destined to be bound strongly by HSF after stress are associated with distinct chromatin marks compared to sites that are unoccupied by HSF after heat shock. Furthermore, chromatin landscape can be changed from a restrictive to a permissive state, allowing inducible HSF binding. These finding suggest that TF binding sites can be predicted based on the chromatin signatures present prior to induced TF recruitment.
| Signal-dependent activation of transcription is a highly regulated process that is dictated by transcriptional activators that selectively identify and function at sequence-specific DNA motifs. The most basic function of sequence specific activators is to discriminate between binding sites in the context of the entire genome [1]–[4], but the mechanism by which this occurs is poorly understood. Two main mechanisms have been proposed that explain the observed in vivo binding specificity (reviewed in [5]): TFs are occluded from cognate site by chromatin structure or TF binding is facilitated by cooperative interactions with cofactors. In vivo, TFs are in competition with chromatin factors, which may limit TF access to cognate binding sites [6], [7]. Early sequence-specific ChIP experiments of homeoproteins revealed that binding sites are preferentially accessible if target motifs are located within active genes [1]. More recently, advances in genome-wide characterization of histone modifications and chromatin structure have begun to identify additional requirements for the binding of TFs. In human cells, it has been shown that the H3K4me1 and H3K4me3 modifications are present at inducible STAT1 binding sites prior to interferon-gamma stimulation [8]. In Drosophila, H3K36me3 has been revealed as an important histone mark for male-specific lethal (MSL) complex binding [9], [10]. However, the Hox proteins primarily discriminate between equivalent predicted binding sites by cooperative interactions with DNA-bound cofactors (reviewed in [11]). These findings indicate that the binding of TFs depend upon the chromatin landscape as well as specific sequence elements, and we set out to determine the extent to which chromatin affects TF binding genome-wide. Characterizing the mechanistic parameters by which TFs locate and bind to target DNA sequences will provide insight into a critical early step in a cell's ability to orchestrate patterns of gene expression in response to developmental, nutritional, and environmental signals.
Heat Shock Factor (HSF) has a conserved function as the master regulator of the heat shock (HS) response from organisms as distantly related as yeast and humans [12]. The HS genes of Drosophila melanogaster are an attractive model system to study the general functions of HSF and its induced transcriptional regulation [13]. HSF is present as a nuclear-localized monomer during non-stress conditions [14]; upon stress, HSF homotrimerization [15] mediates binding to HSF Sequence-binding Element (HSE) motifs within seconds [16], [17], which strongly activates a set of HS genes. While transcription factor binding to DNA is necessary for cis regulation of target genes, not all TF binding is necessarily functional [18]. For instance, HSF has been mapped to over 164 cytological sites on the polytene choromosomes of Drosophila salivary gland cells after HS [19], but only 9 cytological loci exhibit HS-induced transcription elongation factor recruitment and activation [20]–[23]. It remains unclear how HSF discriminates between sites and selectively stimulates functional gene activation.
In this study, we set out to determine the comprehensive set of HSF binding sites in the Drosophila genome and the molecular basis for the binding. We used ChIP (chromatin immunoprecipitation) followed by sequencing [24], adapted for high throughput detection (ChIP-seq) [25]–[27], to map the sites of HSF binding in an unbiased manner with high sensitivity and resolution. We made use of the ChIP-chip datasets from the model organism ENCyclopedia Of DNA Elements (modENCODE) consortium [28], [29], which profiles histone modifications, histone variants [30], insulators [31], [32], and Pol II. These datasets describe critical features of the chromatin landscape in unstressed cells. Using this data, we contrasted the chromatin landscape before HS induction at induced HSF-bound HSE motifs and HSE motifs that remain HSF-free. The roles of many of the modENCODE chromatin features are well established [30]–[33], thus the absence or presence of one or many of these features provides insight into the mechanism of HSF binding.
To determine the comprehensive set of HSF binding sites, we performed two highly correlated, independent ChIP-seq experiments in Drosophila S2 cells [34] for both non-heat shock (NHS) and 20′ HS conditions (Figure S1). We used well-characterized ChIP-grade HSF antiserum [23], [35] which specifically recognizes one HSF-RNAi sensitive Western blot band from whole S2 cell extract (Figure 1A) [35] and generates the expected global HSF-binding pattern observed by indirect immunofluorescence (IF) polytene staining [19], [36], [37]. Despite the specificity observed in these assays, we set out to directly assess specificity in genome-wide ChIP by identifying any HSF-non-specific DNA pull-down. We performed two independent HSF antiserum ChIP-seq control experiments, for each condition (NHS and 20′ HS), in cells that were depleted of HSF by RNAi. This approach approximates a control immunoprecipitation (IP) from cells that lack the factor of interest [38], [39].
HSF-knock down (KD) depleted endogenous levels of HSF to less than 2.5% of control cells as measured by quantitative Western blot (Figure 1A). Importantly, the level of HSF in RNAi depleted cells was reduced at the promoters of well-characterized HS genes, including the highest affinity Hsp83 promoter (Figure S2). Due to the unique presence of tandem HSEs and cooperative HSF binding, the in vitro dissociation constant for the HSF/Hsp83 promoter interaction is on the order of single-digit femtomolar [40], and the Hsp83 promoter harbors the only strongly bound sites during NHS [19] (Figure S1). Since our KD of HSF was successful at reducing HSF levels at the highest affinity binding site, the signal intensity of all HSF-specific peaks should be susceptible to HSF-RNAi depletion as well. Therefore, we discarded peaks that were resistant to HSF depletion, as these are very likely false positives (Figure 1B, Figure S3, Figure S4 and Materials and Methods).
Our analysis of the ChIP-seq data aimed to increase the sensitivity of HSF detection without compromising confidence. To this end, we relied upon two peak calling programs [41], [42] to determine HSF binding sites (see Materials and Methods and Figure S3). Lower confidence peaks were initially considered and later filtered out if found resistant to HSF-RNAi. We detected 464 HSF-specific peaks after 20′ of HS (Dataset S1). We recovered 118 RNAi-sensitive peaks that would have otherwise been discarded because of high false discovery rates (FDR) (Figure S3). In addition, we filtered out 310 non-specific peaks that had FDRs below 0.1 (Figure S3), because they were completely insensitive (and actually increased in intensity) to HSF-KD and exhibit comparable NHS intensity (Figure 1B). Therefore, performing ChIP-seq in cells that were depleted of HSF by RNAi increased the sensitivity and specificity of peak calling.
We derived a position-specific weight matrix (PSWM) [43] and generated an in vivo composite HSF binding site using all 464 HSF peaks occupied after 20′ HS (Figure 2A bottom). Greater than 95% (442/464) of the peaks contained at least one HSE (Figure 2A bottom) with a p-value below 0.001 (Figure S4 and Materials and Methods), indicating that we are primarily detecting HSF directly bound to DNA. In contrast, the distribution of HSE motifs surrounding the HSF-RNAi resistant peaks approximates random expectation (Figure S4). This analysis indicates that the majority of RNAi resistant peaks are false positives that likely result from antiserum cross-reaction with another DNA binding protein, as these peaks are not present in the pre-immune IP. Consistent with the high affinity motif derived by in vitro band shift assays [16] (Figure 2A top), the in vivo HSE is a tandem array of three oppositely oriented five base pair units: AGAAN. In vitro HSF can bind to elements containing three five base pair units, regardless of their orientation relative to one another—although the opposite orientation of three 5 base pair units bound more tightly than direct repeats [16]. Our ChIP-seq study reveals that the opposite orientation of the tandem 5 base pair units is absolutely critical for detectable binding in vivo.
At those peaks that contain HSE motifs, we inferred the HSF binding sites at base pair resolution using the consensus-binding motif derived from this study (Figure 2A bottom). If multiple HSEs were within the 442 HSE containing peaks, the motif closest to the peak center was scored as the HSF binding site (Dataset S2). Our analysis recovered all previously well-characterized HSF binding sites within the promoters of HS responsive genes (Figure 2B, Figure S2, and Dataset S3), including the multi-copy Hsp70 gene (Figure S5). We found that only 20 of the high-confidence HS peaks are detected during NHS conditions, and with a much lower density of tag counts (Figure 2B). Despite the fact that a corresponding NHS peak could not be detected at 422 of the 442 HS peaks, sequence tags are associated with these regions and signal may be above background, but below our threshold for detecting peaks. We considered that true signal should still be susceptible to HSF-KD (Figure S6) and concluded that the majority of these 422 sites are either completely devoid of HSF or contain extremely low, thus undetectable, levels of HSF under NHS conditions. Taken together, our analysis reveals that HSF behaves as we expected from previous molecular analyses of particular genes [16], [17] and from comprehensive, but low resolution, cytological analyses [19]: HSF binds strictly to HSEs and these sites are absent or show drastically reduced occupancy during NHS conditions.
Previous independent reports indicate that ChIP signal intensity positively correlates with motif conformity [3], [4], [44]. We find, however, that HSF binding sites conforming more stringently to the PSWM contain a comparable density of sequence tags as degenerate HSF binding sites (Figure S7A and S7B), suggesting that sequence alone is not driving HSF binding affinity.
Although bona fide HSF binding sites contain highly specific HSE motifs, only a small fraction of potential HSE motifs are occupied by HSF. To search for HSF-free binding sites, we employed a conservative cut-off for conformity to the consensus HSE by using a p-value of 5×10−6 or less [43], while ensuring that the flanking region is mappable [45]. There are 708 HSF-free motifs (Dataset S4 and Figure S8A) that meet these criteria. Less than 15% (107/815) of the mappable HSE motifs with a p-value of 5×10−6 or less are detectably bound by HSF after HS. Upon closer inspection (Figure S6), we find that HSF-free motifs are absolutely HSF-free during NHS, and these same motifs are either unoccupied or infrequently occupied after HS. In contrast, HSF-bound motifs are either very weakly occupied or unoccupied prior to HS, and show strong inducible binding after HS induction. Therefore, these two categories of motifs, HSF-free and HSF-bound, are distinct from one another and are compared below.
We determined the distribution of HSF binding sites relative to annotated genes and promoter regions. Annotated genes account for 60.6% of the Drosophila reference genome (Figure S8B), however, 72% of the HSF-bound motifs are found within gene boundaries (Figure 2C). HSF-bound motifs within promoters (500 bp upstream of a transcription start site (TSS)) were also enriched, accounting for 22% of the total bound motifs (Figure 2C), while such promoter regions only account for 3.4% of the total reference genome (Figure S8B). In contrast, the classification of the 708 HSF-free motifs is much closer to a background distribution; 63% HSF-free motifs are within genes and 5.5% are within promoters (Figure S8A). These results indicate that HSE motifs are not simply enriched within gene and promoter boundaries, but that HSF preferentially interacts with HSEs that are present within genes and promoters.
We hypothesized that HSF discriminates between equivalent HSE sequences in vivo based on the chromatin landscape in which motifs reside. Previous work shows that HSF preferentially binds acetylated nucleosomes in vitro and more recently that the androgen receptor preferentially binds nucleosomes modified with methylated H3K4 in vivo [46], [47]. To determine the extent to which HSF binding is influenced by chromatin in vivo, we compared the NHS chromatin state between the motifs that become HSF-bound or remain HSF-free following HS, excluding the 20 HSF-bound motifs in which HSF was detected during NHS (Dataset S5). Using modENCODE S2 ChIP-chip data [28]–[32], we examined the composite intensity of microarray signal in the region surrounding each HSE. We found that HSF-bound motifs were generally associated with marks of active chromatin, even though these modENCODE signals were generated under NHS conditions (Figure S9). The HSF-free motifs, as a class, were neither enriched nor depleted for any particular factor, histone modification, or histone variant.
Nucleosome occupancy of potential TF binding sites generally restricts TF binding [6], [7], so we examined the distribution of histones and histone variants around HSF motifs. We expected the HSF-bound motifs to be depleted of nucleosomal H3. The composite profiles show that nucleosomal H3 is clearly not depleted (Figure 3); in fact, we observe a slight increase in H3 levels at bound HSEs compared to free HSEs. This observation is in contrast to the general inhibitory nature of nucleosomes and the previous view of HSF binding, as the small set of well-characterized HSF binding sites are devoid of canonical nucleosomes prior to HS [48], [49]. The histone variant H3.3, which associates with active genes [30], displays a peak centered on the HSE motif (Figure 3). These results indicate that HSF binding specificity is not simply dictated by nucleosome-free DNA sequence.
In recent years, considerable attention has focused on the plethora of covalent histone modifications that occur on the N-terminal tails of histones, the enzymes responsible for catalyzing histone modifications, and the functional consequence of each modification. Acetylation of histone residues H3K9, H3K18, H3K27, H4K5, H4K8, and H4K16 were found to associate with HSF-bound motifs (Figure 3 and Figure S10). Each one of these acetylation marks has previously been shown to mark active chromatin [33], [50]. We find that the methylation marks H3K4me3 and H3K79me2, which associate with active genes [33], [51], are also enriched around the HSF-bound HSEs (Figure 3 and Figure S10). Mono-ubiquitylation (Ub) of H2B, a modification that is necessary for methylation of H3K4 [52], correlates with HSF-bound motifs as well (Figure S10). Conversely, marks of repressive chromatin, H3K27me3 and H3K9me3, were found to be depleted or at background levels (Figure 3 and Figure S10).
We considered that HSEs and histone marks cooperate to specify HSF binding. Transcription factors can bind acetylation and methylation marks through specific domains such as bromodomains, chromodomains and PHD domains (reviewed in [53]). For example, the MSL complex harbors a chromodomain, accounting for preferential recognition and binding of the H3K36me3 mark in Drosophila [9]. Interestingly, the HSF protein is devoid of all of these domains, and thus cannot be binding to DNA and histone methyl or acetyl marks cooperatively by any of these well-characterized interactions.
Comparison of HSF-bound motifs with TF binding data reveals that HSF co-localizes with factors that are associated with active transcription. The presence of Pol II is the foremost indicator of an active gene or a gene that is primed to be activated. The composite Pol II profile at HSF-bound HSEs exhibits a striking peak, even in instances where bound HSEs are within intergenic regions (Figure 3). Likewise, we observe a strong BEAF (boundary element-associated factor) signal centered on HSF-bound motifs (Figure 3). BEAF is an insulator that localizes to transcriptionally active and paused polymerase-harboring genes [32]. The multifaceted TF, GAGA Associated Factor (GAF), is associated with both paused polymerases and HSF-bound motifs [54] (Figure 3). Taken together, these profiles indicate that HSF binds to sites that contain hallmarks of open and active chromatin.
These composite profiles provide an average view of HSF-binding, which could potentially be influenced by a small population of binding sites. We used the available “Regions of Significant Enrichment” tracks from modENCODE to determine which motifs (HSF-bound or HSF-free) were present within the significantly enriched regions of a given factor or modification. We employed the Fisher exact test to determine whether HSF-bound motifs were associated with each factor compared to HSF-free motifs and vice versa (Table S1). Depicted in Figure 4 and Figure S11 are the fractions of HSEs that are present within a given region of enrichment (enriched is colored yellow, unenriched is blue). Strikingly, only 30 (7%) inducible HSF-bound sites do not contain any tested activation marks prior to HS. This analysis reveals a statistically significant association (p-value<0.05) of HSF-bound motifs with 17 different histone modifications or chromatin-bound factors that have previously been shown to be associated with active chromatin (Table S1, Figure 4, and Figure S11), regardless of whether the motifs are classified as intergenic, promoter proximal or within genes (Table S2, Table S3, and Table S4). Unlike previous genome-wide TF binding data that show the co-occupancy of many TFs and histone marks, we are able to show that these chromatin features are present before any detectable HSF binding (Figure S12).
We have shown that the presence of activation marks strongly influences the pattern of HSF binding, so we next determined whether quantitative differences in individual marks play a role in the degree of HSF binding. For each HSE that is enriched for a mark or factor in Figure 4, we compared the ChIP-chip intensity of each mark or factor during NHS to the intensity of induced HSF binding following HS. We found a modest, but significant (p-value<0.05), correlation between the intensity of BEAF, tetra-acetylated H4, and H3K18ac with HSF binding intensity (Figure S13).
Considering that the intensity of any one mark only modestly affects HSF binding, we set out to determine whether distinct patterns of TF profiles and histone modifications affect HSF binding intensity. Sets of histone modifications and TFs occur together in distinct combinations on the genome-wide scale in eukaryotic cells [18], [33], [55], [56], and this chromatin landscape can be used to predict and characterize functional regions of the genome [57], [58]. We used cluster analysis [59] to determine whether TF factors and histone modifications showed clear binding patterns at both classes of HSE motifs (Figure 5). This clustering shows that, generally, any single HSF-bound motif is enriched for many activation marks. HSF-free motifs are primarily found in regions with background levels or depleted levels of activation marks. Consistent with our composite profiles, nucleosomal H3 and H2A were not depleted at the bound HSEs prior to HSF binding and H3.3 is generally enriched. Our findings indicate that HSF-accessible chromatin is not synonymous with nucleosome vacancy, but rather, with marks of loose or active chromatin.
Clusters are not absolutely delineated by the presence or absence of a given factor or set of factors; however, we note general properties of individual clusters. For instance, ubiquitous acetylation of histone residues and high levels of H3K4me1 characterize HSF-bound cluster three, while cluster four contains modest levels of every factor and modification tested (Figure 5). Considering that motif conformity does not significantly affect HSF-signal (Figure S7A and S7B), we tested whether clustering HSEs cleanly separated strong and weak binding sites. We observe that cluster four generally exhibits less intense HSF binding, while cluster one, which is driven by intense Pol II and GAF signal, contains stronger HSF binding sites (Figure S7C). These patterns, however, are not sufficient to account for differences in HSF binding intensity, as the HSF intensity in any p-value quartile or cluster overlaps with all other classes. Ultimately, it is likely that the rules that govern TF binding and intensity of binding are a complex nonlinear system, which results in motif accessibility.
The strong correlation between open chromatin and HS-induced HSF suggests that open chromatin dictates HSF accessibility. To test this hypothesis, we directed a change in the chromatin landscape, from the restrictive to the permissive state, at an unbound HSE and then examined HSF binding following HS. HSF has been shown to selectively occupy the ecdysone inducible 75B cytological locus, only when the locus is transcriptionally “puffed”, in salivary gland cells [19]. We found an HSF-free motif that resides within the body of an ecdysone inducible gene isoform, Eip75B, which can be inducibly expressed in S2 cells (Figure 6A) [60]. We confirmed that this motif is minimally bound by HSF after HS, and is below the threshold for peak detection by ChIP-seq (Figure 6B). Ecdysone treatment alone results in RNA Pol II recruitment to the body of the Eip75B gene, but does not affect HSF occupancy of the HSE motif (Figure 6B). H3K9ac and tetra-acetylated H4 increase above the background threshold (top dashed line), while H3 levels are unaffected after a 30′ ecdysone treatment (Figure 6B). Recall that prior to HS, between 70% and 80% of the HSF-bound HSEs are significantly enriched for each RNA Pol II, tetra-acetylated H4 and H3K9ac (Figure 4). A 30-minute ecdysone pre-treatment changes the chromatin landscape and allows HSF to strongly occupy the motif following HS (Figure 6B).
Pre-treatment with ecdysone, followed by HS, not only allows HSF binding at this HSE, but also causes a concomitant increase in local H4 and H3K9 acetylation and decrease in RNA Pol II intensity (Figure 6B and Figure S12). Increased acetylation of histones is consistent with HSF's ability to recruit the acetyltransferase CREB Binding Protein (CBP) to HSF bound sites [61], [62]. At first glance, it is unintuitive that RNA Pol II intensity is compromised following heat shock (Figure 6B and Figure S12). However, this molecular analysis confirms a long-standing observation that following HS, HSF has the ability to repress ecdysone inducible puffs and general protein synthesis [19], [63]. While the mechanism of HSF-mediated repression is unknown in Drosophila, it is tempting to speculate that HSF can act as a roadblock to RNA Pol II within the bodies of active genes (Figure S14).
It has long been known that HSF inducibly binds to many sites and only a subset of sites are transcriptionally activated by HS [19], [64]. These studies, however, did not have the resolution to determine if HSF binding sites did not lead to mRNA production simply because HSF was not promoter-bound. In all well-characterized cases of Drosophila HSF-induced transcription, HSF binds to the promoter. To determine whether promoter-bound HSF is sufficient to upregulate the local gene, we measured mRNA abundance at candidate genes during NHS and after a 20′ HS (Figure 7). Note that HSF is inducibly bound at each gene after 2′ minutes of HS (Figure S15), allowing sufficient time for mRNA accumulation (reviewed in [65]). We observe a continuum of induced mRNA accumulation, from the highly induced Hsp26 gene, to genes that are unaffected by HSF binding (Figure 7).
Previous genome-wide ChIP experiments report that TF binding intensity generally correlates with functional binding [18], [66]. The ChIP-seq signals that we observe are directly comparable to qPCR quantified ChIP material, indicating that the quantitative properties of ChIP were retained in our sample preparation (Figure S16). Because HSF acts as a potent acidic activator [67], we hypothesized that all genes that exhibit inducible and strong promoter binding of HSF would be activated. HSF can activate when bound moderately to the promoters of genes, as is the case for the CG3884 and CG6770 genes (Figure 7 and Figure S15). Surprisingly, HSF binds inducibly and intensely to the CG3016 and CG13025 promoters (Figure S15), but these mRNA levels remain unchanged (Figure 7). Selective activation is not unique to HSF, as both ER and p53 bind the promoters of genes in a signal-dependent manner, but transcription of some local genes remains unaffected [2], [68].
To investigate how HSF may be selectively activating local genes, we used the ChIP-chip data to look for patterns of histone modification and TF binding that separates functional promoter-bound HSF sites, which can activate gene expression, from promoter-bound HSF that does not result in gene activation. We noticed that GAF was present at many up-regulated genes, and in contrast, BEAF was present at unregulated genes (Figure S17). Previous work has shown that GAF is important for the activation of HS genes [69], but our results indicate that GAF is not necessary for HSF activation (Figure S17). BEAF has been shown to function as an insulator [31], [32]; therefore, we speculate that BEAF is blocking the activation function of HSF at unregulated genes. Previous work has implicated paused polymerase as an important criterion for activation from an Hsp70 promoter [70]. Using promoter-proximal enriched Pol II and pausing factor (NELF) data [54], however, we did not see a significant correlation between these pausing hallmarks and activation potential using these 16 genes. In the same way that chromatin signatures affect the binding of HSF to a motif in vivo, we expect that chromatin landscape and individual gene properties act together to dictate the activation potential of activator-bound genes.
We present an experimental approach that increases the sensitivity and power of determining TF-bound sites by ChIP-seq, and we use this approach to characterize the binding profile for HSF under both NHS and HS conditions. Our analysis revealed that HSF binding is dependent upon an underlying HSE motif, although the primary HSE sequence is not sufficient to confer HSF binding. HSF-bound HSEs were found to be associated with a chromatin landscape that harbors active marks prior to HSF binding. Lastly, we demonstrated that promoter-bound HSF is not sufficient to activate local genes.
The ChIP-seq method is used routinely to determine genome-wide factor binding profiles; however, important controls and variations in the ChIP protocol more fully exploit this approach. Our implementation of the control RNAi knockdown of HSF allowed us to eliminate the genome-wide set of false positive signals that were resistant to this knockdown, and prevented the elimination of many true positive binding sites. Another rigorous and complementary control for specificity includes performing independent ChIP experiments with multiple antiserum preparations, each of which is affinity purified with nonoverlapping antigens [18], [38]. The details of ChIP-seq chromatin preparation can also enhance peak detection [71]–[73]. Additional crosslinking agents [74] and crosslinkers that target particular types of protein/DNA interactions, such as exclusively probing direct protein/DNA interactions with UV light [1], [75], can also augment the type and quality of information obtained by the basic ChIP-seq strategy.
The non-sequence dependent specificity observed by TFs can be explained by non-mutually exclusive mechanisms: DNA binding is specifically inhibited by repressive chromatin, aided by active chromatin, or mediated by cooperative interactions with chromatin factors. Here, we report that repressive marks contribute minimally to restrict HSF binding, as only a small fraction of HSF-free motifs are associated with repressive chromatin (Figure S11). Additionally, we observe that chromatin containing background levels of active and repressive marks is unfavorable to inducible HSF binding—the default state of an in vivo HSE can be considered inaccessible. In contrast, HSF inducibly binds to sites that contain TFs and marks of active chromatin prior to HS induction. We have shown that the chromatin landscape can be modified to the permissive state and result in recognition and binding of a previously unbound HSE. This result suggests that HSF does not primarily function to bind DNA cooperatively with other factors, but simply co-occupies the same regions as other TFs, due to the accessible nature of the DNA. These results provide a framework for understanding the binding selectivity of HSF, and we look forward to mechanistic studies that solidify the rules of in vivo binding specificity.
Activators are generally thought to bind to promoters and recruit either Pol II or coactivators to produce productively elongating Pol II. HSF recruits the acetyltransferase CREB Binding Protein (CBP) and a methyltransferase, Trithorax, directly to HS genes [61], [62]. Paradoxically, this study shows that the chromatin landscape at HSF binding sites contains considerable histone acetylation and methylation prior to detectable HSF binding. HSF recruits these enzymes after HS to broaden the domain or increase the level of histone modifications (Figure 6 and Figure S12). Another, non-mutually exclusive, possibility is that cofactors other than histones are the functional targets of recruited transferases. Although we describe the landscape at HSF binding sites prior to HS, it still remains unclear which factors are responsible for setting up or maintaining the accessibility of these motifs. Furthermore, many HSF-binding sites are probably passively occupied because they happen to be accessible and HSF binding is non-deleterious [76], but these sites likely have no function in the HS response. The global chromatin landscape is dynamic throughout development and environmental changes; therefore, we expect that the HSF binding profile at non-functional sites is dynamic as well. Nonetheless, the HS response is a ubiquitous cellular response, so functional sites are likely to be evolutionarily constrained at the sequence level [77], [78], and actively maintained in the accessible state at the level of chromatin organization.
The maintenance of functional HSF binding sites may be occurring as a result of a specific class of activators. Non-traditional activators, such as GAF, are known to recruit cofactors that establish an accessible chromatin state, as opposed to directly activating transcription of the local gene (reviewed in [79]). This general mechanism has been characterized at the phaseolin gene in Arabidopsis [80] and at the PHO5 gene in yeast (reviewed in [81]). Taken together, this suggests a step-wise process whereby a repressed site can be potentiated for activator binding and subsequently activated. Additionally, it has been shown that active marks are not simply a product of transcription, as the active marks that are associated with intergenic DNaseI hypersensitive sites and putative enhancers are not correlated with respective gene expression [33]. Our results suggest that the landscape may be marked with active histone modifications to allow binding of activators that can stimulate transcription; therefore, the presence of a modification would not be expected to correlate with gene expression if the activator has yet to bind. Further investigation of activator binding sites during non-induced conditions will determine the generality of this observation.
Our candidate gene analysis shows that HSF is not sufficient to activate local genes. Although inducibly activated genes are occupied by their cognate transcriptional activator near the TSS [4], [82]–[84], it remains unclear how the majority of activators discriminate between locally bound genes to selectively activate. Strikingly, Caudal exhibits promoter element-specific activation, specifically activating genes that contain the Downstream Promoter Element (DPE) [85]. Previously, we presented evidence that the presence of a paused polymerase facilitates activation from an Hsp70 promoter [70], but it is unclear whether or not this is true for the majority HSF-inducible genes. Combinations of promoter features and gene properties are likely necessary for activation. One certainty, however, is that the recent emergence of genome-wide expression and binding data makes the characterization of complex regulatory mechanisms more exciting and promising than ever.
The ChIP protocol has been previously described [86]. In short, S2-DRSC (lot 181A1) cells were grown in Schneider's media with 10% FBS (lot ASD29137), consistent with modENCODE experiments. Heat shocked cells were instantaneously shifted to 36.5°C by the addition of an equal volume of 48°C media to the 25°C culture. Heat shocked cells were instantaneously cooled to room-temperature and crosslinked with a final concentration of 2% paraformaldhyde for one minute; this shorter duration of crosslinking with higher concentration of paraformaldehyde was found to increase the signal-to-noise ratio. Instant cooling to room temperature and immediate crosslinking allows the heat shock and NHS samples to be crosslinked at the same efficiency and directly compared. We cannot strictly rule out the possibility that instantaneous cooling cells to room-temperature for one minute contributes to the recovery and dissociation of HSF at lower affinity sites, including the 708 HSF-free sites. However, paraformaldehyde penetrates cells quickly to effectively block further cellular changes, and HSF's DNA binding activity is only modestly affected even after a 30 minute recovery from HS [87]. Crosslinking was quenched by the addition of glycine to a final concentration of 250 mM and the extract was sonicated as previously described [86], but for three-times the duration to increase enrichment [88]. The Protein-A beads were blocked with BSA (1 mg/ml) and Polyvinylpyrrolidone (1 mg/ml) prior to the IP and freshly thawed antiserum was used for each IP, which also increased signal compared to noise.
The sample preparation was previously described [89], with some modifications. Only one size selection, after adapter ligation, was performed. Thirteen cycles of PCR were performed. Quant-iT Pico Green (Invitrogen) staining was used to quantify the DNA sample. Samples were submitted to the Cornell DNA Sequencing and Genotyping Lab and run on the Illumina Genome Analyzer II.
RNAi-mediated HSF knockdown was performed as previously described [69]. Primer sequences are available within Dataset S6.
Sequence tags were aligned to the Drosophila melanogaster April 2006 release of the reference genome using MAQ [90]. We considered those tags that aligned uniquely with less than 4 mismatches. A summary of the sequencing tag counts and unique alignment counts for each condition are supplied in Table S5. The text files containing raw sequence tags and uniquely aligned tags were deposited into NCBI's Gene Expression Omnibus (GEO) [91], accession number GSE19025. Two programs [41], [42] (referred to as MACS and SPP, respectively) were independently used to call peaks with the MAQ mapped sequences for each experimental condition. The parameters we used for each program are indicated in Figure S3. The Subpeaks package was further used to dissect the few areas of broad MACS enrichment. Using SPP, we determined that we achieved saturation at this depth of sequencing. Either of two criteria was used to consider a peak RNAi-sensitive: 1) a peak coordinate was called in both the experimental and RNAi dataset and the peak is depleted in the RNAi data more than the Hsp83 promoter depletion; 2) a peak was only called in the experimental dataset and the corresponding region of the RNAi dataset was depleted by at least 3-fold. The intensity used to calculate depletion was defined by the normalized tag count of mapped 5′ ends in the 240 base window centered on the experimental peak center coordinate. SPP and MACS were considered to have called the same peak if the SPP peak center was within the Subpeak enrichment boundary or the broader MACS enrichment boundary. The window that corresponds to the 60 bases flanking each peak center was used as input for MEME [43]. MAST and Tallymer were used in conjunction to determine the 100% mappable (for 40mer tags in the 400 bp window centered on the motif) HSF-free motifs [43], [45].
The individual labs that generated the chromatin landscape data also validated their results. Table S6 provides the respective modENCODE ID or GEO accession number for each dataset used in this study.
Drosophila S2 cells were treated with 1000×20-hydroxyecdysone (20E) in 2% ethanol, at a final concentration of 1 µM for 30 minutes. ChIP was performed immediately after 30 minutes of 20E, for the NHS treated cells, or after a 20 minute HS. Two independent experimental replicates were performed for “20E/NHS” and “20E/HS”. Control cells were treated with 2% ethanol as the vehicle. Two independent control samples were performed, and the values were compared to a no treatment control. Vehicle treatment was comparable to no treatment, so we combined the measurements for a total of three independent biological replicates for both NHS and HS conditions. The error bars indicate the standard error of the mean. Importantly, we calculated two important background measurements. First, we performed ChIP with Rabbit IgG for each condition, to control for non-specific pull-down by IgG or beads. Secondly, we performed ChIP-qPCR at eight regions where each factor or modification is not enriched in untreated conditions [29], which controls for non-specific background pull-down by each antibody. Generally, the background IP by histone modification antibodies is high as measured by raw percent input, presumably do to cross reaction with unmodified histones, so this measurement is necessary in order to assign a threshold for enrichment in ChIP-qPCR assays (the top dashed line).
RNA levels were measured as previously described [92].
Dataset S6 contains the primer sets that were used for measuring mRNA abundance. Table S7 contains the primer sequences that were used for ChIP-qPCR.
|
10.1371/journal.pcbi.1006523 | Self-organized traffic via priority rules in leaf-cutting ants | Ants, termites and humans often form well-organized and highly efficient trails between different locations. Yet the microscopic traffic rules responsible for this organization and efficiency are not fully understood. In previous experimental studies with leaf-cutting ants (Atta colombica), a set of local priority rules were isolated and it was proposed that these rules govern the temporal and spatial organization of the traffic on the trails. Here we introduce a model based on these priority rules to investigate whether they are sufficient to produce traffic similar to that observed in the experiments on both a narrow and a wider trail. We establish that the model is able to reproduce key characteristics of the traffic on the trails. In particular, we show that the proposed priority rules induce de-synchronization into clusters of inbound and outbound ants on a narrow trail, and that priority-type dependent segregated traffic emerges on a wider trail. Due to the generic nature of the proposed priority rules we speculate that they may be used to model traffic organization in a variety of other ant species.
| Ants often form trails to transport food and supplies they find back to their nest. These trails have a function similar to the roads that connect people’s homes with the local mall, but while the traffic rules that cars on our roads are supposed to follow are well known the traffic rules ants use on their trails are still relatively unknown. Earlier experiments with leaf-cutting ants have suggested a set of simple traffic rules that ants may be attempting to follow on their trails. However, it is difficult to experimentally verify the link between the proposed rules and the observed traffic organization. Modeling is a useful way to link the behaviors isolated at the individual level and the pattern recorded at the collective level. Here we present and analyze a computational model based on the proposed traffic rules. We find that, with some modifications, the proposed rules are indeed sufficient to reproduce key features of the overall ant traffic observed in the experiments. Strengthening our belief that these traffic rules might be employed by the leaf-cutting ants to regulate traffic organization and due of their simplicity we speculate that similar rules may be used by other ant species.
| Animal collective movement is a widespread phenomenon that occurs at various spatial and temporal scales in a variety of living organisms from cells to pedestrians [1–4]. Often there are no identifiable leaders or coordinators present and group coordination relies on a completely decentralized process. The global pattern is not explicitly encoded but emerges from numerous interactions between individuals that only have access to local and limited information [5–7].
In many situations the motion within the collective is unidirectional because it is related to migratory phenomena and involve individuals moving in the same direction. Social insects and humans are some of the rare organisms in which the movements within the collective are predominantly bidirectional [8–12]. In particular, ants are central-place foragers and must return to their nest with the food collected after each foraging event which often lead to the formation of trails with a steady stream of traffic between the nest and the food source. In some species the traffic flow on these trails can be extremely high, reaching more than a hundred ants per minute, e.g. red wood ants [13], leaf cutting ants [14] and army ants [15]. When the local concentration of individuals is very high on the trail the high rate of head-on collisions may slow down the individuals [16–18]. These effects can provoke group dysfunctions and reduce the colony’s overall foraging efficiency. Such negative effects can be avoided if ants make use of dispersal mechanisms allowing a better organization of the traffic [8].
Traffic in ants can be organized either on a spatial or on a temporal scale [8]. Spatial organization of traffic is characterized by lane segregation, i.e. the flows of inbound and outbound ants are not completely intermingled [19–21] and the temporal organization of the flow is characterized by a sequence of alternating clusters of inbound and outbound ants [16,22]. To date, the emergence of both the spatial and the temporal organizations observed in these systems remains largely unexplained. In particular, the microscopic traffic rules that individual ants follow when navigating on trails are largely unknown.
In an attempt to isolate the microscopic traffic rules experimental studies on traffic organization in leaf-cutting ants Atta colombica on a narrow trail [22] and on a wide trail [20] were performed. Leaf-cutting ant trails guide workers to and from the foraging site where they cut vegetation into small fragments and transport them back to the nest. These fragments are then incorporated into a fungus on which the colony feed. In the experiments, to reach the leaf source, ants were forced to move on either a narrow trail allowing the passage of only one moving individual at a time [22] or on a wide trail ten times larger [20]. On the narrow trail de-synchronization of inbound and outbound traffic involving the formation of alternating clusters of inbound and outbound ants was observed and on the wide trail a degree of lane segregation with leaf carrying ants travelling almost exclusively on the central section of the trail was described. Summaries of the results obtained in the experiments may be found in the supporting information S1 Table (narrow trail) and S2 Table (wide trail). The authors suggested that both organizations may result from a set of local priority rules observed at the individual level when ants encountered other ants on the trail [20, 22]. However, whether these proposed individual priority rules are sufficient to produce the observed temporal and spatial traffic organization is unknown and to investigate this modelling is required.
There are many well known models for ant traffic, most of which are mean-field models studying the macroscopic properties of the traffic on trails [23–31]. However, to investigate the group level traffic that emerges from repeated local interactions between moving individuals so-called self-propelled particle models are more appropriate. Self-propelled particle models are spatially-explicit individual based models where particles interact locally with each other according to a set of rules. These models range from minimal models used to investigate fundamental properties of collective motion [32–35] to more involved species specific models of collective motion in everything from cells to insects, fish, birds, sheepdogs and pedestrians [9,10,36–44]. The self-propelled particle model approach has been successfully applied to model ant traffic in army ants [19] and black garden ants [45]. However, in these models the characteristic features of outbound and nest-bound ants were not distinguished despite the fact that variation in their maneuverability and speeds exist due to food transport [22, 46].
Here we introduce a self-propelled particle model to reproduce the experiments with leaf-cutting ants Atta colombica on a narrow trail [22] and on a wide trail [20], and to investigate whether the local priority rules proposed in these studies are sufficient to reproduce the traffic organization observed.
To focus our investigation on which macroscopic properties of the traffic the local priority rules alone are responsible for we simplify the model ants (particles) in several ways. In particular, particles are assumed to have constant length 1cm and constant type dependent speeds. Mimicking the experimental setup the particles move on a 300 cm long one-dimensional trail connecting the nest and a leaf source. Particles leave the nest and the leaf source according to a Poisson process characterized by the rate parameter μ. At the beginning of each simulation particles are present at the nest and the leaf source. Following the experimental work we consider three types of particles: outbound (O), inbound unladen (U) and laden particles (L). The probability of a particle being laden when leaving the leaf source is 0.24 as observed in the experiment (S1 Table). Due to the small difference in speed observed between outbound (O) and unladen ants (U) we set the speed to sOU = 2.3 cm/s for both types of particles. The speed of laden particles (L) is set to sL = 1.9 cm/s (S1 Table).
On the narrow trail the positional update formula for each particle is given by
x(t+Δt)=x(t)+δ(t)shΔt,
where x(t) is the x-coordinate of the particle on the trail at time t (the nest is at x = 0 and the leaf source at x = 300), s ∈ {sOU, sL} is the speed of the particle, h ∈ {−1, 1} is the heading of the particle (1 for outbound and -1 for inbound particles), and Δt = 0.1 is the time step. The value of δ(t) ∈ {0, 1} depends on the interactions between particles at time t. δ(t) is 1 if the particle is given the way, and 0 if the particle stops and gives the way. Two particles interact when they are within a distance of 1 + 2 Δt sOU of each other to account for the size of the particles. The interactions between particles are specified by a set of local priority rules identified in [22];
In order to cover all interaction possibilities we also included the following two rules not quantified in [22];
The wide trail experiment [20] was conducted using the same experimental procedure as in the narrow trail experiment and the results are summarized in S2 Table. Ants were forced to move on a 5cm wide and 300cm long trail linking the nest and the leaf source (Fig 1). In addition to the observations made in the narrow trail experiments on the wide trail the spatial organization of the traffic was also quantified. The number and type of ants (O, U and L) passing the midpoint of the trail in three different zones; a central zone (2.5cm wide) and two marginal zones (each 1.25cm wide) were recorded (Fig 1).
The main difference between the narrow and wide trail settings is that there is room for ants/particles to turn on the wide trail so they are not forced to stop when encountering an ant/particle with higher priority. Instead the ant/particle can turn to avoid collision and we adapted the narrow trail model to the wide trail by replacing all instances of stop with turn. The traffic rules for particles on the wide trail are presented in Table 2.
The positional update formula for each particle in the wide trail model is
{ x(t+Δt)=x(t)+scos(θ(t))Δtw(t+Δt)=w(t)+ssin(θ(t))Δt
where x(t) is the x-coordinate of the particle on the trail at time t (the nest is at x = 0 and the leaf source at x = 300), w(t) is the w-coordinate of the particle at time t (the walls are located at w = 0 and w = 5), s ∈ {sOU, sL} is the speed of the particle, Δt = 0.1 is the time step, and θ(t) is the heading of the particle at time t. The heading θ(t) = θd +θI(t) is composed of two components:θd ∈ {0, π} which is equal to 0 for outbound and π for inbound particles, and θI ∈ {−π/2,0,π/2} which is the turning angle resulting from an interaction and depends on the traffic rules (Table 2). During an interaction a particle i will not turn (θI = 0) if it is given the way by particle j. It will turn if it is not given the way by particle j. In that latter case, it will turn up if wi (t) ≥ wj (t) or down if wi (t) < wj (t). Note that when a particle turns its displacement in the x-direction is 0, because cos(±π2)=0, so its arrival to the nest/leaf site is delayed by Δt = 0.1s by each turn time step.
The cooperative rule where unladen following a laden are sometimes allowed passage is not relevant in the wide trail setting because particles are turning instead of stopping. As in the narrow trail case we use the experimentally observed total flow to set the rate parameter μ associated with the Poisson process of ants leaving the nest and leaf source. The average flow in the wide trail experiments was 8803 ants/hour and choosing μ = 0.8 results in a flow of 8999.4 (s.d. 97.8) and we use this value for the simulations. Unlike in the narrow trail case, where all ants entered and moved on the trail at the same vertical position, on the wide trail ants entered the trail from the nest via a small opening centered on the trail and from the leaf source via a narrow wooden stick attached to the center of the trail. So the ants entered close to the center on both sides of the trail and then scattered in the w-direction. To model this we assume that the vertical entry position (w(0)) for each ant is normally distributed with mean 2.5 (center of the trail) and standard deviation 0.8. This ensures that almost all entry positions lie between 0 and 5 and if the generated w(0)>5 we set w(0) = 4.99 and if w(0)<0 we set w(0) = 0.01.
To analyze the model we ran 1000 simulations and calculated the proportion of each particle type in each zone of the trail, in addition to the mean and maximum group sizes, and compared the results with the experimental results (S2 Table). We also ran a set of simulations with turning replaced by stopping as on the narrow trail to investigate the causal effects of turning with respect to the spatial organization on the wide bridge.
The main experimental finding was that a de-synchronization of inbound and outbound traffic occurred on the narrow trail that involved the formation of alternating groups of inbound and outbound ants. De-synchronization of this type also emerged in simulations of the model and the resulting groups share several properties with the experimentally observed groups. See S1 Fig for an illustration of the concept of de-synchronization. In [22] four statistics were used to quantify the traffic organization and structure of the groups: (i) group size distribution, (ii) proportion of laden ants in groups of size N, (iii) proportion of laden ants at position P in a group, and (iv) proportion of groups of size N led by a laden ant. A comparison of the experimentally observed distributions and the distributions generated from simulations are presented in Fig 2. We see that overall the distributions generated by simulations are comparable with those obtained in the experiments. In particular, for group sizes larger than 1 the experimental means are essentially contained within the min-max bounds of the distributions generated by simulations for all four statistics. The main discrepancy between simulations and experiments is the overrepresentation of groups of size 1 in the group size distribution generated by simulations (Fig 2A). In addition, the maximum group size observed in experiments was 74 (S1 Table) and in simulations 77, and the average group size observed in experiments was 5.2 (S1 Table) and in simulations 4.2.
The main finding in the experiment was that a degree of lane segregation occurred on the wide trail. Laden ants returning to the nest travelled mostly in the central zone of the trail, while unladen ants (U) more frequently travelled in the marginal zones of the trail. For outbound ants (O) about two thirds of them travelled in the central zone of the trail. The asterisks in Fig 3A and 3B shows the results for each type of ants in each of the 12 experiments. The boxplots represent the corresponding measurements in simulations with the wide bridge model using the turning rule (Fig 3A) or the stopping rule (Fig 3B). We found that the wide bridge model using the turning rule reproduces the overall trend of laden particles moving almost exclusively in the central zone and outbound particles travelling in the central zone more frequently than inbound unladen particles (Fig 3A). We also found that in the simulations the average group size was 2.0 and the maximum group size 23, compared to the experimental average group size of 1.9 and maximum group size 14 (S2 Table). In addition, Fig 3B shows that when the turning rule is replaced by a stopping rule, lane segregation does not emerge and we conclude that lane segregation is critically dependent on turning in our model.
Our self-propelled particle model based on the local priority rules presented in [22] generates traffic organization that share several characteristics with the traffic organization observed in the narrow and wide trail experiments. In particular, the narrow trail model reproduces de-synchronization of inbound and outbound traffic and the groups that emerge share several features with the experimentally observed groups (Fig 2). The wide trail model generated segregated traffic that share certain properties with the traffic observed in the experiments (Fig 3). In particular, that the proportion of ants of a given type (L, O or U) traveling in the central zone increased with the priority of each type (L>O>U) in both model and experiment.
This suggests that it is plausible that the priority rules proposed in [22] are key drivers of organizing the traffic on these trails because the main features of the observed traffic emerge from them even when the rest of the system is heavily idealized. For example, we use simplified model ants with constant speed and lengths that always follow the rules and the only stochastic components in the models are related to leaving times and entry positions. In particular, we believe that the strict rule following produces the main discrepancy between simulation results and data, i.e. the over representation of groups of size 1 on the narrow trail (Fig 2A). We also know that introducing various types of stochastic rule violations in the model does not solve this problem, and we are confident that the rule violations are far from randomly occurring and new experiments would be required to investigate this. However, while it would be interesting and potentially useful to conduct new experiments to obtain data that allows us to make certain aspects of the model more realistic this would inevitably make the model more complex and thus make it harder to isolate the effects of generic mechanisms underlying traffic organization.
Understanding the basic principles that govern traffic organization on trails and identifying the factors that influence the movements of ants on trails are of fundamental importance in the biology of social insect colonies. When ants were forced to move on a narrow trail the formation of alternating groups of inbound and outbound ants was observed. Groups of inbound ants were frequently headed by laden ants, which are slower, followed by unladen ants. The model replicates this behavior, most likely due to the rule that dictates that inbound unladen ants do not attempt to overtake laden ants in front of them (Rule 1). This behavior may appear detrimental because unladen ants move slower by staying behind a laden ant instead of progressing more rapidly by moving at its desired speed. However, the model also includes the so-called cooperative rule that allows for the possibility of unladen ants following a laden ant to benefit from the passage of the laden (See rules 2 and 3). These unladen ants avoid head-on collisions with outbound ants and thus spare time they would otherwise waste by stopping as they normally do when they meet an outbound ant. Moreover, this organization promotes information transfer about the level of leaf availability by increasing the number of contacts between outbound and inbound laden ants which stimulate the former to cut and retrieve leaf fragments when reaching the end of the trail [47–53]. Following the same idea, on the wide trail, an intermingled flow of outbound and unladen ants instead of a strict lane segregation might appear sub-optimal, but it actually promotes information transfer between ants and stimulate outbound workers to cut and collect leaf material at the end of the trail, thus contributing to increased foraging efficiency [8, 20, 48, 49].
Our findings suggest that the ants may be using the same generic priority rules on both trails and the observed differences results from constraints imposed by the environment. In particular, on a wider trail there is room to turn during an encounter whereas on a narrow trail this option is not available so the ants have to stop when giving way. In fact, due to the simplicity of these rules we believe that they could be valid, with appropriate modifications, for other species of ant under similar conditions. For example, the same priority rule between laden and unladen ants has been observed in another leaf-cutting ant Atta cephalotes [47] and in the red wood ant Formica rufa [13]. In addition, similar types of priority rules are likely to be operating in army ants because returning laden ants are known to be less mobile and have less maneuverability than unladen outbound ants [46]. We also note that there is a correspondence between the priority rules and the potential utility of each type of ant with respect to leaf collection. Laden ants have the highest priority and they are collecting leaves, outbound ants have the second highest priority and they are potentially going to collect leaves, and inbound unladen ants have the lowest priority and they are not collecting leaves. We speculate that priority rules in other species are likely to correspond to the potential utility of each type of ant with respect to the colony’s foraging activity.
Our model distinguishes itself from earlier spp-models of ant traffic in several ways. In particular, we model three types of ants (outbound, unladen and laden) whereas [19,45] only include two; inbound and outbound, and the two types are essentially identical except for different avoidance turning rates in [19]. Furthermore, the avoidance turning rate is the same for all individuals of a certain type, i.e. outbound or inbound, despite the fact that variation in maneuverability and speeds exist in real ants due to food transport [22, 46]. Our model includes this variability and our priority rules are flexible enough to model traffic on both narrow and wide trails. In [19,45] only traffic on wider trails are modeled and while the avoidance turning rate approach may be modified to work on narrow trails, which presumably both army ants [19] and black garden ants [45] occasionally travel on in the wild, we predict that unless the inbound flow is separated into unladen and laden ants with different behaviors the model will not be able to generate traffic consistent with the real ant traffic [16].
One often thinks about the similarities between ant traffic, pedestrian traffic and vehicular traffic. These analogies have inspired multiple investigations [54–58]. However, even if at first sight traffic on ant trails may appear similar to human traffic there are important differences to consider when comparing their traffic organization. First, ant traffic is of a cooperative nature because all ants share a common objective, namely harvesting food for the colony. Second, ants do not have the same mechanical constraints as pedestrians or vehicles. Because of their small mass they have a low inertia and are not damaged by collisions, allowing a certain degree of mixing of opposite flows on foraging trails. Despite this, ant traffic remains an important source of inspiration for various researchers working with large groups of interacting particles in disciplines as diverse as molecular biology [59], statistical physics [60] and telecommunication sciences [61].
|
10.1371/journal.pntd.0001689 | Sensitive and Specific Detection of Trypanosoma cruzi DNA in Clinical Specimens Using a Multi-Target Real-Time PCR Approach | The laboratory diagnosis of Chagas disease is challenging because the usefulness of different diagnostic tests will depend on the stage of the disease. Serology is the preferred method for patients in the chronic phase, whereas PCR can be successfully used to diagnose acute and congenital cases. Here we present data using a combination of three TaqMan PCR assays to detect T. cruzi DNA in clinical specimens.
Included in the analysis were DNA extracted from 320 EDTA blood specimens, 18 heart tissue specimens, 6 umbilical cord blood specimens, 2 skin tissue specimens and 3 CSF specimens. For the blood specimens both whole blood and buffy coat fraction were analyzed. The specimens were from patients living in the USA, with suspected exposure to T. cruzi through organ transplantation, contact with triatomine bugs or laboratory accidents, and from immunosuppressed patients with suspected Chagas disease reactivation. Real-time PCR was successfully used to diagnose acute and Chagas disease reactivation in 20 patients, including one case of organ-transmitted infection and one congenital case. Analysis of buffy coat fractions of EDTA blood led to faster diagnosis in six of these patients compared to whole blood analysis. The three real-time PCR assays produced identical results for 94% of the specimens. The major reason for discrepant results was variable sensitivity among the assays, but two of the real-time PCR assays also produced four false positive results.
These data strongly indicate that at least two PCR assays with different performances should be combined to increase the accuracy. This evaluation also highlights the benefit of extracting DNA from the blood specimen's buffy coat to increase the sensitivity of PCR analysis.
| Chagas disease is endemic in several Latin American countries and affects approximately 8 to 11 million people. The protozoan parasite, Trypanosoma cruzi, is the agent of Chagas disease, a zoonotic disease that can be transmitted to humans by blood-sucking triatomine bugs. Other routes of infection include congenital transmission, blood transfusion, organ transplantation, accidental inoculation of the parasite during laboratory research and by consuming food and juice contaminated with the parasite. This study focused on the evaluation of three quantitative PCR (QPCR) assays for the diagnosis of Chagas disease. The evaluation was based on the analysis of 349 specimens submitted for confirmatory diagnosis of Chagas disease to the Centers for Disease Control and Prevention from 2008 to 2010. By using such assays we were able to diagnose acute and Chagas disease reactivation in 20 patients, including one case of organ-transmitted infection and one congenital case. The paper also highlights the benefit of extracting DNA from the blood specimen's buffy coat to increase the sensitivity of diagnostic PCR analysis. The results obtained in this study strongly indicate that at least two QPCR assays with different performance characteristics should be combined to increase diagnostic accuracy.
| Chagas disease is a vector-borne infectious disease caused by the parasite Trypanosoma cruzi. It is endemic in several countries of Central and South America. In endemic areas the disease is spread by certain species of triatomine bugs that excrete the parasites in their feces while feeding on human hosts. Humans get infected when feces from infected triatomines contaminates wounds, allowing the parasite to enter the bloodstream. Other routes of infection include congenital transmission, blood transfusion, organ transplantation, accidental inoculation of the parasite during laboratory research and by consuming food and juice contaminated with the parasite. As efforts to control vector-borne and blood transmission are successful, congenital and oral transmission paths are becoming increasingly important [1].
After a short acute phase when the parasite can be found circulating in the blood, the disease enters the chronic phase when the amastigote stage develops and multiplies in organ tissues, primarily in the heart. The chronic phase is characterized by two forms; patients first develop the indeterminate form of chronic infection which can last for decades and the patients are typically asymptomatic during this time. An estimated 30–40% of patients may develop clinical disease, with manifestations such as cardiomyopathy or digestive megasyndromes [1]. Chronically infected patients that become immunosuppressed may experience a reactivation of the disease, a condition characterized by increasing parasitemia and atypical presentations such as epidermic lesions and compromise of the central nervous system [2].
The options for laboratory diagnosis of Chagas disease depend on the disease phase. Serology is the method of choice to diagnose chronic infections. Acute infections can be diagnosed by detecting motile organisms in fresh blood preparations, by culture or by detection of parasite DNA by PCR [3]. The latter methods are also recommended to detect increasing parasitemia in cases of reactivation following immunosuppression [4]. In cases of acute infections or reactivation of chronic disease it is important to use sensitive diagnostic methods since early detection and treatment results in a more favorable outcome. PCR-based methods are generally considered to be more sensitive than microscopy and have lately been increasingly used to diagnose Chagas disease [4]. However, the use of PCR is also challenging as there is no “gold standard” method for the diagnosis of Chagas disease [5], [6] and the diagnostic performance can vary widely depending on the type of PCR assay. The most widely used PCR assays used for diagnostic purposes target either the kinetoplast genome (kDNA), also called the minicircle, or a nuclear mini-satellite region designated TCZ [4], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16]. Both of these targets are present in multiple copies in the parasite genome, which increases the sensitivity of detection [15], [17]. However, assays that target these regions have been reported to cross-amplify non-T. cruzi DNA [10], [16], [18], [19], [20]. Assays that amplify other genes may show better specificity but they are generally less sensitive [4], [9], [16], [21].
One important use of PCR as a diagnostic tool is to provide a sensitive method to detect reactivation in chronically infected patients with immunosuppression. Patients with chronic Chagas heart disease often require a heart transplant [22]. Current recommendations state that these patients should be monitored at regular intervals after the transplant for signs of increasing parasitemia [3]. Another category of patients for whom PCR testing is beneficial is patients who receive organs from chronically infected donors. Since only a fraction of organ recipients will develop an acute T. cruzi infection, preventive drug treatment is not recommended. In such cases the use of PCR can allow for early detection of those cases where transmission has occurred.
Recently, an international collaborative study focusing on standardization and validation of PCR for diagnostic detection of T. cruzi DNA was conducted [13]. The study relied on the use of DNA specimens from genetically distinct cultured T. cruzi strains plus blood specimens from chronically infected patients. The specimens were coded at a coordinating laboratory and shipped to 26 participating laboratories that performed PCR testing according to their own standard operating procedures. Results were then sent back to the coordinating laboratory and performance characteristics were calculated for each PCR assay. The study found a high degree of variability in accuracy and performance among the included PCR tests and identified and further evaluated two DNA extraction methods and four PCR assays that performed better than the others. Two of the best-performing assays were real-time PCR assays.
To continue these efforts we here present results from a diagnostic testing algorithm involving three of the real-time PCR assays included in the international validation study mentioned above. Real-time PCR has several advantages over conventional PCR, e.g. shorter turnaround times and less risk of amplicon carry-over contamination [23], both of which can be advantageous in diagnostic laboratories. One of the real-time PCR assays included in this study was ranked among the four best-performing assays in the international validation study; a real-time PCR assay targeting the mini-satellite TCZ region. The second real-time PCR assay was selected because it was the best-performing real-time PCR assay targeting the kDNA included in the international validation study. The third real-time PCR assay was included in this study because it targets the small subunit ribosomal RNA (18 S rRNA) gene, which is generally suitable for diagnostic assays because it is highly conserved.
In contrast to the international validation study we mainly used specimens from patients with suspected acute or reactivating Chagas disease since PCR testing is more relevant for early diagnosis or monitoring in this patient group than in chronic patients, whose diagnosis relies on serological methods. The majority of the specimens tested were EDTA blood samples; we performed real-time PCR on DNA extracted from buffy coat preparations in addition to whole blood to determine the effect of buffy coat concentration on the sensitivity of the PCR analysis.
All the specimens used in this study were submitted to CDC for confirmatory diagnosis of Chagas disease during years 2008–2010 from state public health laboratories, hospitals and private clinics in the United States. The tests were performed on 349 laboratory specimens from 119 patients, who lived in the United States at the time of specimen collection. A breakdown of the specimen types and the conditions that prompted the diagnostic requests are presented in Table 1. Samples analyzed in this study were anonymized by removing identifiers after diagnostic results were reported, in accordance with the CDC IRB, protocol number 3580, entitled “Use of Human Specimens for Laboratory Methods Research”. All of the patients included in this study were evaluated for serology status using the Chagatest recombinante v. 3.0 (Wiener Laboratorios, Rosario, Argentina) and a CDC in-house IIF test.
DNA extraction was performed from all specimens within 24 hours of arrival at the laboratory. DNA was extracted from whole blood specimens using the QIAamp blood mini DNA kit (QIAGEN, Valencia, Calif.). The volume of whole blood used was 0.2 ml and if the remaining volume exceeded 1 ml, the buffy coat fraction was separated as follows: up to 2 ml of whole blood was centrifuged at 2,500×g for 10 minutes. The plasma was removed and the buffy coat layer plus some of the erythrocyte pellet were transferred to a clean tube. DNA was then extracted from that material in parallel with the whole blood aliquot using the same method mentioned above. Three EDTA blood specimens had enough volume left after initial DNA extraction to allow for one or more additional buffy coat preparations. Two-ml aliquots of these specimens were stored at 4°C for one, two or four weeks and then processed as described above. DNA from tissue specimens was extracted with the DNeasy blood and tissue DNA kit (QIAGEN). For cerebrospinal fluids (CSF), approximately half of the total volume received (0.5–1 ml) was centrifuged for 5 minutes at 6000×g. Most of the supernatant was carefully removed until 0.2 ml remained and DNA was extracted from this remaining volume (plus any pellet) with the DNeasy blood and tissue DNA kit (QIAGEN). All the DNA extraction procedures were performed following manufacturer's instructions for the different types of samples. Previous experiences with these methods in our laboratory had ensured that they efficiently removed potential PCR inhibitors from the specimen types included in this study (data not shown). One negative extraction control was included in each batch of DNA extractions to monitor for potential cross- contamination among samples and contamination of kit reagents.
All three real-time PCR assays were included in the international validation study [13]. Table 2 summarizes validation data for the PCR assays as presented in that study, plus specificity data for two other Trypanosoma spp. obtained in our laboratory. The real-time PCR assays were performed and analyzed in an Mx3000P QPCR system (Agilent Technologies, Calif.). Each DNA sample was added to the PCR mix in two different concentrations (corresponding to 5 µl and 1 µl of undiluted DNA). All PCR runs included two or more negative amplification controls (adding water instead of template DNA) plus two positive amplification controls (DNA extracted from a culture of the Y strain in two different dilutions). The risk of false positive results due to contamination was minimized by the following procedures: using separate rooms for DNA extraction, pre-and post-amplification processes; having a uni-directional workflow; and using enzymatic removal of contaminating amplicons before real-time PCR amplification.
TCZ TaqMan real-time PCR (designated as method LbF1 in the international validation study [13]): This TaqMan assay was performed as described in Piron 2007 [11], except that the Platinum qPCR supermix was used instead of the Universal mastermix from Applied Biosystems.
kDNA TaqMan real-time PCR (designated as method LbG/3 in the international validation study [13]): The reaction mix consisted of 1× Platinum qPCR supermix, 0.4 µM of each PCR primer 32F, 5′-TTT GGG AGG GGC GTT CA-3′, and 148R, 5′-ATA TTA CAC CAA CCC CAA TCG AA-3′, plus 0.1 µM of the LNA TaqMan probe 71P, 5′-CA TCTC AC CCG TACA TT-3′, where the LNA nucleotides [24] are underlined. Total reaction volume was 20 µl. Thermocycling structure was as follows: 2 minute incubation at 50°C to activate UDG degradation, 2 minute incubation at 95°C to activate the hot-start DNA polymerase, and 40 cycles of 95°C for 15 seconds and 58°C for 60 seconds.
18 S rRNA TaqMan real-time PCR (designated as method LbS/4 in the international validation study [13]): The reaction mix consisted of 1× Platinum qPCR supermix, 0.2 µM of each PCR primer TcF1042, 3′-GCA CTC GTC GCC TTT GTG-3′, and TcR1144, 5′-AGT TGA GGG AAG GCA TGA CA-3′ plus 0.05 µM of the TaqMan probe TCP1104, 5′-AA GAC CGA AGT CTG CCA ACA ACA C-3′. Total reaction volume was 20 µl. Thermocycling structure was as follows: 2 minute incubation at 50°C to activate UDG degradation, 2 minute incubation at 95°C to activate the hot-start DNA polymerase, and 40 cycles of 95°C for 15 seconds and 60°C for 60 seconds.
This study focused on diagnostic specimens tested from 2008 to 2010. Fifty of the 349 samples produced positive results in at least one of the real-time PCR assays. The three real-time PCR assays produced identical results for 329 samples (94%), of which 30 were PCR positive, while the remaining 20 samples had discrepant results in the three assays.
Of the 50 samples with at least one positive PCR result, 37 samples were collected from 13 chronically infected patients with reactivation disease, six samples from three transplant recipients of organs from chronically infected donors, four samples from a patient with acute Chagas disease acquired during travel in an endemic region, one sample from a congenitally transmitted infection and two samples from two patients under evaluation for severe cardiomyopathy. Tables 3, 4, and 5 list the detailed real-time PCR results from specimens collected from a selection of these patients, as outlined below.
Specimens from 25 patients with chronic Chagas disease (as determined by positive serology) were received for evaluation of reactivation disease during the study period. Eighteen of the patients had received a heart transplant, five were HIV infected and two had undergone a bone marrow transplant. Sixteen patients had one or more PCR-positive samples, including sporadic PCR-positive results in seven patients who had received a heart transplant prior to 2008. Table 3 lists a selection of the samples analyzed from the remaining nine patients with at least one PCR-positive test result during this study. Seven patients were tested for reactivation following transplants; four of these were monitored on a regular basis by PCR. Two of the five HIV-positive patients were diagnosed with re-activated Chagas disease (patients 8 and 9); one of them had cerebral Chagas, confirmed by the presence of T. cruzi DNA in CSF.
Real-time PCR was used to test blood specimens from 14 previously non-infected transplant patients who received organs from a donor with suspected or confirmed chronic Chagas disease. It is recommended to closely monitor these patients with PCR or other sensitive technique in order to detect potential transmission as soon as possible. Three organ recipients had one or more PCR positive results (see Table 4). However, only patient 12, a heart recipient, was actually infected with T. cruzi. The PCR positive results for the other two patients (patients 10 and 11) were reported as equivocal and were most likely false positive results because of the following circumstances. Patient 10 received a kidney from a donor with borderline positive serology results with the Ortho T. cruzi ELISA test (Ortho-Clinical Diagnostics, Raritan, New Jersey). Since this could have been interpreted as indicative of infection in the donor, regular PCR testing was started on patient 10. However, subsequent serology testing of the donor associated with this case could not confirm the preliminary results; i.e., the T. cruzi RIPA was indeterminate and both the Wiener and the IIF test were negative on repeated serum samples. It was therefore concluded that the donor was not infected with T. cruzi and additional PCR follow up of patient 10 was unnecessary. However, before the final donor serology status had been determined, weak positive signals in the kDNA and TCZ TaqMan assays were verified in blood samples from patient 10. Unexpectedly, each subsequent specimen obtained from this patient showed a signal that was weaker than the signal obtained for the previous sample; i.e. the opposite of what was expected from an acute T. cruzi infection in an immunocompromised patient. At six weeks post-transplant patient 10 was no longer positive in any of the real-time PCR assays. Patient 11 received a kidney from a donor that was confirmed to be serologically positive for T. cruzi. The blood sample from patient 11 collected on the 3rd week post-transplant tested weakly positive in the kDNA and TCZ TaqMan assays, with only the whole blood aliquot being positive and not the buffy coat fraction. The blood collected a week later was PCR negative in both whole blood and buffy coat. Neither patient 10 nor 11 had any clinical signs of T. cruzi infection. Their blood smears were constantly negative for parasites and they did not receive anti-trypanosomal drugs.
We received 48 blood samples from 13 healthy patients who had been bitten or in close contact with triatomine bugs plus 9 laboratory workers that had been accidentally exposed to T. cruzi via needle stick accidents or animal bites during research activities. None of these were PCR positive. We tested 22 specimens from 20 children (aged newborn to 8 years) with sero-positive mothers for possible congenital transmission and detected T. cruzi DNA in the blood of a 19-days-old infant (patient 14 in Table 5). Twenty-eight specimens were received from 18 adult patients with symptoms of acute T. cruzi infection (fever and malaise after traveling to endemic region and/or having close contact with triatomine bug; three had a swollen eye that could be chagoma). Only one of these patients tested positive for T. cruzi by PCR and was treated for acute Chagas disease (patient 13 in Table 5). Follow-up specimens from this patient again tested positive in PCR after completed drug treatment but unfortunately the patient was lost to follow-up.
Thirty-five of the PCR-positive blood specimens (from 16 patients) had enough volume to allow for DNA extraction from both whole blood aliquots and buffy coat fraction. Of these, 26 specimens (from 10 patients) had PCR-detectable levels of T. cruzi DNA in both whole blood and buffy coat, with a relatively higher concentration in the buffy coat based on the quantitative output (the Cq value) from the real-time PCR assays. The remaining 9 specimens (from 6 patients) were positive only in the buffy coat fraction. Thus, 26% of the PCR-positive specimens would have been reported as being negative for T. cruzi if no buffy coat analysis had been performed. For three patients the analysis of buffy coat was crucial: Chagas disease reactivation in two patients was detected two weeks earlier by testing the buffy coat sample as compared to whole blood (patients 3 and 6 in Table 3) and the patient who acquired Chagas disease through transplantation (patient 12 in Table 3) was identified as positive one week earlier by testing buffy coat as compared to whole blood.
Three of the PCR-positive blood samples had enough volume to allow for analysis of more than one buffy coat preparation. Aliquots of these three samples were stored at 4°C for up to four weeks and then processed as described. Figure 1 depicts the quantitative real-time PCR results obtained from these samples over time. The results suggested that storage of EDTA blood for a limited time had minor effect on the quality of T. cruzi DNA obtained from buffy coat preparations, at least for the kDNA and TCZ genetic regions. Although these are only preliminary data that need confirmation with a larger set of samples, it removes some of the uncertainty whether to accept EDTA-blood samples that for various reasons are delayed in transport to the diagnostic laboratory.
The laboratory diagnosis of Chagas disease relies mainly on serology, microscopic identification of trypomastigotes in blood or buffy coat, hemoculture and PCR [3]. Several PCR assays with variable diagnostic sensitivity and specificity have been developed and used as diagnostic tests [4], [7], [8], [9], [10], [11], [15], [16], [21]. A complicating factor for PCR assays is the high genetic variability of T. cruzi strains; there are currently six genotype groups or discrete typing units (DTU) described that differ significantly in genetic content and gene copy numbers [25], [26]. Since some DTUs are more common than others in various endemic regions, the same PCR assay can perform differently depending on the geographic origin of the specimen [16], [25], [27], [28], [29]. One way to circumvent these accuracy problems is to combine two or more PCR assays that target different genes.
The reference diagnostic laboratory at CDC employs a multi-target PCR testing algorithm consisting of three real-time PCR assays that are performed in parallel on all specimens. The three assays target different genomic regions in T. cruzi and have therefore variable sensitivity and specificity. The rationale for including all three assays in the testing algorithm is to ensure the highest accuracy possible by combining assays that complement each other. The kDNA TaqMan assay seems to be the most sensitive assay but it can amplify non-T. cruzi DNA, e.g. T. rangeli, and thus lead to false positives. The TCZ TaqMan assay has better specificity but as shown in this study can produce false positive PCR results as well. The kDNA and TCZ TaqMan assays are both much more sensitive than the 18 S rRNA TaqMan but the main advantage of including the 18 S rRNA assay in the testing algorithm is that it seems to be 100% specific. According to the CDC protocol, if a specimen tests positive in all three real-time PCR assays it will be reported as positive for T. cruzi, but any specimen that is only positive in the kDNA and/or TCZ TaqMan assays and negative in the 18 S rRNA TaqMan assay require additional confirmation by other tests or clinical data in order to be reported as positive for T. cruzi. If confirmatory data is absent or does not support a diagnosis of T. cruzi infection, the PCR results are reported as equivocal and a new specimen is requested to repeat the molecular analysis.
Diagnostic sensitivity can be enhanced by maximizing the amount of target DNA in the aliquot used for DNA extraction. For multi-copy PCR targets this can be obtained by mixing blood specimens with guanidine HCl-EDTA solution that lyses the parasites and releases their genetic content, thus making it possible to detect as little as one parasite in a large volume of blood [30], [31]. It has also been reported that sensitivity could be enhanced if blood clot was used as starting material [32]. An alternative method is to concentrate the parasites in the buffy coat fraction [33] prior to DNA extraction; this has been reported to increase the sensitivity compared to analysis of frozen EDTA-blood and guanidine HCl-EDTA treated blood [32], [34]. During this study, we compared the PCR results obtained from buffy coat concentration with results from fresh EDTA- blood and found that analysis of buffy coat allowed earlier detection of increasing levels of circulating parasite genome in three cases: two reactivation cases and one organ-transmitted acute infection. Thus, appropriate drug treatment for these patients could be initiated 1–2 weeks sooner.
Analyzing both the buffy coat fraction and a whole blood aliquot in parallel can also give helpful information to ensure test validity and to troubleshoot suspicious false positive PCR results. DNA extraction from the buffy coat fraction of a blood sample containing T. cruzi trypomastigotes should produce more T. cruzi DNA than the corresponding volume of whole blood. If that is not the case, there could be a problem with the quality of the blood specimen, the DNA extraction process or the PCR accuracy. One of the false positive PCR results obtained in this study was immediately flagged as suspicious because only the whole blood fraction was positive while buffy coat was negative. Nevertheless, more data must be accumulated during a longer period of time for a more robust assessment about the advantages of analyzing both whole blood and buffy coat.
In conclusion, we propose that in reference laboratories with the adequate infrastructure, the use of two or more real-time PCR tests with different performance characteristics combined with the analysis of buffy coat and whole blood can strengthen the use of PCR for accurate diagnosis of Chagas disease.
|
10.1371/journal.pntd.0002212 | Post-Exposure Therapeutic Efficacy of COX-2 Inhibition against Burkholderia pseudomallei | Burkholderia pseudomallei is a Gram-negative, facultative intracellular bacillus and the etiologic agent of melioidosis, a severe disease in Southeast Asia and Northern Australia. Like other multidrug-resistant pathogens, the inherent antibiotic resistance of B. pseudomallei impedes treatment and highlights the need for alternative therapeutic strategies that can circumvent antimicrobial resistance mechanisms. In this work, we demonstrate that host prostaglandin E2 (PGE2) production plays a regulatory role in the pathogenesis of B. pseudomallei. PGE2 promotes B. pseudomallei intracellular survival within macrophages and bacterial virulence in a mouse model of pneumonic melioidosis. PGE2-mediated immunosuppression of macrophage bactericidal effector functions is associated with increased arginase 2 (Arg2) expression and decreased nitric oxide (NO) production. Treatment with a commercially-available COX-2 inhibitor suppresses the growth of B. pseudomallei in macrophages and affords significant protection against rapidly lethal pneumonic melioidosis when administered post-exposure to B. pseudomallei-infected mice. COX-2 inhibition may represent a novel immunotherapeutic strategy to control infection with B. pseudomallei and other intracellular pathogens.
| Burkholderia pseudomallei is the etiologic agent of melioidosis, a severe disease endemic in Southeast Asia and Northern Australia. B. pseudomallei is also classified as a Tier 1 select agent due to the threat of malicious use of the organism. Treatment of melioidosis is complicated by the inherent multidrug resistance of B. pseudomallei, leading to high case fatality rates or disease relapse. New therapeutic strategies are urgently needed to improve patient survival and to protect against a deliberate release of B. pseudomallei. Immunotherapeutics that can enhance the host immune response and delay disease progression represent a significant area of research interest. A number of immunomodulatory agents delivered locally to the lung prior to B. pseudomallei infection have afforded significant protection against pulmonary disease in animal models of melioidosis; however, their protective capacity significantly wanes upon post-exposure administration. In this work, we identify the PGE2 pathway as an immunotherapeutic target in pulmonary melioidosis and show that post-exposure COX-2 inhibition provides significant protection against lethal B. pseudomallei lung infection in mice. Further research examining FDA-approved COX-2 inhibitors as post-exposure prophylaxis for B. pseudomallei is warranted, as this may represent a safe, affordable, and efficacious immunotherapeutic strategy.
| Development of new therapeutics effective against intracellular bacterial pathogens remains a high priority. In addition to the global impact of intracellular bacterial infections on public health, the alarming increase in multidrug resistant strains and the potential threat of biological attack with select agents, such as Burkholderia pseudomallei, highlight the urgent need for safe and effective therapies against this collective group of pathogens. B. pseudomallei is a Gram-negative, facultative intracellular bacillus and the causative agent of melioidosis, a disease associated with high morbidity and mortality in Southeast Asia and Northern Australia. Although melioidosis is not endemic in the United States, B. pseudomallei is classified as a Tier 1 select agent due to its ease of respiratory transmission, high mortality rate, multidrug resistance, and the absence of a protective vaccine [1]. Furthermore, malicious use of B. pseudomallei and B. mallei during World Wars I and II provides historical precedence for use of these agents as bioweapons and validates the need for post-exposure therapeutics that can be quickly administered to military personnel and civilians [2].
The inherent antibiotic resistance of B. pseudomallei limits chemotherapeutic options for melioidosis and the particular choice of antibiotic regimen has not been shown to impact mortality within the first 48 hours of hospitalization [3]. Current treatment requires intravenous administration of ceftazidime or meropenem, with or without trimethoprim-sulphamethoxazole (TMP-SMX), for two weeks of intensive phase therapy. The intensive phase of treatment may be extended up to eight weeks for deep-seated infections. Upon completion of this intensive phase, an eradication phase utilizing oral TMP-SMX or doxycycline for outpatient use is recommended for 8–12 weeks in order to prevent relapse. Despite this aggressive therapy, case fatality rates for severe melioidosis approach 40% in Thailand and 15% in Australia [4]. Therefore, it is necessary to develop new modalities of treatment that can replace or complement existing antibiotics to improve patient survival.
An appealing alternative as a first line therapeutic strategy is to enhance the host innate immune response during the early course of bacterial infection. In human trials, complementary use of granulocyte colony-stimulating factor improved the duration of survival for melioidosis patients with severe sepsis but did not decrease mortality rates [5]. In pre-clinical studies, treatment of BALB/c mice with cationic liposomal DNA complexes (CLDC) 24 h prior to intranasal B. pseudomallei challenge enhanced natural killer (NK) cell recruitment and afforded complete protection from a lethal infectious dose [6]. Similarly, treatment of BALB/c mice with the TLR9 agonist, CPG ODN, 48 h prior to B. pseudomallei infection led to significantly lower tissue bacterial burdens and improved overall survival [7], [8]. Combining vaccination with CpG treatment that was given up to 18 h post-infection provided significantly greater protection against B. pseudomallei than either treatment alone, indicating that immune modulation with CpG can also enhance the efficacy of other countermeasures [9]. In contrast, post-exposure prophylaxis with CpG alone was not effective against B. pseudomallei, as initial control of bacterial growth appears dependent upon prior recruitment of inflammatory cells to the lung [8]. Since bacterial infection cannot be predicted, it is imperative to identify immunotherapeutics that can mediate protection when administered post-exposure.
In the present study, we identify the prostaglandin E2 (PGE2) pathway as a novel therapeutic target during pneumonic melioidosis. PGE2 is a potent lipid mediator derived from cyclooxygenase (COX) metabolism of the cell membrane fatty acid, arachidonic acid [10]. PGE2 is produced in response to inflammation via the COX-2 enzyme and is a key mediator of immunopathology in chronic disease, autoimmunity, and cancer [10]. While PGE2-mediated immunoregulation is essential for maintaining homeostasis, its suppressive effects on innate and adaptive immunity may be counter-productive during infection. In this work, we demonstrate that B. pseudomallei rapidly induces macrophage COX-2 expression and PGE2 production which establishes a permissive environment for B. pseudomallei intracellular persistence. Pulmonary infection with B. pseudomallei leads to increased concentrations of lung PGE2, and lung PGE2 levels significantly correlate with disease progression in mice. Post-exposure administration of a COX-2 inhibitor provides significant protection against lethal pulmonary challenge with B. pseudomallei. This is the first demonstration of a non-antibiotic post-exposure therapeutic that provides significant protection on its own against lethal pulmonary infection with B. pseudomallei. Therapeutic strategies targeting the PGE2 pathway may delay disease progression in pneumonic melioidosis and afford a window of opportunity for antibiotic intervention and/or development of adaptive immunity. Furthermore, COX-2 inhibition may represent a novel and universal immunotherapeutic strategy against other intracellular pathogens.
Ethics Statement: Animal experiments were performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Tulane University Institutional Animal Care and Use Committee (protocol number 4042). Six to eight week old, female BALB/c mice (Charles River) were maintained under pathogen-free conditions and fed sterile food and water ad libitum. Infections utilizing Bps were performed under Animal Biosafety Level 3 containment.
Burkholderia pseudomallei strain 1026b (BEI Resources) was used in this study. For infectious challenge, mice were anesthetized with Ketamine/xylazine (88 mg/kg) (Fort Dodge Animal Health). The bacterial inoculum contained 3×103 cfu (∼4 LD50) suspended in 40 µl sterile saline and 20 µL was delivered to each nostril via pipet. Bacterial cfu were confirmed by plating the inoculum on LB agar. Euthanasia endpoints used in this study included loss of >20% body weight, hunched posture and decreased movement or response to stimuli, or paralysis. In a subset of experiments, mice were treated with the selective COX-2 inhibitor, (N-[2-(cyclohexyloxy)-4-nitrophenyl]-methanesulfonamide) (NS398) (Cayman Chemicals), 3 h post-exposure. Mice received 50 µl of NS398 (15 mg/kg) dissolved in DMSO or vehicle control (DMSO) by intraperitoneal injection. Treatments were repeated for two consecutive days. After euthanasia, tissues were removed, weighed and homogenized in 1 ml 0.9% sterile saline. Serial dilutions of tissue homogenates were plated on LB agar and bacterial cfu were counted after 2–4 days of incubation at 37°C.
J774A.1 murine macrophage-like cells were obtained from ATCC. Cells were propagated in media containing DMEM (Invitrogen) with 10% FBS (Atlanta Biologicals), 1% Pen/Strep (Invitrogen) and 1% sodium bicarbonate (Invitrogen). Bone marrow-derived macrophages (BMDM) were extracted from 8–10 week old BALB/c mice as previously described [11]. BMDM were propagated in RPMI (ATCC), containing 15% L929 fibroblast-conditioned media, 2 g/L D-glucose (Invitrogen), 10% FBS, 5% horse serum (Invitrogen), 1% Pen/Strep and 2 mM L-glutamine (Invitrogen). Prior to each experiment, the cytotoxic dose of bacteria and chemical treatments were pre-determined using a colorimetric assay for LDH release (Clontech). Intracellular survival assays were performed as previously described [12]. In some experiments, cells were treated with 100 µM of NS398, 100 µM nor- Nω-hydroxy-L-arginine (nor-NOHA; Cayman Chemical), or 1 µM PGE2 (Sigma). PGE2 was measured in cell culture supernatants and lung homogenates by competitive ELISA (Pierce). Nitric oxide was measured as its stable end product nitrite by Griess assay (Invitrogen).
Fold-change in mRNA expression of 84 genes central to TLR-mediated signal transduction and innate immunity were measured by PCR array following the manufacturer's protocol and data analysis software (SABiosciences).
RT-PCR was conducted using an iCycler (BioRad) with iScript cDNA Synthesis Kit (BioRad). 1 µg of RNA was converted to cDNA following the manufacturer's protocol. 1 µl of cDNA was added to 12.5 µL of iQ SYBR Green Super Mix containing 350 nM of each forward and reverse primer. Primer sequences were as follows: GAPDH: forward, 5′-ACAGCCGCATCTTCTTGTGCAGTG-3′; reverse, 5′-GGCCTTGACTGTGCCGTTGAATTT-3′; Arg1: forward, 5′-GGGCTGGACCCAGCATTCACCCCG-3′; reverse, 5′TCACTTAGGTGGTTTAAGGTAGTC-3′; Arg2: forward, 5′-GACCCTAAACTGGCTCCAGCCACA-3′; reverse, 5′-CTAAATTCTCACACATTCTTCATT-3′; iNOS: forward, 5′-ATGACCAGTATAAGGCAAGC-3′; reverse, 5′-GCTCTGGATGAGCCTATATTG-3′; COX-2: forward, 5′-GGAGAGAAGGAAATGGCTGCA-3′; reverse, 5′-ATCTAGTCTGGAGTGGGAGG-3′. Nuclease-free water was added to bring the total reaction volume to 25 µL. PCR was performed using the following conditions: reverse transcriptase inactivation (95°C, 3 min) followed by 40 PCR cycles (95°C, 15 seconds and 60°C, 30 seconds) followed by melt curve analysis. Fold change (up- or down- regulation) relative to base line expression in uninfected cells was calculated using the ΔΔCt method using Ct values for arginase 1, arginase 2, iNOS, COX-2, and GAPDH.
For Western blot, equal amounts of protein (50 µg) from cell lysates or lung homogenates were resolved by SDS-PAGE and transferred to nitrocellulose using an iBLOT (Invitrogen). Detection of COX-2 enzyme was performed using a 1∶1000 dilution of rabbit polyclonal anti-COX-2 (Cell Signaling Technology), followed by peroxidase-conjugated, donkey anti-rabbit IgG (1∶5000). Arg2 was detected using a 1∶200 dilution of rabbit polyclonal anti-mouse Arg2 (sc-20151, Santa Cruz) and Arg1 was detected using a 1∶500 dilution of Arg1 antibody 18351 (Santa Cruz). Mouse Arg2-transfected 293T cell lysate (Santa Cruz) was used a positive control for Arg2 and 5 µg of mouse liver extract was used a positive control for Arg1. β-actin, a protein loading control, was detected using 1∶1000 dilution of polyclonal rabbit anti-mouse β-actin (Cell Signaling). Detection of bound antibodies was visualized by a chromogenic reaction using Opti-4CN Substrate (BioRad).
Statistical analyses were performed using Prism 5.0 software (Graph Pad). Kaplan–Meier survival curves were compared by log-rank analysis. All other data were analyzed using a one-way or two-way ANOVA followed by the Bonferroni post-test to determine statistical differences between groups. p<0.05 was considered statistically significant. All data are representative of at least two independent experiments.
B. pseudomallei is remarkable in its ability to establish chronic infection that can reactivate decades after the initial infection and yet virtually nothing is known regarding the mechanisms by which B. pseudomallei evades immune clearance [1]. In order to identify host cell signaling pathways that might contribute to B. pseudomallei intracellular persistence, we performed a Toll-like receptor (TLR) PCR array on J744A.1 macrophages infected with B. thailandensis. B. thailandensis is a commonly used biosafety level 2 surrogate organism for the study of B. pseudomallei and, with the exception of capsular polysaccharide, possesses all of the known B. pseudomallei virulence determinants such as Type 3 and Type 6 secretion systems [13]–[17]. Although B. thailandensis is 1,000- to 100,000-fold less virulent than B. pseudomallei in animal models, the organisms behave very similarly in vitro. B. thailandensis and B. pseudomallei induce pyroptosis in macrophages as early as 8 h post infection at a multiplicity of infection (MOI) 10 or greater [18]. In pilot experiments, we determined that J774A.1 macrophages infected with B. thailandensis at MOI 10 or 1 displayed 80% and 28% cytotoxicity, respectively at 8 h post-infection (not shown). Therefore, experiments utilizing J774A.1 macrophages or primary bone marrow-derived macrophages (BMDM) were limited to an eight hour experimental time course using B. thailandensis or B. pseudomallei at MOI 1 or lower (0.1).
Consistent with previous reports [19]–[22], B. thailandensis upregulated expression of TLR1 and TLR2 by two h post-infection, and increases in TLR1, TLR2, TLR3, TLR4, and TLR5 mRNA expression were observed by eight h post-infection (Supporting information, Table S1). No change in mRNA expression was observed for TLR6, 7, 8, or 9. One of the most striking changes in expression occurred in COX-2, the enzyme responsible for the production of PGE2. A rapid increase (430-fold) in COX-2 mRNA expression occurred by two h post-infection and further increased by >16,000-fold at eight h (Supplementary Table S1).
To confirm the TLR array results obtained for B. thailandensis-infected J774A.1 macrophages, BMDM were infected with B. thailandensis and B. pseudomallei (MOI 1) and COX-2 mRNA expression was measured by RT-PCR. B. thailandensis and B. pseudomallei both up-regulated COX-2 mRNA expression in BMDM to a similar extent (Fig. 1A), although the levels of mRNA expression were lower than that observed for B. thailandensis-infected J774A.1 cells and may reflect differences between immortalized and primary cell lines (Fig. 1A, Supplementary Table S1). COX-2 enzyme and its end product, PGE2, were also produced by macrophages in response to B. pseudomallei in a time- and dose-dependent manner (Fig. 1B–C). Since lipopolysaccharide (LPS) of Gram-negative bacteria is known to induce COX-2 and PGE2 production, we evaluated whether the PGE2 response of infected macrophages was simply a passive signaling event mediated by TLR4 recognition of LPS. Notably, heat inactivation of B. pseudomallei significantly abolished COX-2 and PGE2 expression (Fig. 1A–C) indicating that viable bacteria and/or bacterial proteins are required for early PGE2 production by macrophages.
Because PGE2 has been shown to suppress macrophage bactericidal mechanisms [23], we assessed the impact of COX-2 activation and PGE2 production on B. pseudomallei intracellular survival using the selective COX-2 inhibitor, NS398. Preliminary dose-response experiments were conducted using 10 to 200 µM NS398 (not shown). BMDM treated with ≥100 µM NS398 demonstrated enhanced intracellular killing of B. pseudomallei compared to non-treated cells by six h post-infection (Fig. 2A). To verify the specificity of NS398 and that endogenous PGE2 is responsible for the suppression of bacterial killing, exogenous PGE2 was added to NS398-treated cells. Addition of PGE2 to the cell cultures restored B. pseudomallei intracellular survival (Fig. 2A) confirming that PGE2 promotes a favorable environment for B. pseudomallei.
Previous work has shown that macrophage bactericidal activity against B. pseudomallei is mediated to a large extent by reactive nitrogen species and to a lesser extent by reactive oxygen species (ROS) [24], [25]. PGE2 has been shown to suppress nitric oxide (NO) synthesis in Kupffer cells, hepatocytes, murine peritoneal macrophages, and RAW 264.7 murine macrophages [26]. Therefore, we evaluated the downstream effect of PGE2 on the macrophage NO response to B. pseudomallei infection. Treatment of BMDM with the COX-2 inhibitor NS398 led to a significant increase in nitrite, the stable end product of NO (Fig. 2B). This effect was not drug-specific because similar results were obtained using the COX inhibitor, indomethacin (not shown). Conversely, the addition of exogenous PGE2 to NS398-treated macrophages significantly reduced nitrite levels in B. pseudomallei-infected cells (Fig. 2B). This suggests that PGE2-mediated suppression of NO production may partially contribute to B. pseudomallei intracellular survival.
We next examined the effect of endogenous PGE2 production on the expression of iNOS, which is required for the synthesis of NO. We did not observe any significant change in iNOS mRNA expression in NS398- or PGE2-treated cells compared to controls infected with B. pseudomallei (Fig. 3A). This suggested that PGE2 did not directly regulate iNOS in B. pseudomallei-infected cells and that other mechanisms were responsible for the reduced levels of NO.
Since the enzymes arginase 1 (Arg1) and 2 (Arg2) compete with iNOS for the substrate, L-arginine, we postulated that PGE2 induction of arginase could alter the level of NO production during B. pseudomallei infection. PGE2 induction of macrophage arginase promotes tumor cell growth by suppressing NO-mediated tumor cytotoxicity [27], [28]. Arg1 expression was not detected after four h of B. pseudomallei-infection, but the expression of Arg2 was significantly increased (155-fold) in B. pseudomallei-infected BMDM (Fig. 3A). NS398-treated macrophages demonstrated a significant reduction in Arg2 expression while treatment with exogenous PGE2 increased Arg2 expression by 376-fold (Fig. 3A). These data suggest that endogenous PGE2 may interfere with NO production by enhancing Arg2 expression.
Modulation of the arginase pathway contributes to the intracellular survival of multiple pathogens, including Salmonella and Mycobacterium spp. [29]. To determine whether Arg2 directly interferes with NO production and enhances B. pseudomallei intracellular survival, we treated macrophages with the arginase inhibitor, nor-NOHA. A significant decrease in B. pseudomallei intracellular survival was observed in nor-NOHA-treated BMDM (Fig. 3B) and this corresponded to a significant increase in nitrite levels (Fig. 3C). Collectively, these results indicate that Arg2 expression promotes B. pseudomallei intracellular survival, in part, through suppression of macrophage NO synthesis.
Inhalational infection with B. pseudomallei is a natural route of exposure and represents the most likely route of infection in a deliberate biological attack [2]. In order to evaluate the role of PGE2 during pneumonic melioidosis, genetically-susceptible BALB/c mice were challenged by the intranasal route with a lethal dose of B. pseudomallei (3×103 cfu) [30]. Pulmonary infection with B. pseudomallei progressed rapidly in mice leading to greater than 20% weight loss by 72 h post-infection (Fig. 4). A significant increase in lung PGE2 was observed by 72 h post-infection and significantly correlated with disease progression (p = 0.029 by Pearson statistical analysis) (Fig. 4). These results indicate that PGE2 may play an important role in pneumonic melioidosis during the early stages of infection.
Because PGE2 inhibition enhanced bacterial clearance in vitro and because PGE2 is elevated in B. pseudomallei-infected lungs, we evaluated the efficacy of COX-2 inhibition as a post-exposure therapeutic strategy against lethal B. pseudomallei pulmonary challenge. Mice were given NS398 or mock control by i.p. administration three h after B. pseudomallei intranasal infection, and treatments were repeated for two consecutive days. Initiation of therapy within three h is clinically relevant in the case of a known biological exposure to B. pseudomallei, such as a laboratory accident. A daily maximum dose of 15 mg/kg of NS398 was selected based upon previously documented pharmacological efficacy in mice (particularly in reducing lung PGE2) without any associated toxicity [31]. Mock-treated mice infected with B. pseudomallei rapidly displayed signs of pulmonary disease and all had to be euthanized within 72 h (Fig. 5). Lungs of mock-treated mice all contained greater than 106 cfu of B. pseudomallei at the time of euthanasia. In contrast, none of the NS398-treated mice showed signs of illness until day 5 post-infection. On day 5, one mouse in the NS398-treated group displayed hind leg paralysis and was humanely euthanized. This was observed again in another animal on day 7. No bacteria were recovered from the lungs of either animal. Intranasal infection of mice with B. pseudomallei often manifests in colonization of the brain with subsequent neurologic complications [30], and we believe that this, and not pulmonary disease, likely accounted for the animals' morbidity. By day 10, all of the remaining NS398-treated mice appeared to have recovered from the infection. NS398-treated mice showed no evidence of weight loss throughout the study (not shown). No bacteria were recovered from the lungs of NS398-treated mice at the study endpoint with the exception of one animal that contained 104 cfu. All of the mice were colonized with 20–100 cfu in the spleen and liver, indicating that bacterial dissemination from the lung had occurred. These results indicate that host PGE2 production promotes the pathogenesis of B. pseudomallei during pneumonic melioidosis and that inhibition of COX-2 enhances bacterial clearance from the lung and improves host survival. Consistent with these findings, COX-2 inhibition also significantly reduced tissue bacterial burdens and pulmonary inflammation in mice infected with B. thailandensis (Supporting information, Figures S1, S2).
B. pseudomallei infection led to increased PGE2 and Arg2 expression in macrophages and both PGE2 and Arg2 enhanced B. pseudomallei intracellular survival. Furthermore, PGE2 positively regulated Arg2 expression in response to B. pseudomallei in vitro. We therefore evaluated Arg2 expression in the lungs of mice in response to bacterial infection and COX-2 inhibition. Similar to our in vitro observations, an increase in lung Arg2, but not Arg1, was observed in B. pseudomallei-infected animals compared to uninfected animals (Fig. 6). Upon COX-2 inhibition, a reduction in lung Arg2 was observed in B. pseudomallei-infected mice as evident by Western blot and densitometry analysis (Fig. 6). These results corroborate our observations in murine macrophages and advocate a supporting role for Arg2 in PGE2-mediated immunosuppression during B. pseudomallei infection.
Enhancement of non-specific innate immunity represents an attractive therapeutic strategy to combat infection with multidrug resistant bacterial pathogens. In this study, we demonstrate a critical role for PGE2 in the early pathogenesis of B. pseudomallei pulmonary infection and identify the PGE2 pathway as an immunotherapeutic target in melioidosis. PGE2 can negatively regulate innate immunity by suppression of leukocyte activation [32], macrophage microbicidal activity [23], and NK cell function [33]. Over-production of PGE2 has also been observed in a number of clinical conditions associated with an increased susceptibility to bacterial infection, including AIDS [34]. Therefore, the results presented here may be applicable to other intracellular bacteria, particularly those that infect the lung. PGE2 is a potent pro-inflammatory mediator in most tissues but plays an opposite role in the lung and gastric mucosa in order to limit inflammation and tissue injury upon mucosal insult [35], [36]. PGE2 concentrations in the lung are much higher than in plasma [37] so the bacterial survival advantage afforded by PGE2 in the present study may be exclusive to pulmonary infection and requires further study.
Macrophages play an important role in early host defense against B. pseudomallei as macrophage-depleted mice display an accelerated mortality during experimentally-induced melioidosis [38]. Our in vitro studies demonstrated that endogenous and exogenous PGE2 promoted B. pseudomallei intracellular survival, while inhibition of COX-2 eliminated endogenous PGE2 production and restricted bacterial growth in macrophages. The therapeutic efficacy of COX-2 inhibition at the cellular level may account for its significant protective efficacy against pulmonary melioidosis when administered three h after B. pseudomallei challenge. A number of studies have shown the clinical effectiveness of various immunostimulants delivered intranasally to mice prior to B. pseudomallei pulmonary challenge but none have demonstrated significant protective efficacy as a stand-alone post-exposure therapeutic [6], [8], [9], [39]. For example, CpG ODN was ineffective when given as early as one h after intranasal B. pseudomallei challenge due to a delay in recruitment of inflammatory monocytes and neutrophils to the lung [8]. Unlike these studies, the COX-2 inhibitor was administered intraperitoneally to mice after B. pseudomallei pulmonary delivery, indicating that its efficacy does not rely upon local administration and subsequent inflammatory cell recruitment. We have not yet assessed the post-exposure window of efficacy for COX-2 inhibition but we postulate that its direct action on infected macrophages, with less dependence on additional phagocyte recruitment, may allow a greater time frame for therapeutic intervention than we have already shown.
PGE2-mediated suppression of macrophage bactericidal ability has been observed in other bacterial pulmonary infections and is not restricted to the mouse model. For example, pre-treatment of rat alveolar macrophages with the COX-1/2 inhibitor, indomethacin, or antagonists of the PGE2 receptors, EP-2 and EP-4, augmented NADPH oxidase and ROS production and improved killing of Klebsiella pneumoniae [23]. It has been proposed that the timing and concentration of PGE2 determines the macrophage bactericidal response to stimuli such as IFN-γ and LPS [40]. Therefore, rapid production of high concentrations of PGE2 in response to B. pseudomallei may suppress macrophage control of bacterial growth early in infection. While our results demonstrated that PGE2 suppressed NO and enhanced Arg2, the mechanism(s) by which PGE2 exerts its suppressive effects on B. pseudomallei-infected macrophages may involve inhibition of additional reactive oxygen [23], [41] and nitrogen species [42] or pro-inflammatory cytokines [43] and warrants further study. Nonetheless, our studies add B. pseudomallei to the growing list of intracellular pathogens that utilize host arginase to facilitate their survival [29]. PGE2-mediated immunosuppression through arginase induction is well-established in cancer [44]. Our study highlights a similar mechanism operating during an intracellular bacterial infection. The clinical benefit of COX-2 inhibitors against lung carcinoma [44] and other cancers [10], [45] lends support to the potential use of this class of inhibitors against respiratory bacterial infection.
Interestingly, heat-inactivated B. pseudomallei did not induce appreciable levels of COX-2 expression or PGE2 production by macrophages indicating that the response is not stimulated by bacterial LPS. Although bacterial flagellin has been shown to induce COX-2 expression through TLR5 recognition and p38 MAPK signaling [46], P. aeruginosa strains lacking flagellin induce COX-2 expression comparable to wild type strains [41]. In addition, the majority of our in vitro studies were performed using murine BMDM which have been shown to be unresponsive to flagellin due to an absence of TLR5 [47] [48]. Collectively, these findings suggest that the macrophage PGE2 response is predominantly regulated by active bacterial processes as opposed to a passive signaling event mediated by bacterial PAMPs. In support of this view, COX-2 expression in the lungs of mice infected with Pseudomonas aeruginosa was dependent upon viable bacteria and the presence of the type three secretion system effector, ExoU, a member of the phospholipase A family [41]. Although the B. pseudomallei effector(s) responsible for the induction of PGE2 remain to be identified, it is plausible that secreted bacterial phospholipases accelerate phospholipid release and turnover in the host cell leading to increased COX-2 and PGE2 expression [49].
To our knowledge, this study is the first to characterize a post-exposure immunotherapeutic that provides significant protection against lethal B. pseudomallei pulmonary infection in mice. In an experimental mouse model of tuberculosis, inhibition of PGE2 reduced bacillary loads and increased granuloma formation, concomitant with increased IFN-γ, TNF-γ, and iNOS expression, suggesting that PGE2 may contribute to M. tuberculosis persistence by down-regulation of cell-mediated immunity (CMI) [50]. Similar findings were reported for F. tularensis live vaccine strain (LVS) pulmonary challenge in mice [51]. Inhibition of PGE2 reduced bacterial loads in the tissues and enhanced CMI responses. COX-2−/− and EP2−/− mice demonstrated accelerated clearance of Pseudomonas aeruginosa from the lungs compared to wild type mice, and PGE2 signaling via EP2 suppressed macrophage ROS production in vitro [41]. COX-2 inhibition also improved bacterial clearance [32], [41] and enhanced host survival [41] during intratracheal infection with P. aeruginosa. Collectively, these studies and ours suggest that PGE2 production promotes bacterial pathogenesis in the lung and that inhibition of COX-2 may represent a broad-spectrum immunotherapeutic against multiple bacterial pathogens. These results compel further investigation of the role of PGE2 in human melioidosis, particularly in patients with pneumonia. Use of commercially-available selective COX-2 inhibitors as an adjunct therapy to antibiotic treatment should also be explored in animal models of melioidosis as combination therapy may further eradicate persistent bacteria.
|
10.1371/journal.ppat.1004047 | Parasite Fate and Involvement of Infected Cells in the Induction of CD4+ and CD8+ T Cell Responses to Toxoplasma gondii | During infection with the intracellular parasite Toxoplasma gondii, the presentation of parasite-derived antigens to CD4+ and CD8+ T cells is essential for long-term resistance to this pathogen. Fundamental questions remain regarding the roles of phagocytosis and active invasion in the events that lead to the processing and presentation of parasite antigens. To understand the most proximal events in this process, an attenuated non-replicating strain of T. gondii (the cpsII strain) was combined with a cytometry-based approach to distinguish active invasion from phagocytic uptake. In vivo studies revealed that T. gondii disproportionately infected dendritic cells and macrophages, and that infected dendritic cells and macrophages displayed an activated phenotype characterized by enhanced levels of CD86 compared to cells that had phagocytosed the parasite, thus suggesting a role for these cells in priming naïve T cells. Indeed, dendritic cells were required for optimal CD4+ and CD8+ T cell responses, and the phagocytosis of heat-killed or invasion-blocked parasites was not sufficient to induce T cell responses. Rather, the selective transfer of cpsII-infected dendritic cells or macrophages (but not those that had phagocytosed the parasite) to naïve mice potently induced CD4+ and CD8+ T cell responses, and conferred protection against challenge with virulent T. gondii. Collectively, these results point toward a critical role for actively infected host cells in initiating T. gondii-specific CD4+ and CD8+ T cell responses.
| CD4+ and CD8+ T cells are critical for controlling many infections. To generate a T cell response during infection, T cells must encounter the microbial peptides that they recognize bound to MHC molecules on the surfaces of other cells, such as dendritic cells. It is currently unclear how dendritic cells acquire the antigens they present to T cells during infection with many intracellular pathogens. It is possible that these antigens are phagocytosed and processed by dendritic cells, or antigens may be presented by cells that are infected by pathogens such as Toxoplasma gondii, which invades host cells independently of phagocytosis. To differentiate these pathways, we developed a novel technique to track the fate of T. gondii in vivo that distinguishes actively infected cells from those that phagocytosed parasites. This technique was used to examine each of these cell populations. We also used pharmacological inhibitors of parasite invasion, and the transfer of sort-purified infected or uninfected dendritic cells and macrophages to determine what roles phagocytosis and active invasion have in the initiation of T cell responses. Our results demonstrate that phagocytosis of parasites is not sufficient to induce CD4+ or CD8+ T cell responses, whereas infected cells are critical for this process.
| Toxoplasma gondii is an intracellular protozoan parasite of medical and veterinary significance that can induce acute disease in its host and is an important opportunistic pathogen in immunocompromised individuals [1], [2]. Successful control of this pathogen requires a rapid TH1 immune response, characterized by the production of the cytokine IL-12, which promotes the ability of parasite-specific CD4+ and CD8+ T cells to produce the cytokine Interferon-γ (IFN-γ) [3], [4], [5]. The initiation of CD8+ T cell responses is a complex process which requires that professional antigen presenting cells acquire antigens and present them in the context of Major Histocompatibility Complex (MHC) I, and multiple models have been proposed to explain how this may occur during toxoplasmosis [6], [7]. For example, in other systems, foreign antigens are acquired through the pinocytosis of soluble antigens, the phagocytosis of large particulate antigens, or the phagocytosis of host cells containing foreign antigens, and subsequently presented to CD8+ T cells through cross-presentation [8], [9]. A role for cross presentation during toxoplasmosis is supported by in vivo imaging studies showing that uninfected dendritic cells interact extensively with parasite-specific CD8+ T cells [6], [10], [11]. Alternatively, since T. gondii is an intracellular parasite, actively infected dendritic cells may acquire parasite-derived antigens from their intracellular environment independently of phagocytosis and directly prime naïve CD8+ T cells. Indeed, the ability of cells actively infected by T. gondii to prime or present antigen to CD8+ T cells has been observed in vitro [12]–[14] and the critical role of perforin in immunity to T. gondii implicates the cytolysis of infected host cells as a mechanism of defense, thus arguing that infected cells can present antigen to effector CD8+ T cells in vivo [15]. However, several caveats must be acknowledged in interpreting these studies. Firstly, the ability of infected cells to present antigens to reporter cells lines or activated effector CD8+ T cells does not necessarily indicate that infected cells can prime naïve CD8+ T cells, and events that occur in vitro may not represent the in vivo situation. Additionally, it can be difficult to distinguish actively infected host cells from those that have phagocytosed the parasite by flow cytometry, thus confounding experimental interpretation. Furthermore, like many intracellular pathogens, T. gondii has been reported to inhibit the expression or upregulation of molecules involved in antigen presentation such as MHCI, CD40, CD80, and CD86 on infected cells, suggesting that the ability of infected cells to prime naïve CD8+ T cells may be compromised [16]–[18].
Antigens presented to CD4+ T cells in the context of MHCII may also be derived from the extracellular or intracellular environment of the host cell. Endocytosed antigens can be presented in the context of MHCII, and this pathway is considered to be the primary mechanism by which antigens are acquired for presentation to CD4+ T cells [19]. However, intracellular antigens can also be presented in the context of MHCII, as cytosolic peptides are presented in the context of MHCII by B cells and macrophages [20]. Similarly, in vitro studies have demonstrated that viral or model antigens expressed intracellularly can be presented to CD4+ T cells independently of phagocytosis [21]–[29]. Despite these findings, the role of infected cells in presenting antigen to CD4+ T cells in vivo during any infection remains unclear [30]. In the case of T. gondii, downregulated expression of MHCII and other molecules involved in antigen presentation has been observed on infected cells, and cells infected with T. gondii exhibit decreased ability to present antigen in vitro [16]–[18]. Furthermore, in vitro studies have observed that antigens from heat-killed or invasion-inhibited parasites incubated with dendritic cells can be presented in the context of MHCII, consistent with a role for phagocytosis-dependent antigen presentation to CD4+ T cells [12].
There are several difficulties involved with addressing the relative contributions of phagocytosis versus active invasion to antigen presentation in vivo during many infections. For example, interfering with these pathways can result in changes in pathogen burden and inflammation that confound experimental interpretation, and the parasite-mediated lysis of host cells and re-infection may obscure the analysis of the earliest cell populations that interact with the pathogen. In addition, there are limited tools to distinguish host cells that have phagocytosed pathogens from those that have been productively infected. In the present study, these issues are addressed using a non-replicating uracil auxotrophic vaccine strain of T. gondii (the cpsII strain) [31]–[33] and a novel assay that tracks the fate of parasites and distinguishes active invasion from phagocytosis in vivo. Using these approaches, cpsII parasites were found to infect large numbers of macrophages and dendritic cells, and dendritic cells were found to be necessary for optimal cpsII-induced CD4+ and CD8+ T cell responses. Infected dendritic cells displayed an activated phenotype, characterized by high levels of CD86 and MHCI expression, which was unique from the phenotype of dendritic cells that had phagocytosed T. gondii. Furthermore, the administration of heat-killed or invasion-blocked parasites did not induce CD4+ or CD8+ T cell responses, thus demonstrating that phagocytosis of parasites is insufficient to activate naïve T cells. Lastly, the selective transfer of infected dendritic cells or macrophages, but not those that had phagocytosed T. gondii, to naïve mice resulted in robust CD4+ and CD8+ T cell responses and protection from challenge with a virulent strain of T. gondii. These findings point toward a critical role for infected cells in initiating the adaptive immune response to T. gondii.
To distinguish between parasites that are phagocytosed by host cells and those that actively infect host cells, differences in sensitivity to pH between the fluorescent markers mCherry and CellTrace Violet were exploited. When mCherry-expressing parasites were labeled intracellularly with CellTrace Violet and incubated overnight in buffer solutions of varying pH, mCherry fluorescence was retained (Figure 1a). In contrast, violet fluorescence intensity was maintained at pH 7.0 but was decreased at low pH (Figure 1a). The ability of this system to distinguish active invasion from phagocytosis was demonstrated in vitro by incubating Violet-labeled, mCherry-expressing cpsII parasites with macrophages and examining fluorescence by flow cytometry 1 hour and 18 hours post-infection. At one hour after incubation with parasites, two distinct macrophage populations were present: One displayed mCherry and Violet fluorescence, while the other was negative for both markers (Figure 1b). However, by 18 hours, two distinct mCherry+ve populations were apparent. One population displayed no loss of mCherry or Violet fluorescence (mCherry+veViolet+ve), while the other population had decreased mCherry fluorescence associated with a complete loss of violet fluorescence (mCherry+veViolet−ve). Utilizing ImageStream flow cytometry to generate images of individual cells from each of these populations revealed that the mCherry+veViolet+ve cells contained intact parasites, while the mCherry+veViolet−ve cells contained dimmer and more diffuse mCherry fluorescence (Figure 1c, Figure S1). Instances in which cells contained both diffuse fluorescence and intact parasites were rare (<3% of infected cells). Furthermore, pre-treatment of parasites with the irreversible inhibitor of invasion 4-p-bromophenacyl bromide (4-p-bpb) (thus making parasites targets for phagocytosis) [12], [34]–[36], resulted in the complete loss of the mCherry+veViolet+ve population at 18 hours post-infection (Figure 1b,c, Figure S1). Staining with LysoTracker, a fluorescent dye that specifically stains acidified compartments [37], enabled parasites that localized to acidified compartments to be distinguished from those that persist in non-acidified compartments. Both of these populations of parasites (LysoTracker+ve and LysoTracker−ve) were apparent when untreated (invasion competent) parasites were incubated with bone marrow-derived macrophages one hour post-infection (Figure S2). In contrast, when invasion was pharmacologically inhibited parasites localized exclusively to the acidified compartments at these early time points, and at later time points the diffuse mCherry+ve fluorescence localized most commonly to a LysoTracker−ve compartment. Collectively, these results are consistent with a model in which phagocytosed parasites are degraded, and the acidic environment of the phagosome leads to a loss of Violet fluorescence, while mCherry fluorescence is retained. In contrast, when the parasite actively invades host cells and persists in the less acidic environment of the parasitophorous vacuole (PV), both Violet and mCherry fluorescence are retained.
The ability to distinguish active invasion from phagocytosis was then utilized to determine the fate of cpsII parasites in vivo. When C57BL/6 mice were vaccinated intraperitoneally with Violet-labeled, mCherry-expressing parasites, mCherry+veViolet+ve and mCherry+veViolet−ve populations were apparent in the Peritoneal Exudate Cells (PECS) 18 hours post-vaccination, and the presence of the mCherry+veViolet+ve population was abrogated by pre-treating the parasites with 4-p-bpb (Figure 1d). Furthermore, when Violet+ve cells were sorted and cytospins were examined, they were found to contain intact parasites (Figure 1e). ImageStream analysis also revealed that the mCherry+veViolet+ve population contained intact parasites whereas the mCherry+veViolet−ve population displayed diffuse mCherry fluorescence (Figure 1f, Figure S3). Collectively, these studies demonstrate that the use of fluorescent markers with differing pH sensitivities can be used to distinguish cells that have phagocytosed T. gondii from those that have been actively infected.
To measure the persistence of cpsII parasites in vivo, bioassays were performed in which tissues from vaccinated mice were cultured in the presence of exogenous uracil and examined by microscopy for the presence of cpsII parasites. Using this method, cpsII parasites were detected in all mice examined at day 3 post-infection. However, by day 5 post-infection, 50% of mice had cleared the infection, and by day 10 post-infection, no parasites could be detected. These data suggest that cpsII parasites are ultimately cleared from the host, and are consistent with previous studies, in which parasite DNA could not be detected in the peritoneal cavities or spleens of cpsII-vaccinated mice when measured 3 weeks post-infection [38].
To determine the mechanisms by which cpsII parasites may ultimately be cleared from host cells, their fate within infected host cells was examined in vitro. Since IFN-γ (in combination with LPS or TNF-α) can induce the recruitment of immune enzymes such as the Immunity Related Guanosine Triphosphatases (IRGs) to the PV, and these enzymes have been implicated in the rupture of the PV which leads to the xenophagic elimination of the parasite [39], the colocalization of the parasite with Irgb6 (a member of the IRG family) and LAMP-1 (which is expressed on lysosomes) in IFN-γ–activated cells and untreated cells was examined using immunofluorescence microscopy, to determine if IFN-γ induced the elimination of cpsII parasites within infected cells. When the subcellular localization of live cpsII parasites was examined, it was apparent that these parasites did not colocalize with either Irgb6 or LAMP-1 in IFN-γ-activated or untreated macrophages, at any time point examined (ranging from 3 hours post-infection to 5 days post-infection) (Figure 2a–b). In contrast, LAMP-1 colocalized with heat-killed parasites, consistent with the idea that heat-killed parasites are phagocytosed. These data argue against the notion that cpsII parasites are eliminated by xenophagy, and demonstrate that these parasites can persist within infected cells for long periods of time. Electron microscopy was also utilized to examine the integrity of the PV, since IFN-γ can induce the blebbing and rupture of the PV during infection with replicating strains of T. gondii [40], [41]. Using this approach, cpsII-infected macrophages were consistently observed to contain intact PVs and blebbing was not apparent (Figure 2c). Additionally, some cpsII parasites showed atypical morphology, indicative of non-productive cell division (Figure 2d). Collectively, these results confirm that cpsII parasites cannot replicate within host cells, and suggest that cpsII parasites can persist within infected cells, evading IFN-γ-mediated destruction, although they are eventually cleared from the host.
To better understand the fate of cpsII parasites in vivo, mice were challenged intraperitoneally with Violet-labeled, mCherry-expressing cpsII parasites, and flow cytometry was performed on the PECS 18 hours later to characterize the cell populations that had phagocytosed T. gondii or were actively infected. The largest population of mCherry+veViolet+ve cells to be infected was CD11bHI macrophages, which comprised 44.0±16.7% of infected cells. Dendritic cells (which have been previously implicated in the induction of T cell responses to cpsII [42]) comprised 8.3±2.8% of infected cells (Figure 3a,b). Of the infected dendritic cells the vast majority (97.8±2.0%) belonged to the Gr-1−veCD11bHI subset (data not shown). Although T. gondii is capable of infecting any nucleated cell, when the frequencies of CD11bHI macrophages and dendritic cells within the population of infected cells (44.0±16.7% and 8.3±2.8%, respectively) were compared to their frequencies within the total population of peritoneal cells in vaccinated mice (11.3±7.9% and 1.3±0.4%, respectively), it was apparent that macrophages and dendritic cells are overrepresented among cells infected by the parasite (Figure 3c). Analysis of the population that had phagocytosed T. gondii revealed 46.0±20.6% of these cells were CD11bHI macrophages, whereas dendritic cells represented 6.2±3.2% of this population (Figure 3a,b). Additionally, 23.4±9.9% of the cells that had phagocytosed the parasite stained positive for markers for T, B or NK cells (CD3, CD19 and NK1.1, respectively). Further sub-setting revealed these cells to be B cells, consistent with previous reports identifying a population of phagocytic B cells in the peritoneal cavity (Figure 3b, data not shown) [43], [44]. Parasites were not detected in lymph nodes or spleens by flow cytometry, and parasites could not be cultured from these tissues at days 3,5 or 10 post-vaccination.
The phenotype of infected cells and those that phagocytosed the parasite was compared by analyzing expression levels of MHCI and MHCII, as well as the costimulatory molecules CD86 and CD40. Although vaccination with cpsII resulted in an overall increase in expression of MHCI on CD11bHI macrophages, macrophages that had phagocytosed the parasite and those that were infected displayed similar levels of MHCI to the total population present in the PECS of vaccinated mice. In contrast, dendritic cells that had phagocytosed cpsII and those that were infected by the parasite displayed higher levels of MHCI relative to the total dendritic cell population in the peritoneal cavity (Figure 4a). Vaccination with cpsII induced no significant changes in MHCII expression on dendritic cells, although infected macrophages had lower levels of MHCII than the total population in the PECS (Figure 4b). Expression of CD86 was markedly higher on macrophages and dendritic cell populations that were infected by the parasite, but not the populations that had phagocytosed the parasite (Figure 4c). While vaccination induced increased CD40 expression on the total dendritic cell population, infected cells displayed similar expression levels to the total population, and those that phagocytosed the parasite exhibited the highest levels of expression (Figure 4d). Collectively, these results reveal a complex pattern demonstrating that infected macrophages and dendritic cells display activated phenotypes, characterized by the upregulation of MHCI and CD86, and constitutive expression of CD40 and MHCII, which is distinct from the phenotype of cells that phagocytosed T. gondii.
Given the activated phenotype of dendritic cells infected with cpsII versus those that had phagocytosed the parasite, studies were performed to determine the role of dendritic cells in the development of CD4+ and CD8+ T cell responses to this strain. Mice that express the diphtheria toxin receptor under the control of the CD11c promoter (CD11c-DTR mice) were used to test the requirement for dendritic cells to prime T cells [45]. In these experiments, CD11c-DTR mice were treated with diphtheria toxin, which resulted in a 70–90% reduction in dendritic cells (Figure 5a). One day following the administration of diphtheria toxin, mice were challenged with a strain of cpsII engineered to express Ovalbumin (cpsII-OVA) [38]. At eight days following vaccination, CD4+ and CD8+ T cell responses were measured using MHCII tetramers, which bind CD4+ T cells specific for the endogenous T. gondii epitope CD4Ag28m combined with magnetic enrichment for the tetramer+ve population [46], [47], and MHCI tetramers for OVA-specific CD8+ T cells. Additionally, the surface molecule CD11a, which is upregulated on antigen-experienced CD4+ and CD8+ T cells [48], [49], and the intracellular molecule Ki67, which is indicative of cellular proliferation [50], were used to estimate the total CD4+ and CD8+ T cell responses to T. gondii. Indeed, vaccination with cpsII induced a two-fold increase in the frequency of CD11aHIKi67HI cells and an expansion in the number of CD11aHI CD4+ T cells specific for the CD4Ag28m epitope, but depletion of dendritic cells inhibited these responses (Figure 5b). Similarly, cpsII vaccination induced an increase in CD11aHIKi67HI and OVA-specific CD8+ T cells, however these responses were decreased in mice depleted of dendritic cells (Figure 5c). Furthermore, when Flt3L−/− mice (which have global defects in numbers of dendritic cells [51]) or Batf3−/− mice (which have a defect in numbers of CD8a+ dendritic cells [52]) were challenged with cpsII-OVA, both mice displayed marked defects in tetramer-specific and total CD4+ and CD8+ T cell responses (Figure S4,S5).
Given the numbers of macrophages that were either infected or which had phagocytosed T. gondii, experiments were performed to assess their role in the cpsII-induced T cell responses. However, attempts to deplete macrophages using clodronate liposomes also resulted in significant depletion of dendritic cells, making it difficult to assess the specific contribution of macrophages (data not shown). However, because monocytes were observed to interact with parasites (Figure 3b), and these populations can develop into dendritic cells that express CD11c, experiments were performed to assess their role in generating CD4+ and CD8+ T cell responses following cpsII vaccination. Therefore, mice deficient in the chemokine receptor CCR2, which promotes the recruitment of inflammatory monocytes to sites of inflammation during toxoplasmosis [53], were immunized with cpsII-OVA parasites. Despite having a defect in monocyte recruitment to the peritoneum, CCR2−/− mice had similar cpsII-induced CD4+ and CD8+ T cell responses to WT control mice (Figure S6), thus arguing against a critical role for inflammatory monocytes in presenting antigen to CD4+ and CD8+ T cells following cpsII-vaccination. Collectively, these results establish a role for dendritic cells in the generation of CD4+ and CD8+ T cell responses following cpsII vaccination.
To assess the contribution of phagocytosis to the generation of CD4+ and CD8+ T cell responses, mice were challenged with live cpsII-OVA parasites, heat-killed cpsII-OVA parasites, or parasites pre-treated with the irreversible inhibitor of invasion 4-p-bpb. As expected, vaccination with live parasites induced a robust CD4+ T cell response, however these responses were abrogated when parasites were killed or invasion was inhibited (Figure 6a). Similarly, CD11aHIKi67HI and OVA-specific CD8+ T cells were detected when mice were administered live, but not heat-killed or invasion-inhibited parasites (Figure 6b). Indeed, even when the dose of heat-killed parasites was increased to 107 parasites (100× the typical dose of live parasites used in these experiments), no CD4+ or CD8+ T cell responses could be detected (Figure S7). Additionally, gp91−/− mice, which have a defect in cross-presenting antigens to CD8+ T cells [54], developed normal CD8+ T cell responses following cpsII-vaccination (data not shown). Collectively, these data indicate that phagocytosis of parasites is insufficient to induce CD4+ and CD8+ T cell responses, and point toward a critical role for infected cells in these processes.
To determine whether infected dendritic cells were sufficient to generate CD4+ and CD8+ T cell responses, bone marrow-derived dendritic cells cultured in GM-CSF (which are CD11bHICD8α−ve) were infected with violet-labeled, mCherry-expressing cpsII parasites in vitro overnight, and FACS sorting was used to purify the uninfected (mCherry−veViolet−ve) and infected cells (mCherry+veViolet+ve) from the same cultures, and each of these fractions was then administered to naïve mice. In addition, bone marrow-derived dendritic cells were cultured with invasion-blocked parasites, and the populations of DCs that had phagocytosed the parasite (mCherry+veViolet−ve) were also isolated by FACS sorting, and administered to mice. This experiment allowed a direct comparison of the ability of infected dendritic cells and dendritic cells that phagocytosed T. gondii to induce CD4+ and CD8+ T cell responses in vivo. In mice administered uninfected dendritic cells cultured with parasites, or dendritic cells that had phagocytosed parasites, there was no detectable increase in Ki67+veCD11aHI, antigen-experienced CD4+ or CD8+ T cells (Figure 7a,b). In contrast, mice administered cpsII-infected dendritic cells developed CD4+ and CD8+ T cell responses as determined by tetramer-binding as well as expression of Ki67 and CD11a (Figure 7a,b). Furthermore, when vaccinated mice were challenged 6 weeks later with a highly virulent strain of T. gondii, only those mice administered cpsII-infected dendritic cells displayed a ∼90% reduction in parasite burden (Figure 7c). Similar results were obtained using splenic dendritic cells, which are composed of both CD8α+ and CD8α− dendritic cells (data not shown). Moreover, the transfer of sort-purified infected bone marrow-derived macrophages to mice also induced CD4+ and CD8+ T cell responses and protected mice from challenge, whereas the transfer of macrophages that had phagocytosed parasites did not induce T cell responses or protection (Figure S8). Collectively, these results demonstrate a key role for infected cells in the induction of CD4+ and CD8+ T cell responses, and protective immunity upon re-challenge.
There are many fundamental questions about the mechanisms of antigen presentation that lead to the activation of CD4+ and CD8+ T cells during toxoplasmosis and multiple studies have addressed the ability of actively infected cells to present antigen [12]–[14], [55]. The present work highlights that following challenge in vitro or in vivo with live parasites there are high rates of phagocytosis and the combination of flow cytometry and parasites that express a single fluorescent reporter protein are not sufficient to distinguish infected cells from those that phagocytose T. gondii. Rather, the ability to combine parasites that express a pH insensitive reporter such as mCherry protein with a pH sensitive dye and analysis by high throughput imaging and flow cytometry provide a unique opportunity to examine parasite fate and host cell phenotype. This approach should be broadly applicable to determining the fate of other intracellular fungal, bacterial and parasitic pathogens [56]–[62]. Regardless, the ability to distinguish active invasion from phagocytosis revealed that macrophages and dendritic cells infected by T. gondii have unique activation phenotypes when compared to those that have phagocytosed the parasite. Previous reports have indicated that infection with T. gondii inhibits the maturation of professional antigen presenting cells [6], [16], [18], [63], but the data presented here are more consistent with the idea that infection induces DC maturation [36], [55], [64]–[66]. The experiments in which dendritic cells were selectively depleted, or pre-infected dendritic cells were transferred to mice highlight the important role of these accessory cells in generating CD4+ and CD8+ T cell responses following cpsII-vaccination. However, these findings do not rule out the possibility that other cell types are also involved. Indeed, the transfer of infected bone marrow-derived macrophages could also induce CD4+ and CD8+ T cell responses, suggesting that resident macrophages may also contribute to the T cell responses that occur following cpsII vaccination.
In current paradigms, the direct phagocytosis or endocytosis of soluble and particulate non-infectious antigens is the major pathway that allows antigens to be presented in the context of MHCII to CD4+ T cells [19]. Similarly, phagocytosed antigens are thought to be presented to CD8+ T cells through the process of cross-presentation [8]. However, the multiple approaches presented here indicate that phagocytosis of T. gondii is not sufficient to generate T cell responses. The finding that infected dendritic cells and macrophages display activated phenotypes and are able to promote CD4+ and CD8+ T cells responses in vivo distinguishes them from populations that phagocytose T. gondii. These observations suggest that live (as opposed to phagocytosed) parasites may uniquely activate innate sensing mechanisms that are linked to antigen presentation. This may relate to the persistence of parasites that occurs in infected cells, or to the engagement of mechanisms that allow the host to distinguish viable parasites from those that had been phagocytosed and would be killed [67]. The failure of cells that phagocytose the parasite to upregulate expression of CD86 is consistent with this idea. Another possibility is that dendritic cells actively infected with T. gondii display a hypermotile phenotype and enhanced migration to lymph nodes, a process that is considered essential for T cell priming [68]–[72]. Differences in cellular motility between infected cells and those that phagocytose parasites may account for the apparent discrepancy between the previous studies that showed that phagocytosis of parasites is sufficient to prime CD4+ T cells in vitro [12] and our finding that this process is not sufficient in vivo.
Regardless of the reasons that cells that phagocytose T. gondii fail to prime T cells, the data presented here are consistent with models in which infected cells either directly prime CD4+ T and CD8+ T cells, or are taken up by efferocytosis (i.e. the phagocytosis of apoptotic cells), leading to antigen presentation. Since T. gondii resides in a specialized non-fusogenic vacuole, it is unclear how parasite antigens may escape the PV for processing and presentation by infected cells. One possibility is that parasite antigens are acquired for presentation from the intracellular environment through the xenophagic elimination of cpsII parasites. Indeed, autophagic machinery has been implicated in the elimination of T. gondii [40], [73], [74], and antigen acquired through autophagy can be subsequently presented [23], [24], [75]. However, the lack of recruitment of Irgb6 and LAMP-1 to the PVs containing cpsII parasites argues against this idea. Other possible mechanisms that would allow parasite material to enter antigen processing pathways include the fusion of the PV with the endoplasmic reticulum [12], the secretion of antigen into the cytoplasm during invasion [76], or leakage of antigen out of the PV [14]. More recent work has shown that T. gondii can secrete antigens into host cells without subsequently infecting these cells [77]. This population of injected-but-uninfected cells may also contribute to the host immune response, and the ability to track these abortive invasion events in vivo, as well as the ability to divorce injection from infection through modulation of the parasite, may provide further insight into the pathways involved in antigen processing during cpsII vaccination.
Given the lack of overt inflammation observed during infection with cpsII parasites, the absence of parasite-driven cytolysis of host cells, and limited antigen load, it remains surprising that relatively low numbers of these parasites are able to generate strong protective CD4+ and CD8+ T cell responses, comparable to those seen during live infection [31]–[33], [38], [42], [78]. Increased antigenic burden is generally associated with increased T cell responses, and inflammatory signals can promote pathways involved in antigen presentation, T cell proliferation, and T cell survival [79]–[81]. Caution is therefore required when extrapolating these findings to natural infection with replicating parasites. Regardless, the finding that phagocytosis is insufficient to induce antigen presentation in this system highlights the importance of alternative approaches to deliver antigens for vaccine design and immunotherapies, such as those that target antigens to the host cell cytosol [82]. Furthermore, while many studies have utilized models of murine infection to elucidate the factors involved in the generation of T cell responses and the formation of memory T cells, vaccination with cpsII parasites allows these processes to be studied in a setting in which overt inflammation is limited. Thus, this experimental system may prove valuable to dissect basic principles that lead to the generation of long-lived T cell responses that translate easily to vaccine design, where inflammation should also be limited.
All procedures involving mice were reviewed and approved by the Institutional Animal Care and Use Committee of the University of Pennsylvania (Animal Welfare Assurance Reference Number #A3079-01) and were in accordance with the guidelines set forth in the Guide for the Care and Use of Laboratory Animals of the National Institute of Health.
Flt3L−/− mice were obtained from Taconic Farms (Germantown, NY). Batf3−/− mice, CCR2−/− and CD11c-DTR mice were obtained from Jackson Laboratories. C57BL/6 mice were obtained from Jackson Laboratories or Taconic Farms. All mice were kept in specific-pathogen-free conditions at the School of Veterinary Medicine at the University of Pennsylvania. For experiments in which dendritic cells were depleted, CD11c-DTR or WT control mice were administered 100 ng of Diphtheria Toxin (Sigma-Aldrich) diluted in 100 µL of PBS (Invitrogen) intraperitoneally ∼12 hours prior to vaccination. Depletion efficiency was typically 90%.
All experiments were performed using cpsII parasites, cpsII-OVA parasites [38], cpsII-OVA-mCherry parasites, or RH-OVA-Tomato parasites. RH-OVA-Tomato parasites [83] and cpsII-OVA parasites [38], [84] have been previously described. CpsII-OVA parasites and were derived from the RHΔcpsII clone, which was provided as a generous gift by Dr. David Bzik [31]. CpsII-OVA-mCherry parasites were derived from the cpsII-OVA clone using the previously described methods [76], [77], with the exception that parasites were selected using zeomycin as previously described [85]. Parasites were cultured and maintained by serial passage on human foreskin fibroblast cells in the presence of parasite culture media [71.7% (Corning), 17.9% Medium 199 (Invitrogen), 9.9% Fetal Bovine Serum (FBS)(Invitrogen), 0.45% Penicillin and Streptomycin (Invitrogen)(final concentration of 0.05 units/ml Penicillin and 50 µg/ml Streptomycin), 0.04% Gentamycin (Invitrogen)(final concentration of 0.02 mg/ml Gentamycin)], which was supplemented with uracil (Sigma-Aldrich)(final concentration of 0.2 mM uracil) in the case of cpsII, cpsII-OVA and cpsII-OVA-mCherry parasites. For infections, parasites were harvested and serially passaged through 18, 20 and 26 gauge needles (BD) before filtration with a 5 µM filter (Sartorius Stedim). Parasites were washed extensively with PBS and mice were injected intraperitoneally with 105 or 106 parasites suspended in PBS. In vitro experiments were performed at an MOI of 0.5 or 1. For experiments in which CellTrace Violet (Invitrogen) was utilized to track the fate of parasites, CellTrace Violet was diluted in 200 µL of DMSO to obtain a 0.5 mM stock solution. Parasites were washed once with PBS before incubation in 0.5 µM CellTrace Violet diluted in PBS for 10–25 minutes at 37°C. This reaction was quenched by the addition of ∼40 volumes of complete media [88.5% RPMI 1640 (Corning), 8.8% FBS (Invitrogen), 0.9% Sodium Pyruvate (Gibco), 0.9% Penicillin and Streptomycin (Invitrogen)(final concentration of 0.1 units/ml Penicillin and 100 µg/ml Streptomycin), 0.9% MEM Non-essential Amino Acids Solution (Gibco) and 0.18% beta-2-mercaptoethanol (Gibco)] and parasites were washed extensively. In experiments in which 4-p-bromophenacyl bromide (4-p-bpb) was utilized to inhibit parasite invasion, 4-p-bpb (Sigma-Aldrich) was prepared fresh for each experiment and dissolved in DMSO (Sigma-Aldrich) to make a 0.1 M stock solution. Parasites were incubated in a 100 µM solution of 4-p-bpb in Fetal Bovine Serum at a concentration of 107 parasites/ml for 10 minutes, and the reaction was quenched by the addition of ∼40 volumes of complete media, followed by extensive washing [12]. To heat-kill parasites, parasites were incubated at 60°C for 1 hour in PBS [86]. Death was confirmed using Trypan Blue staining (Corning).
Peritoneal exudate cells were obtained by peritoneal lavage with 5 ml of PBS. Splenocytes and lymphocytes were obtained by grinding spleens and lymph nodes over a 40 µM filter (Biologix) and washing them in complete media. Red blood cells were then lysed by incubating for 5 minutes at room temperature in 5 ml of lysis buffer [0.864% ammonium chloride (Sigma-Aldrich) diluted in sterile de-ionized H2O)], followed by washing with complete media. Bone marrow-derived macrophages were obtained using previously described methods [83], [87]. Immortalized macrophages from C57BL/6 mice were obtained by transforming bone marrow-derived macrophages with the J2 Virus and were cultured in macrophage media [88].
Tetramer-specific CD4+ T cells were measured using the protocol previously described [46]. MHCII Tetramer was obtained as generous gifts from Drs. Marc Jenkins and Marion Pepper, and subsequently from the NIH Tetramer Core Facility, and was used at a final concentration of 10 nM. APC-MHCI-SIINFEKL Tetramer was obtained from Beckman-Coulter. Cells were washed with FACS Buffer [1× PBS, 0.2% bovine serum antigen (Sigma), 1 mM EDTA (Invitrogen)], stained with LIVE/DEAD Fixable Aqua Dead Cell marker (Invitrogen) and incubated in Fc block [99.5% FACS Buffer, 0.5% normal rat serum (Invitrogen), 1 µg/ml 2.4G2 (BD)] prior to staining. The following antibodies were used for staining: Ki67 Alexa Fluor 488 (BD, B56), CD3 APC-eFluor 780 (eBioscience, 17A2), CD8 eFluor 450 (eBioscience, 53-6.7), CD11a PerCP-Cy5.5 (Biolegend, H155-78), MHCII PE (eBioscience, M5/114.15.2), NK1.1 PE (BD, PK136), CD19 PE (eBioscience, 1D3), Foxp3 eFlour 450 (eBioscience, FJK-16a), CD4 Pe-Cy7 (eBioscience, GK1.5), CD3 FITC (BD, 145-2C11), NK1.1 FITC (eBioscience, PK136), CD19 FITC (eBioscience, 1D3), Gr-1 PerCP-Cy5.5 (eBioscience, RB6-8C5), CD11c PE-Cy7 (eBioscience, N418), CD11b APC-eFluor 780 (eBioscience, M1/70), MHCII AF700 (Biolegend, M5/114.15.2), MHCI APC (AlexaFlour647 AF6-88.5), CD86 APC (eBioscience, GL1), CD40 APC (eBioscience 1C10), CD8 eFlour 650 NC (eBioscience, 53-6.7), CD45.2 APC-eFluor 780 (eBioscience, 104), polyclonal rabbit anti-T. gondii [a generous gift from Fausto G. Araujo (Palo Alto Medical Foundation, Palo Alto, CA)], and polyclonal Goat anti-Rabbit Alexa Fluor 680 (Jackson). Intracellular staining was performed using the Foxp3/Transcription Factor Staining Buffer Set (eBioscience) following the manufacturer's instructions. Samples were run on a FACSCanto (BD) or LSR Fortessa (BD) and analyzed using FlowJo Software (TreeStar). Images were obtained using the ImageStream and analysis was performed using IDEAS software (Amnis).
Splenic dendritic cells were obtained from mice injected subcutaneously with Flt3L-secreting b16 tumor cells [89], [90] and magnetically enriched using CD11c microbeads (Miltenyi Biotech) and LD MACS separation columns (Miltenyi Biotech), following the manufacturer's instructions. Bone marrow-derived dendritic cells were obtained by culturing bone marrow cells in the presence of 40 ng/ml of GM-CSF, which was added at days 0,3,6 and 9 post-seeding. Dendritic cells or bone marrow-derived macrophages were cultured overnight with parasites at 37°C and collected the following day. Dendritic cells were then stained for MHCII, CD11c, CD45, and free parasites, and sorted for mCherry+veViolet+ve, mCherry+veViolet−ve or mCherry−veViolet−ve populations that were CD45+MHCIIHICD11cHI, and negative for free parasites using the FACSAria (BD). Macrophages were stained for CD45 and free parasites and sorted into mCherry+veViolet+ve, mCherry+veViolet−ve or mCherry−veViolet−ve populations that were CD45+ and negative for free parasites.
Bone marrow-derived macrophages from C57BL/6 mice were activated with IFN-γ and LPS for 18–24 hours or left untreated in macrophage media lacking uracil [DMEM (Gibco) supplemented with 4 mM L-glutamine (Sigma) and 10% dialyzed fetal bovine serum (Hyclone)]. Where indicated, cells were infected with freshly egressed parasites, washed three times with PBS then fixed at 2 hours or 24 hours post-infection. For ultrastructural analysis, cells were fixed in 2% paraformaldehyde/2.5% glutaraldehyde (Polysciences Inc., Warrington, PA) in 100 mM phosphate buffer, pH 7.2 for 1 hour at room temperature, processed and examined as described previously [91].
Immunofluorescence assays were performed in C57BL/6 bone marrow-derived macrophages. Bone marrow-derived macrophages for these experiments were derived as described previously [91]. Cells were activated with 100 U/ml IFN-γ and 0.1 ng/ml LPS in macrophage media lacking uracil. Macrophages were infected with freshly egressed parasites at an MOI of 1, washed at 3 hours post-infection five times with PBS, and incubated in uracil-free media supplemented with IFN-γ and LPS for the indicated time. Heat-killed parasites were incubated at 65°C for 10 minutes and infected at an MOI of 5. Cells for immunofluorescence were fixed in 4% formaldehyde, permeabilized with 0.05% saponin, and stained using primary antibodies as described. Parasite vacuoles were localized using mouse monoclonal Tg17-43 against GRA1 or rabbit polyclonal sera against GRA7. Host LAMP-1 was localized with rat monoclonal antibody 1D4B and Irgb6 was localized using rabbit polyclonal sera raised against recombinant protein [92]. All secondary antibodies used in immunofluorescence were highly-cross adsorbed Alexa Fluor conjugated antibodies (Invitrogen). Samples were visualized using a Zeiss Axioskop 2 MOT Plus microscope equipped for epifluorescence and using a 63× PlanApochromat lens, N.A. 1.40 (Carl Zeiss, Inc., Thornwood, NY). Images were acquired with an AxioCam MRm camera (Carl Zeiss, Inc.) using Axiovision v4.6, and processed using similar linear adjustments for all samples in Photoshop CS4 v9.
Bone marrow-derived macrophages were cultured with invasion-blocked or untreated mCherry-expressing cpsII parasites (MOI = 1) and LysoTracker Green DND-26 (Life Technologies) was added prior to imaging, following the manufacturer's instructions. Images were collected using a Leica DMI4000 microscope equipped with a Yokogawa CSU10 spinning disk confocal unit and a Hamamatsu ImagEM EMCCD camera. Images were analyzed using ImageJ software.
Statistical analysis was performed using PRISM software (Graphpad Software). Significance was calculated using an unpaired two-tailed student's t-test except when otherwise noted.
|
10.1371/journal.pcbi.1000814 | Nonlinearity of Mechanochemical Motions in Motor Proteins | The assumption of linear response of protein molecules to thermal noise or structural perturbations, such as ligand binding or detachment, is broadly used in the studies of protein dynamics. Conformational motions in proteins are traditionally analyzed in terms of normal modes and experimental data on thermal fluctuations in such macromolecules is also usually interpreted in terms of the excitation of normal modes. We have chosen two important protein motors — myosin V and kinesin KIF1A — and performed numerical investigations of their conformational relaxation properties within the coarse-grained elastic network approximation. We have found that the linearity assumption is deficient for ligand-induced conformational motions and can even be violated for characteristic thermal fluctuations. The deficiency is particularly pronounced in KIF1A where the normal mode description fails completely in describing functional mechanochemical motions. These results indicate that important assumptions of the theory of protein dynamics may need to be reconsidered. Neither a single normal mode nor a superposition of such modes yields an approximation of strongly nonlinear dynamics.
| Biological cells use a variety of molecular machines representing enzymes, ion channels or pumps, and motors. Motor proteins are nanometer-size devices generating forces and actively moving or rotating under the supply of chemical energy through ATP hydrolysis. They are crucial for many cell functions and promising for nanotechnology of the future. Although such motors represent single molecules, their operation cycles cannot be followed in detail in simulations even on the best modern supercomputers and some approximations need to be employed. It is often assumed that conformational dynamics of motor proteins is well described within a linear response approximation and corresponds to excitation of normal modes. We have checked this assumption for two motor proteins, myosin V and kinesin KIF1A. Our results show that, while both these biomolecules respond by well-defined motions to energetic excitations, these motions are essentially nonlinear. The effect is particularly pronounced in KIF1A where relaxation proceeds through a sequence of qualitatively different conformational changes, which may facilitate complex functional motions without additional control mechanisms.
| Protein machines, which may represent enzymes, ion pumps or molecular motors, play a fundamental role in biological cells and understanding of their activity is a major challenge. Operation of these machines is based on slow conformational motions powered by external energy supply, often with ligands (such as ATP). In molecular motors, binding of ATP and its subsequent hydrolysis induce functional mechanochemical motions, essential for their operation. These motions, which follow after an energetic activation, are conformational relaxation processes.
Large-scale conformational changes may take place in proteins as a result of ligand binding [1]. Despite the large magnitude of such changes, they are nonetheless often considered in the framework of the linear response theory [2] and the normal mode approximation [3]–[7]. The normal mode analysis is furthermore broadly employed in the elastic-network studies of proteins [7]–[18]. However, there is no general justification to assume that relaxation processes in proteins are linear and this assumption has to be verified for particular macromolecules.
It is known that relaxation processes in complex dynamical systems may be strongly nonlinear and deviate much from simple exponential relaxation. As an example borrowed from a distant field, we can mention the Belousov-Zhabotinsky reaction which exhibits a great variety of spatiotemporal patterns (pacemakers, rotating spiral waves) that are however only complicated transients accompanying relaxation to the equilibrium state [19], [20]. Conformational relaxation in single protein molecules may also be a complicated process, comprising qualitatively different kinds of mechanochemical motions.
While partial unfolding and refolding, associated with ligand binding, are known for some protein machines, such as the enzyme adenylate kinase [21], usually functional conformational motions in molecular machines and, specifically, in motor proteins are elastic. This means that the pattern of contacts between the residues in a protein is not changed upon ligand binding and preserved during the relaxation process, as generally assumed in the elastic network modeling (ENM).
Here, we provide detailed analysis of conformational relaxation processes, associated with ligand binding and hydrolysis, in two motor proteins — myosin V [22], [23] and kinesin KIF1A [24]. Our investigations, performed in the framework of the ENM approximation, reveal that nonlinearity is characteristic for both macromolecules and the normal mode description is not really applicable for any of them. For KIF1A, a monomeric motor protein from the kinesin superfamily, nonlinear effects are found to dominate completely functional mechanochemical motions which turn out to be qualitatively different from the normal mode predictions. Despite the nonlinearity, well-defined conformational relaxation paths, robust against perturbations, have been found in both motor proteins.
Within the coarse-grained ENM approach, a protein is modeled as a network of point-like particles, corresponding to residues, which are connected by a set of elastic links [9], [10]. A link between two particles is present if the distance between them in the equilibrium conformation of the considered macromolecule is shorter than a cutoff length. The elastic energy of the network is , where is the stiffness constant of the network links, is the matrix of connections inside the network, is the distance between particles and , and is the respective distance in the equilibrium reference state. The characteristic time scales of functional mechanochemical motions in motor proteins are in the millisecond range and slow conformational relaxation motions on such timescale should be overdamped [25]. Neglecting hydrodynamic interactions, relaxation dynamics is then described by equations for the coordinates of the particles, where is their mobility. Relaxation dynamics for elastic networks of proteins has been previously considered [26].
Despite a wide-spread misunderstanding, elastic dynamics is generally nonlinear. For example, macroscopic objects, such as ribbons or membranes, can still exhibit pronounced nonlinear effects of spontaneous twisting or buckling, while fully retaining their elastic behavior and not undergoing plastic deformations [27]. The energy of an elastic network is quadratic in terms of the distances and the forces acting on the particles are linear in terms of such distances. However, the distance is itself a nonlinear function of the coordinates and and this makes the forces also nonlinear functions of dynamical variables. The presence of nonlinear effects in conformational relaxation of proteins in the ENM approximation has been previously demonstrated [28], [29].
Explicitly, relaxation dynamics of considered proteins is described by equations (3) in the Methods section, where further details are also given. To study conformational relaxation, these equations were numerically integrated starting from various initial conditions.
The reference conformation, used to construct the elastic network, was that of the ATP(analog) bound state (Protein Data Bank (PDB) ID code: 1W7J, with MgADP-BeFx as the ATP analog [30]). As the initial condition, the conformation corresponding to the nucleotide-free state was taken (PDB ID: 1OE9 [31]). The elastic network had 855 particles connected by 7261 links. Note that only the residues whose -carbon positions are contained in both PDB data sets have been taken to construct the network. Additionally, relaxation processes starting from randomly generated initial conditions (see Methods) have been considered. For visualization purposes, motions of three particles (Asp122 in chain A, and Val22 and Ser135 in chain B) have been traced (Figure 1A). Thus, each relaxation process was characterized by a certain trajectory in the space of distances between the three chosen labels.
Figures 1B,1C display 100 conformational relaxation trajectories, each starting from a different random initial condition. Although the initial conditions were generated by applying relatively strong deformations (without unfolding) to the reference state, almost all of them were leading back to that reference state, with just a few metastable states found. Furthermore, one can observe that the trajectories converge to a well-defined relaxation path.
The red trajectory in Figures 1B,1C is for the relaxation starting from the nucleotide-free conformational state of myosin V (so that the mechanochemical motion following ATP binding is simulated). After a transient, this special trajectory joins the well-defined relaxation path. This functional trajectory is robust against perturbations, as shown by Figure 1D. Several snapshots of the conformation along this trajectory are shown in Figure 2 (see also Video S1).
The attractive path corresponds to a deep energy valley in the energy landscape of myosin V. Once this valley is entered, the conformational relaxation motion becomes effectively one-dimensional and characterized by a single mechanical coordinate. The profile of the elastic energy along the bottom of such energy valley determines the dependence of the elastic energy on the collective mechanical coordinate (see Methods).
Figure 3A shows the dependence of the elastic energy along the special attractive relaxation path starting from the nucleotide-free state and leading to the ATP-bound state. Markers indicate positions along the trajectory in Figure 1C. For , the elastic energy is approximately quadratic in terms of the mechanical coordinate , i.e. . Because , this implies that then and the relaxation is exponential. Only within such harmonical neighborhood of the reference state, the normal mode description becomes applicable (see Methods for further discussion).
The dotted blue line in Figure 1D shows the direction of the distance changes corresponding to the slowest normal mode (see Methods). The nucleotide-free state of myosin V lies away from this direction and also outside of the harmonical neighborhood. The initial stage of the functional mechanochemical motion (until time ) cannot be quantitatively analyzed in terms of the normal modes.
The reference conformation for KIF1A is the ADP-bound state (PDB ID: 1I5S, with MgADP [32]). Relaxation starting from the initial condition, corresponding to the ATP(analog)-state (PDB ID: 1I6I, with MgAMPPCP as an ATP analog [32]) and from randomly generated initial conditions was considered. The elastic network has 320 particles and 2871 links. Only the residues whose -carbons are in both PDB data sets have been used. Visualization labels are Glu233, Ala286, and Asn211 (Figure 4A).
100 relaxation trajectories, starting from random initial conditions, are shown in Figures 4B,4C. The presence of an attractive relaxation path, corresponding to a deep energy valley, can be noticed.
The red lines in Figures 4B,4C display the special relaxation trajectory starting from the ATP-bound state. Surprisingly, we find that, in contrast to myosin V, this trajectory is different from the typical relaxation path. By applying small random initial perturbations to the initial ATP-bound state and integrating the dynamical equations, it can be demonstrated that this trajectory is, however, also stable with respect to the perturbations (Figure 4D). The dotted blue line in Figure 4D shows the direction of the distance changes in the slowest normal mode of KIF1A.
Thus, in KIF1A the deep energy valley leading to the reference ADP-bound state gets branched at some distance from it. The path corresponding to the functional mechanochemical motion from the ATP-bound state belongs to the side branch. Only at the final relaxation stage, in the immediate vicinity of the equilibrium, the valleys merge and the functional motion begins to coincide with the typical relaxation motion in this protein.
The branching of the energy valley is already an indication of strong nonlinearity in the relaxation dynamics. We have also determined the profile of the elastic energy as a function of the mechanical coordinate along the path connecting the ATP- and ADP-bound states (Figure 3B). The profile becomes quadratic only starting from time , very close to the equilibrium reference state.
Figure 5 shows snapshots of KIF1A along the special attractive relaxation path (see also Video S2). At the early relaxation stage (until ), the relaxation motion represents a combination of the rotation of the switch II helix and of the sliding of the switch I loop. Relaxation at the end of such initial stage is apparently hindered, as revealed in the presence of a plateau in the dependence of the elastic energy on the mechanical coordinate in Figure 3B near . Only once the sliding is completed, further local structural reorganization, representing a transition from the loop to the -helix, becomes possible and is indeed observed approximately after time .
The normal mode description is broadly used in structural studies of proteins. The analysis of thermal fluctuations and the interpretation of the respective experimental structural data are traditionally performed assuming that fluctuations are linear and, hence, correspond to thermal excitation of various normal modes (see, e.g., [3], [4]). The linear response of a protein macromolecule to structural perturbations, such as ligand binding, is an often used assumption [2]. To a large extent, the elastic-network analysis of ligand-induced macromolecular motions is based on determining normal modes in the elastic networks of the considered proteins (see, e.g., [7]). The patterns of atomic displacements in such normal modes are further compared with the experimentally measured atomic displacements in the same proteins that are induced by a change of the chemical state, such as binding of an ATP molecule [7], [14]–[17]. Large overlaps with only a few slowest normal modes are seen as the evidence for the applicability of the elastic-network ansatz, whereas the wide distribution of overlaps is considered as the indication that the elastic network description fails for a particular macromolecule. Specifically, strong overlaps between ligand-induced conformational changes and atomic displacements in the few slowest modes have been found for scallop myosin and -ATPase, while such overlaps were absent for kinesin KIF1A [18].
Our numerical investigations of elastic conformational motions in two motor proteins (myosin V and KIF1A) have revealed however that in both of them the nonlinearities play an essential role. While slow conformational relaxation motions in myosin can still be qualitatively characterized in terms of the normal modes, the normal mode description breaks down completely for KIF1A. The observed breakdown of the normal mode description does not however mean that conformational motions become irregular. We have seen that ordered and robust mechanochemical motions are characteristic for both protein motors, even though they cannot be described in terms of the linear response.
We want to emphasize that, when the dynamics is nonlinear, neither a single normal mode, nor a combination of many such modes can reproduce the motions. Thus, the normal mode description fails completely in this case and the problem is not that many normal modes must be taken into account. Actually, as we have shown, even for KIF1A, one normal mode would be sufficient to describe long-time relaxation within the harmonic domain — however, this domain is restricted to a tiny neighbourhood of the equilibrium state.
Thermal fluctuations have not been explicitly included into our dynamical ENM simulations. However, such fluctuations are effectively generating random conformational perturbations. In our study, relaxation processes starting from random conformational perturbations have indeed been considered.
In myosin V, one well-defined nonlinear conformational relaxation trajectory, leading to the equilibrium state, has been identified. Starting from an arbitrary initial conformation (but still without unfolding), rapid convergence to this special trajectory takes place. While the motion corresponding to the special attractive trajectory is initially nonlinear, it becomes harmonical later and a substantial part of the ordered conformational relaxation process is within the harmonic domain of the equilibrium state. Similar behavior has been previously noted by us [29] for scallop myosin and -ATPase, but its detailed analysis has not yet been performed.
The situation is more complex for the monomeric kinesin KIF1A. Instead of a unique deep energy valley leading to the reference ADP-bound state, two such valleys, both leading to the equilibrium state, are present. These valleys correspond to two kinds of ordered conformational motions possible in the protein.
The first of them is relatively wide and, when thermal conformational fluctuations are excited, they would typically proceed along it. However, the conformational relaxation motion starting from the ATP-bound state follows a different path, which corresponds to the second energy valley branching from the typical fluctuation path already at very small deviations from the equilibrium state. Note that the branching takes place as the change in the distance between the molecular labels Glu233 and Ala286 is still less than an angstrom, which is much smaller than the intensity of typical thermal fluctuations for such a distance. Thus, the nonlinear effects in KIF1A are strong even for the typical thermal fluctuations.
Remarkably, such second relaxation path is also stable with respect to perturbations, i.e. structurally robust. Our numerical investigations reveal that motion along this path can be divided into two qualitatively different stages. At the first of them, sliding of the switch I loop is observed, whereas at the second stage a transition from the loop to the -helix is realized. Structural reorganization, corresponding to this transition, is not possible until the sliding motion is completed, lifting a restriction through the backbone chain. Recent crystallographic studies suggest that the switch I loop/helix plays an important role in control of the motor function through interaction with and switch II [33].
Thus, in contrast to myosin, a single ATP binding event induces in KIF1A a complex, but ordered conformational motion characterized by two qualitatively different consequent phases. As we conjecture, this special dynamical property of KIF1A may be needed for the processive motion of this single-headed molecular motor [24].
In myosin V, conformational motions driven by random thermal fluctuations are similar in their properties to the relaxation motion from the nucleotide-free state. This may facilitate exploitation of such fluctuation motions for the motor operation, as suggested by recent single-molecule experiments [34]. In KIF1A, where the energy valley splits into two branches, typical thermal conformational fluctuations are qualitatively different from the relaxation motion starting from the previous ATP-bound state. The latter motion is entropically hindered for thermal fluctuations and cannot be reversed through them. This may turn out to be important for the understanding of the operation of the monomeric kinesin as a molecular motor. Latest experimental techniques permit simultaneous observation of stepping motion and conformational changes of a motor [35]. The coarse-grained modeling, including our present study, can contribute further suggestions for the design (e.g., by determining positions for fluorescent labeling) of such experiments.
Finally, we note that our study has been based on the elastic network approximation for proteins. More detailed descriptions, such as, e.g., G-like models, can also be used to consider conformational relaxation processes [36]. We expect that similar behavior will then be observed.
In this study, we employ elastic network models where material points are connected by a set of elastic springs [8]–[11]. Each particle corresponds to a residue in the considered protein. The equilibrium positions of the particles are determined by the locations of -carbon atoms in the reference state of the protein, taken from the PDB database. Two particles in a network are connected by an elastic spring if at equilibrium the distance between them is less than a certain cutoff length . The natural length of an elastic link is equal to the equilibrium distance . The cutoff distance has been used in our study.
The elastic forces obey the Hooke law and all springs have the same stiffness constant . Elastic torsion effects are not included. Thus, the force acting on particle is(1)where is the total number of particles in the network, is the actual position of the particle and is the actual distance between two particles and . The adjacency matrix of the network is defined as having if and otherwise. The total elastic energy of the network is(2)
Because slow conformational dynamics of proteins in the solvent is considered, the motions are overdamped (see [25]) and the velocity of a particle is proportional to the force acting on it, i.e. where is the mobility. We assume that the mobilities of all particles are the same. Hydrodynamical effects are neglected (they can be however incorporated into the elastic network models as shown in ref. [37]).
Explicitly, the relaxation dynamics is described by a set of differential equations:(3)Here, time is rescaled and measured in units of . Hence, the relaxation dynamics of a network is completely determined by its pattern of connections (matrix ) and the equilibrium distances between the particles. Equations (3) were integrated to determine conformational relaxation motions.
To prepare random initial conditions, the following procedure has been employed. Random static forces , acting on all particles in the network have been independently generated with the constraint that . The equations of motion were integrated in the presence of such static forces for a fixed time . The conformation which was thus reached has been then used as the initial condition for the relaxation simulation. The parameters were , for myosin V and , for KIF1A. With these parameter values, relatively large overall deformations ( typical) could be reached, while still avoiding unfolding. In the deformed states, the lengths of the links did not exceed for myosin V and for KIF1A.
When relaxation from specific conformations has been considered, initial positions of all particles were allocated according to the respective PDB structures. When robustness of a relaxation path starting from a specific conformation was investigated, the initial condition was prepared by randomly shifting the positions of all particles with respect to their locations in that conformation with a certain root-mean-square displacement . We have chosen for myosin V and for KIF1A.
To visualize conformational motions, three particles labeled as , and were chosen and the distances and were monitored in the simulations. Thus, the relaxation motion was represented by a trajectory on the plane .
The choice of the visualization labels is essentially arbitrary. In a simulation, motions of all residues were traced (see, e.g., Videos S1 and S2) and different residues could be selected for a specific visualization. If a molecule has a low-dimensional attractive relaxation manifold, this is a property of the respective dynamical system and it cannot depend on the visualization method. When selecting the labels, one should only pay attention to the fact that the distances between them should significantly vary during the relaxation process. If, by chance, two labels belonging to the same stiff domain in a protein have been taken, the distance between them would remain almost constant, so that such a choice would be inconvenient. When the normal mode description approximately holds and, furthermore, relaxation is well described by a few slowest modes, one can choose the labels so that the distances between them reveal variations characteristic for the first few normal modes. Such selection was previously made [29] for scallop myosin and -ATPase, and it has been adopted in the present study for myosin V. For KIF1A, the labels have been chosen in such a way that motions in switch I and switch II regions are well resolved.
The collective mechanical coordinate along a relaxation path was defined by requiring that its dynamics obeys the equation and that as . Multiplying both parts of this equation by , we find that it is equivalent to the equation(4)Equation (4) can be used to determine the coordinate along a given relaxation trajectory and the dependence of the elastic energy on this coordinate.
For each point along the trajectory, the time when it is reached in the relaxation process is known. Moreover, the actual network configuration corresponding to this point is also known from the simulation. Therefore, for each point specified by time the respective elastic energy is determined. The mechanical coordinate , reached at time , is given by the integral(5)
We provide a summary of the results on the normal mode description of conformational relaxation processes. If deviations from the reference conformation are small for all particles, the nonlinear equations (3), describing conformational relaxation of an elastic network, can be linearized:(6)where .
Equations (6) can be written in the matrix form as(7)where is the linearization matrix:(8)where .
The general solution of these linear differential equations is given by a superposition of exponentially decaying normal modes, i.e.(9)Here, and are the eigenvalues and the eigenvectors of the linearization matrix, i.e.(10)This matrix has eigenvalues, but 6 of them must be zero, corresponding to free translations and rotations of the entire network.
Generally, all normal modes are initially present. As time goes on, first the normal modes with the larger eigenvalues decay. In the long time limit, relaxation is characterized by the soft modes corresponding to the lowest eigenvalues.
Figure 6 shows the computed eigenvalue spectra of myosin V and KIF1A. The eigenvalues are normalized to the lowest nonzero eigenvalue and the logarithmic representation is chosen.
Note that in both motor proteins a significant gap, separating the soft mode from the rest of the spectrum, is present. This means that, in the linear approximation, long-time relaxation in these proteins is effectively characterized by a single degree of freedom, representing the amplitude of the soft mode. The pattern of displacements of particles (i.e., residues) from the reference positions is determined by the eigenvector of the soft mode.
In the plane of the distances between the labels , and , used by us for the visualization of conformational motions, the exponential relaxation motion corresponding to the soft mode should proceed along the direction defined by the vector with the components and . Such directions are indicated by dotted blue lines in Figure 1D and Figure 4D.
When relaxation is reduced to a single soft mode, the elastic potential is quadratic in terms of the mechanical coordinate, i.e. .
Note that the representation of the relaxation process as a superposition (9) of normal relaxation modes holds only in the harmonic domain, i.e. when linearization (6) of full nonlinear relaxation dynamics equations (3) is valid. If dynamics is nonlinear and the linearization does not hold, relaxation dynamics cannot be viewed at all as a superposition of any normal modes. Whether just one normal mode or many of them should be included into a description of long-time relaxation dynamics is determined by the properties of the eigenvalue spectrum and not related to the possible invalidity of the harmonic approximation.
As an extension, iterative normal mode analysis has been proposed [21], [38]. This method is applied to obtain an optimal sequence of conformational states, transforming an initial given conformation into a target conformation, which may be known with a low resolution or only partially, and thus to reconstruct missing details of that structure. Each next conformation in the sequence is obtained by making a step into the direction maximizing similarity with the target, restricted however to a superposition of a certain number of the lower normal modes. At the next iteration step, the previous conformation is chosen as a new reference state and a new set of normal modes is determined. This prediction method is useful and provides valuable results, e.g., in the refinement of low-resolution structures from electron microscopy [38]. It should be however emphasized that the sequence of conformational states yielded by such a method is generally different from the path along which conformational relaxation from the target to the reference state would proceed. Even in the normal mode approximation, dynamics of conformational relaxation depends not only on the eigenvectors, but also — and very significantly — on the eigenvalues of normal modes. Generally, the next iteration state in this method would not be the next conformation along the actual relaxation path. This difference can be clearly demonstrated by considering the example of KIF1A. The conformational relaxation path transforming the initial ATP-bound state into the (equilibrium) ADP-bound state is non-monotonous (Figure 4). It proceeds via intermediate states (particularly of the switch I region) which cannot be obtained by gradual interpolation maximizing similarity of the structures along the optimization path.
|
10.1371/journal.ppat.1002640 | Airborne Signals from a Wounded Leaf Facilitate Viral Spreading and Induce Antibacterial Resistance in Neighboring Plants | Many plants release airborne volatile compounds in response to wounding due to pathogenic assault. These compounds serve as plant defenses and are involved in plant signaling. Here, we study the effects of pectin methylesterase (PME)-generated methanol release from wounded plants (“emitters”) on the defensive reactions of neighboring “receiver” plants. Plant leaf wounding resulted in the synthesis of PME and a spike in methanol released into the air. Gaseous methanol or vapors from wounded PME-transgenic plants induced resistance to the bacterial pathogen Ralstonia solanacearum in the leaves of non-wounded neighboring “receiver” plants. In experiments with different volatile organic compounds, gaseous methanol was the only airborne factor that could induce antibacterial resistance in neighboring plants. In an effort to understand the mechanisms by which methanol stimulates the antibacterial resistance of “receiver” plants, we constructed forward and reverse suppression subtractive hybridization cDNA libraries from Nicotiana benthamiana plants exposed to methanol. We identified multiple methanol-inducible genes (MIGs), most of which are involved in defense or cell-to-cell trafficking. We then isolated the most affected genes for further analysis: β-1,3-glucanase (BG), a previously unidentified gene (MIG-21), and non-cell-autonomous pathway protein (NCAPP). Experiments with Tobacco mosaic virus (TMV) and a vector encoding two tandem copies of green fluorescent protein as a tracer of cell-to-cell movement showed the increased gating capacity of plasmodesmata in the presence of BG, MIG-21, and NCAPP. The increased gating capacity is accompanied by enhanced TMV reproduction in the “receivers”. Overall, our data indicate that methanol emitted by a wounded plant acts as a signal that enhances antibacterial resistance and facilitates viral spread in neighboring plants.
| The mechanical wounding of plant leaves, which is one of the first steps in pathogen infection and herbivore attack, activates signal transduction pathways and airborne signals to fend off harmful organisms. The mechanisms by which these signals promote plant immunity remain elusive. Here, we demonstrate that plant leaf wounding results in the synthesis of a cell wall enzyme, pectin methylesterase (PME), causing the plant to release methanol into the air. Gaseous methanol or vapors from wounded PME-transgenic plants induced resistance to the bacterial pathogen Ralstonia solanacearum in the leaves of non-wounded neighboring “receiver” plants. To investigate the mechanism underlying this phenomenon, we identified the methanol inducible genes (MIGs) in Nicotiana benthamiana, most of which fell into the category of defense genes. We selected and isolated the following genes: non-cell-autonomous pathway protein (NCAPP), β-1,3-glucanase (BG), and the previously unidentified MIG-21. We demonstrated that BG, MIG-21 and NCAPP could enhance cell-to-cell communication and Tobacco mosaic virus (TMV) RNA accumulation. Moreover, gaseous methanol or vapors from wounded plants increased TMV reproduction in “receivers”. Thus, methanol emitted by a wounded plant enhances antibacterial resistance as well as cell-to-cell communication that facilitate virus spreading in neighboring plants.
| Plants are exposed to a diverse range of abiotic and biotic stresses [1]–[3]. Physical damage to a plant is a potential threat because it provides an opportunity for pathogen entry. Localized tissue damage elicits the expression of an array of antimicrobial phytochemicals [4], proteins [5], and systemic defense responses against microbial pathogens [6], [7] and herbivore attack [1], [8]–[14]. Systemic defense responses provide an attractive model for the study of cell-to-cell signal transduction pathways that operate over long distances [15], [16]. The molecular mechanisms of systemic wound signaling are not yet fully understood, but several of the non-cell autonomous signals that are released from damaged cells have been studied. In response to pathogen attack or physical damage, several plant species emit volatile organic compounds (VOCs), including ethylene [17], methyl salicylate [18], methyl jasmonate [19], [20], nitric oxide [21], [22] and cis-3-hexen-1-ol [23], which upregulate pathogen-related (PR) genes [14], [23], [24].
Pectin methylesterase (PME, EC: 3.1.1.11) [25] is a PR protein [26] and is the first barrier of defense against invading pathogens [26]–[31] and herbivores [32]–[34]. In higher plants, PME is a ubiquitous multifunctional enzymatic component of the plant cell wall (CW). The PME gene encodes a pro-PME precursor with an N-terminal extension of variable length [35]–[37]. The tobacco pro-PME contains a long N-terminal leader with a transmembrane domain, which is important for PME delivery into the CW [37], [38]. PME participates in CW modulation during general plant growth [39]–[42], nematode infection [43] and pollen tube growth [44]–[47].
PME interacts with the movement protein of the Tobacco mosaic virus (TMV) [48], [49], suggesting that PME may be involved in the cell-to-cell movement of plant viruses [50]. PME also efficiently enhances virus- and transgene-induced gene silencing (VIGS and TIGS) via the activation of siRNA and miRNA production [51], [52]. In the case of bacterial and fungal phytopathogens, PMEs act as virulence factors that are necessary for pathogen invasion and spreading through plant tissues [53]. The general structure of plant PME is very similar to that of the enzymes produced by phytopathogens [54]. Due to this structural similarity, transgenic plants overexpressing PME can be used as a model of host responses to pathogenic attack. A transgenic tobacco plant (Nicotiana tabacum L.) expressing a fungal PME exhibited a dwarf phenotype, modified CW metabolism [40] and a two-fold increase in leaf sap methanol levels.
The pectin demethylation directed by PME is likely to be the main source of methanol, which has long been assumed to be a metabolic waste product [55]–[57]. Methanol can accumulate in the intercellular air space at night after the stomata have closed [57]. Methanol emission peaks have been observed in the morning, when the stomata open [58]. Wounding and herbivore attack increase methanol emission levels [34], [59]–[61]. Transgenic plants with a silenced PME gene had a 50% reduction of PME activity in their leaves and a 70% reduction of methanol emissions compared with wild type (WT) plants. This result demonstrates that herbivore-induced methanol emissions originate from pectin demethylation by PME [33]. However, there is no direct evidence that de novo synthesized PME participates in methanol synthesis. In a study of VOC emissions from Nicotiana attenuata plants attacked by Manduca sexta larvae [34], [60], methanol was detected in the headspace of leaves very quickly (10 min) after leaf wounding. Therefore, it was concluded that the methanol detected was produced by PME that had been deposited in the CW before the leaf damage occurred.
To investigate the metabolism of methanol in plants, Downie et al. [62] used foliar sprays to apply methanol stimulation to Arabidopsis thaliana and studied the resulting changes in gene expression in leaves harvested 1, 24, and 72 h after methanol treatment using a 26,090 element oligonucleotide microarray. A concentration of 10% (v/v) methanol containing Silwet surfactant was used, to expose plants to a methanol concentration in essential excess of endogenous levels. A total of 484 (1.9%) transcripts were shown to be regulated in response to the methanol treatment. A group of genes encoding detoxification proteins, including cytochrome P450s, glucosyl transferases and members of the ABC transporter family, was the most strongly regulated group. Those authors concluded that a foliar spray of 10% methanol affects the expression of hundreds of genes, activating multiple detoxification and signaling pathways.
Here, we show that wounding results in drastic de novo PME synthesis. The analysis of methanol in plant emissions presents serious technical challenges. To avoid the underestimation of methanol emissions, we developed a method of methanol registration based on the high solubility of methanol in water [59]. The usage of water traps in a hermetically sealed water-drop system and a flow-through system revealed a 20-fold increase in the emission of gaseous methanol 180 min after leaf injury. To clarify the role of methanol in antibacterial resistance, we examined plant susceptibility to infection with Ralstonia (Pseudomonas) solanacearum [63], which causes wilt. Bacterial wilt is a devastating plant disease that affects several economically important hosts, including potatoes, tomatoes, bananas, and tobacco [64].
We showed that the methanol emitted by wounded and PME transgenic plants induced antibacterial resistance in non-wounded neighbor plants. We then identified more than three hundred methanol inducible genes (MIGs) that were upregulated in methanol-treated N. benthamiana plants. We further studied the function of three abundant MIGs: β-1,3-glucanase (BG), non-cell-autonomous pathway protein (NCAPP), and a previously unknown gene, designated MIG-21. Quantitative real-time PCR (qPCR) analysis of mRNA from a plant treated with gaseous methanol confirmed changes in the expression of these MIGs and revealed a specific “wave” of MIG mRNA accumulation. The wave of MIG mRNA accumulation consisted of a peak followed by attenuation. We also showed that methanol and the selected MIGs (NCAPP, MIG-21, and β-1,3-glucanase) induced an increase in the plasmodesmata (Pd) size exclusion limit (SEL). This was demonstrated in experiments using two tandem copies of green fluorescent protein (GFP) (2×GFP) as an indicator of Pd SEL. In addition to methanol-induced Pd gating, we also observed enhanced TMV reproduction in methanol-exposed plants and in neighbors of PME-transgenic and wounded plants. We hypothesize that methanol-mediated MIG upregulation and enhanced viral reproduction are unintended consequences of plant mobilization against bacterial pathogens.
Leaf wounding is often used as an experimental model of mechanical injuries sustained by a plant after wind, rain, hail, or herbivore feeding. However, serious leaf damage caused by, for example, crushing the leaf lamina with forceps [65] or puncturing leaves [34] has only a mild effect on PME gene expression. In nature, pathogen penetration of leaf tissue can occur via microdamage to the leaf cuticle, trichome or CW. Microdamage can be induced by wind-mediated leaf rubbing or insect attack. To test whether the expression of the endogenous PME gene is modulated by external mechanical stress, we rubbed N. benthamiana leaves with an abrasive water suspension of Celite. This approach is commonly used for plant virus inoculation. The 1.7-kb PME transcript was not detectable in intact leaves but was clearly induced after Celite rubbing at 1 hpi and was increased at 8 hpi (Figure 1A). TMV inoculation increased the accumulation of PME transcripts, suggesting a role for viral infection in increased PME mRNA levels.
We wanted to evaluate the effect of enhanced PME mRNA accumulation on methanol emission. We hypothesized that methanol from wounded leaves is produced by two forms of PME: pre-existing PME deposited in the CW before wounding, which allows rapid methanol release [34], [60], and PME synthesized de novo after wounding (Figure 1A), which likely generates methanol for an extended period (more than 8 h). Until now, the quantification of methanol emission by a plant leaf was conducted using methods based on the detection of gas-phase methanol [34], [57]–[61]. Methanol is a polar, soluble compound that is easily lost due to condensation in sampling lines and traps. Methanol mixes readily with water, a property that we exploited by using water as a trap for methanol measurement. The methanol released by wounded leaves was measured in the headspace of either a hermetically sealed jar (the water-drop system) (Figure 1B) or a glass flow chamber (Figure 1C). A drop of methanol added to the bottom of the jar will vaporize rapidly and dissolve in the water. The methanol content in the water phase may thus be used to estimate the methanol content of the leaf headspace. In the reconstruction experiment, we measured the methanol content in the water drop at different times following evaporation of various quantities of methanol that had been added to the jars. At 24°C, methanol was detected in the water drop 30 min after its addition. The water drop reached more than 80% of its saturation point after 3 h. Using calibration curves and the previously determined methanol recovery correction factors, we calculated the methanol emission of wounded leaves. Leaf wounding resulted in gaseous methanol emission, which was 20-fold higher than the methanol emission by the control intact leaf at 3 h of incubation (Figure S1). The water-drop and flow-through approaches yielded similar results for methanol emission after wounding (Figure 1D). Analysis using the unpaired two-tailed Student's t-test confirmed a statistically significant difference in methanol emission between the control leaves and the wounded leaves.
To determine whether methanol reabsorption might complicate our analysis, we measured the methanol content in the sap of control leaves and wounded leaves. No statistically significant methanol increase in leaf sap was detected (Figure 1E). This result indicates that essentially all methanol generated by the wounded leaves was emitted into the air.
Collectively, our data show that leaf wounding causes a rapid increase in the production of gaseous methanol.
Biologically, wound-induced PME gene expression and the subsequent methanol emission should lead to increased resistance to pathogens, including pathogenic bacteria [26]. To determine whether the methanol emitted by wounded plants serves as a signal for antibacterial resistance, we developed an approach (Figure 2A) in which a wounded N. benthamiana plant (an “emitter”) was placed in a hermetically sealed 20-l desiccator along with an intact N. benthamiana “receiver” plant. The “receiver” plant, which had been stored adjacent to the “emitter” plant, was removed from the desiccator, and its leaves were injected with a suspension of R. solanacearum, which infects a wide range of host plants. Because both whole plants were confined within the sealed container, the available CO2 may have been depleted. Because CO2 depletion could cause several types of stress, we also tested for bacterial growth in the “receiver” plant stored together with an intact plant. Figure 2B shows that, as expected, incubation with the wounded “emitter” plants led to decreased R. solanacearum growth in the “receiver” plants (diagram bar #3) compared with control plants (diagram bar #1). In control experiments, methanol evaporating from a piece of methanol-soaked filter paper also suppressed bacterial growth (diagram bar #4). We also tested N. tabacum as a “receiver” and confirmed bacterial growth suppression (data not shown).
We also examined whether green-leaf VOC (GLV) emission, which is known to be enhanced by plant wounding [34], [66]–[68], suppressed bacterial growth [69]. GLVs are lipoxygenase metabolic pathway products that include six-carbon aldehydes and alcohols. Unlike terpenoids, GLVs are rapidly, immediately and likely passively released from wounded leaves [70], [71]. Our gas chromatography (GC) analysis confirmed the presence of methanol (Figure 3A). In line with the data of von Dahl et al. [34], our GC analysis revealed that cis-3-hexen-1-ol is emitted in the headspace of wounded leaves (Figure 3C). We did not detect methyl salicylate or methyl jasmonate in the headspace of wounded leaves (data not shown). Ethylene emission was detected, but there was no statistically significant difference in ethylene emission between the control and wounded leaves (Figure 3B).
Thus, the suppression of R. solanacearum growth observed in the “receiver” plants could be caused by gaseous methanol or by GLV. Indeed, cis-3-hexen-1-ol evaporated in the desiccator also resulted in decreased bacterial growth in target plants (Figure 2B, diagram bar #5). However, GLVs rapidly released from wounded leaves may stimulate PME-generated methanol production, and their influence on bacterial growth may thus be indirect. To examine the role of cis-3-hexen-1-ol in the emission of methanol from leaves, we measured the methanol content in a water trap system in which an N. benthamiana leaf was exposed to continuous airflow from an evaporator containing cis-3-hexen-1-ol for 3 h (Figure 4, upper). The diagram (Figure 4, bottom) shows that the methanol content in the water trap increased after cis-3-hexen-1-ol treatment. We suggest that methanol emission induced by GLV may be responsible for the suppression of R. solanacearum growth in “receivers”.
To further refine the role of methanol, GLV was excluded from the gaseous mixture emitted by wounded leaves. We used the previously engineered PME-transgenic tobacco line, pro1, which has increased PME gene expression and resistance to TMV [52]. Transgenic plants produced higher levels of methanol in the leaf sap than did the control plants (Figure S2). Consistent with our expectations, the increased PME gene expression in the transgenic plants also resulted in a higher production of gaseous methanol, whereas cis-3-hexen-1-ol was not detected (Figure 3C). To determine whether the methanol emitted by PME-transgenic plants serves as a signal for antibacterial resistance, we employed a hermetically sealed desiccator. Incubation with the PME-transgenic “emitter” plants slowed the growth of R. solanacearum compared to the control plants (Figure 2B, diagram bar #2). Although the retardation of R. solanacearum growth caused by a neighboring PME-transgenic plant was less than that caused by a wounded plant, the reduction in growth correlated with the level of methanol emission (Figure 3A). The unpaired two-tailed Student's t-test confirmed the statistical significance of the differences in R. solanacearum growth retardation between the “receivers” of control and PME transgenic plant (Figure 2B, diagram bar #2).
To further clarify the role of methanol as an airborne signal of antibacterial resistance, we again used a flow-through system that allows continuous airflow from PME-transgenic or wounded tobacco plants to intact target N. benthamiana plants (Figure 5A). Control plants were exposed to air from a desiccator containing intact N. tabacum plants. After exposure, the target “receiver” plants were inoculated with R. solanacearum. “Receiver” plants exposed to air from the desiccator with evaporated methanol, PME-transgenic or wounded plants acquired antibacterial resistance (Figure 5B). The evaporation from the wounded PME-transgenic plants had even greater effect on antibacterial activity. The unpaired two-tailed Student's t-test confirmed the statistical significance of the differences in decreased R. solanacearum growth.
Collectively, these data indicate that gaseous methanol is an airborne factor that may induce antibacterial resistance in neighboring plants.
In an effort to understand the mechanisms by which methanol can stimulate antibacterial resistance in “receiver” plants, we constructed forward and reverse suppression subtractive hybridization (SSH) cDNA libraries from N. benthamiana plants exposed to methanol.
A total of 359 differentially expressed transcripts were identified; of these, 39 appeared to be more abundant in intact leaves, and 320 appeared to be upregulated after methanol treatment (Table S1). The cloned ESTs of genes that responded to the methanol treatment were considered for sequencing. The EST sequences of the upregulated genes were deposited in the NCBI dbEST database with accession numbers. Most of the ESTs identified (Table S1) (i.e., 167) fell into the category of stress gene transcripts. We identified only one novel EST (FN432041), methanol-inducible gene-21 (MIG-21) (GenBank AC GU128961), which was unrelated to all other nucleotide sequences in GenBank. MIG-21 contains an ORF encoding a protein with a repetitive amino acid sequence (Figure S3). The methanol-specific upregulation of the SSH-identified genes was validated by a Northern blot analysis hybridized with 32P-labeled probes, which were prepared from randomly selected differential clones that were found by differential screening. We selected and isolated the most abundant SSH-identified genes for further analysis (Table 1) [72]–[76]. We validated the changes in gene expression observed by SSH by performing quantitative real-time PCR (qRT-PCR) to determine the mRNA levels from a plant treated with gaseous methanol. MIG mRNA accumulation depended on both the methanol concentration (Table 2) and the length of treatment (Figure S4). PME is not likely to be a MIG because its mRNA accumulation was not significantly altered after methanol treatment. The level of BG mRNA accumulation increased with time, up to 400-fold after 18 h (compared with the untreated control, Table 2). The accumulation of the NCAPP mRNA (GenBank AC FN432039) increased by almost 50-fold at 18 h and the level of NCAPP mRNA accumulation was the highest compared to BG and MIG-21 after 6 h of treatment (Figure S4).
Our model proposes that a burst of methanol from wounded leaves should elicit an extended MIG induction in neighboring leaves. We exposed N. benthamiana plants to methanol vapors (160 mg) applied to filter paper within a sealed 20-l desiccator for 3 h. RNA for qRT-PCR analysis was isolated from leaves at different times after the plant was withdrawn from the methanol atmosphere. Figure 6 shows the decaying wave of MIG mRNA accumulation after methanol treatment. MIGs mRNA accumulation reached a maximum at 24 h after methanol treatment and decreased slowly thereafter. Moreover, increased BG and NCAPP mRNA levels were observed as long as 5 days after methanol treatment.
The suppression of R. solanacearum growth in “receiver” plants in a sealed desiccator (Figure 2B) suggests that MIGs may be involved in plant antibacterial resistance. We examined the accumulation of MIGs mRNA in N. benthamiana “receivers” that were kept together with wounded WT or PME-transgenic tobacco plants in a sealed desiccator (Figures 7 A,B). The unpaired two-tailed Student's t-test confirmed the statistical significance of the differences in MIG induction between the “receivers” of control and wounded or PME-transgenic plants.
Constant PME expression and increased methanol production in PME-transgenic tobacco was predicted to result in increased MIG mRNA accumulation. Indeed, RNA analysis of PME transgenic leaves (Figure 8) confirmed this expectation, though the general profile was different from that of methanol-treated plants. The nearly 70-fold increase observed in PI-II mRNA accumulation is likely to be a response to long-term PME overproduction.
It has been demonstrated previously [23] that several plant species emit VOCs, including ethylene, methyl salicylate, methyl jasmonate, and cis-3-hexen-1-ol, in response to pathogen attack and plant damage. In “receiver” plants, the emitted VOCs can upregulate PR genes, such as the basic type PR-3 (chitinase), acidic type PR-4 (thaumatin-like), lipoxygenase (LOX), phenylalanine ammonia-lyase (PAL), and farnesyl pyrophosphate synthetase (FPS). We studied gene expression in plants treated with methanol and compared those results to the gene expression of plants treated with the VOCs listed above. As shown in Figure S5, the expression of LOX, PR-3, PR-4, FPS and PAL genes increased slightly in methanol-treated plants. Treatment with cis-3-hexen-1-ol stimulated the accumulation of FPS mRNA, but ethylene, methyl salicylate, and methyl jasmonate treatment primarily upregulated the PAL and PR-4 mRNAs accumulation.
Thus, the methanol emitted from a wounded plant most likely potentiates the antibacterial resistance of neighboring plants by increasing the MIG mRNA accumulation.
Bacterial pathogens do not cross plant cell wall boundaries because they inhabit the intercellular spaces in plants. In contrast, viral pathogens require intercellular movement for local and systemic spread [16]. However, plasmodesmata (Pd) play an important role in both bacterial effector molecule spreading and host defense responses [77].
To evaluate cell-to-cell communication in leaves treated with methanol, a reporter macromolecule was used to test movement through Pd in different states of dilation. We chose a reporter containing two fused green fluorescent proteins (2×GFP) to query the non-targeted Pd transport of macromolecules [78]. Mature source leaves have generally been considered closed to 2×GFP (54 kDa) because their Pd size exclusion limit (SEL) does not permit proteins with a size of 47 kDa [79]. To establish a system to monitor cell-to-cell transit, we exploited an “agroinjection strategy” to deliver the 2×GFP plasmid into the cell nucleus [80], [81]. To monitor single infection sites, N. benthamiana plants were agroinjected with a diluted (1∶1000) bacterial suspension. Plants were then exposed to methanol vapors and examined by fluorescent light microscopy 30 h after agroinjection. Counting the number of epidermal cells surrounding the initial Agrobacterium-transformed cell that display fluorescence provides a quantitative measure of 2×GFP movement. When the Pd were closed, 2×GFP was detected mainly in single cells (Figure 9A, upper). However, fluorescent signals were distributed in 2- or 3-cell clusters (Figure 9A, bottom) when Pd were dilated. In the control plant, approximately 6% of the signal was distributed in 2- to 3-cell clusters (Figure 9B). These observations were consistent with the known rate of 2×GFP movement through plant tissues [78]. When methanol-treated plants were examined, more than 20% of the signal was distributed in 2 - to 3-cell clusters, indicating that the ability to support cell-to-cell movement of 2×GFP was enhanced. Specifically, whereas only 1% of the signal was found in 3 cell-clusters in the control leaves, with methanol-treatment, this value was increased up to 7%. The unpaired two-tailed Student's t-test confirmed the statistical significance of the differences in the cell-to-cell movement of 2×GFP between the control plant and plants treated with methanol (Figure 9B).
Collectively, these data indicate that methanol acts as a signal that facilitates the movement of 2×GFP between cells.
To examine the role of MIGs in Pd dilation (gating), we monitored the relative cell-to-cell spreading of 2×GFP within the epidermis of N. benthamiana leaves co-agroinjected with binary plasmids encoding BG, NCAPP or MIG-21 directing the synthesis of the respective mRNAs, as tested by qRT-PCR (data not shown). Figure 9C shows that in the control leaves, which were co-agroinjected with an empty Bin19 vector, approximately 7% of the signal was distributed in 2- to 3-cell clusters. When leaves were co-agroinjected with NCAPP, MIG-21 or BG, the movement of 2×GFP was enhanced: more than 23, 26 or 22% of the 2×GFP signal was detected in cell clusters. The unpaired two-tailed Student's t-test confirmed the statistical significance of the differences in the cell-to-cell movement of 2×GFP between the vector-only control and leaves co-agroinjected with NCAPP, MIG-21 or BG (Figure 9C).
Collectively, these data imply that gaseous methanol may trigger leaf Pd dilation (gating) by inducing the mRNA accumulation of MIGs such as NCAPP, MIG-21 and BG.
Our model suggests that methanol-triggered Pd dilation should enhance viral spread within the plant. To examine this possibility, we inoculated plants with a crTMV binary vector that carries an autofluorescent tag GFP (crTMV:GFP) in the place of its coat protein gene [82] and treated the transfected plants with methanol, as shown in Figure 10A. Figure 10B shows the quantification of GFP foci in leaves at 3 dpi. Methanol treatment reduced the number of GFP foci per cm2 presumably due to the induction of antibacterial resistance, which was consistent with our data showing that methanol exposure inhibited R. solanacearum growth (see Figure 2B). Importantly, the stimulation of local viral movement by methanol was indicated by the appearance and spread of the GFP signal. Figure 10C shows that while viral foci became visible in all plants approximately at the same time (3 dpi) after inoculation, viral reproduction representing viral RNA replication and RNA cell-to-cell movement occurred more rapidly in the methanol-treated plants than in the control plants. Figure 10D summarizes the results of the statistical analysis of the data, with the horizontal red lines across the boxes representing the median size of the GFP expression foci (µm2×104). ANOVA confirmed the statistical significance of the differences in focus size between the control and methanol-treated leaves (P = 0.005).
Because BG, NCAPP and MIG-21 can enhance cell-to-cell movement, they may also increase viral RNA movement and/or replication. Therefore, BG, NCAPP and MIG-21 may increase TMV-directed GFP accumulation due to viral reproduction. We tested this hypothesis using crTMV:GFP and binary vectors encoding BG, NCAPP and MIG-21 through co-agroinjection of N. benthamiana leaves. At five days after co-agroinjection with vectors encoding BG, NCAPP and MIG-21, the GFP accumulation in whole leaves increased by 13–23 fold (Figure 10E). These results suggest that BG, NCAPP and MIG-21 enhance viral reproduction. A change in the accumulation of GFP expressed from the viral vector can be caused by a change in viral RNA movement and/or a change in viral replication.
Under natural conditions, viral RNA directly enters the cytoplasm of a negligible number of cells following leaf wounding. Agrobacterium-delivered plant viral vectors exploit the host RNA polymerase II–mediated nuclear export system, which includes 5′-end capping, splicing and 3′-end formation [83]. To test whether methanol or vapors from wounded plants can enhance viral reproduction in TMV-inoculated leaves, we used an experimental design that mimicked the natural condition of viral entry, excluding Agrobacterial participation.
In contrast to controls, plants incubated with wounded N. benthamiana in a hermetically sealed desiccator exhibited increased sensitivity to TMV, as reflected by TMV RNA accumulation (Figure 11A). The same effect occurred when methanol was evaporated in the desiccator. The unpaired two-tailed Student's t-test confirmed the statistical significance of the differences in TMV RNA accumulation between the “receivers” of intact plants, wounded plants or methanol.
The use of the flow-through system to provide continuous airflow from wounded N. benthamiana plants to intact target N. benthamiana plants (Figure 11B, upper panel) confirmed the results of experiments with the hermetically sealed desiccator. “Receiver” plants exposed to air from the desiccator containing wounded plants acquired increased sensitivity to TMV in comparison to control plants (Figure 11B, bottom).
The statistical significance of the differences in TMV RNA accumulation in the inoculated (48 and 72 h after TMV inoculation) or systemically infected leaves (120 h after TMV inoculation) between the “receivers” of intact or wounded plants were confirmed by the unpaired Student's t-test.
These data indicate a role for methanol in triggering MIG expression, which leads to enhanced viral spread and/or reproduction.
The amazing capacity of plants to recognize pathogens through strategies that involve both conserved and variable pathogen elicitors has been previously reported [5], [84], [85]. However, the molecular mechanism by which plants protect themselves against bacterial pathogens remains obscure. This is mainly due to a lack of knowledge about the long-distance signals that trigger systemic reactions in plants. One recent study suggested that a long-range factor, GLV, may increase resistance to the bacterial pathogen Pseudomonas syringae [68]. Here, we characterized another VOC, methanol, which induces a protective reaction against R. solanacearum.
Methanol is a natural plant product that accumulates in the leaf tissue and is emitted when the stomata open in the morning [57], [58]. Our data reveal that leaf wounding stimulates additional methanol emission. Five aspects of wound-stimulated methanol production are especially interesting. First, there is a direct correlation between de novo PME synthesis and methanol emission (Figure 1A,D). We observed a 20-fold increase in the emission of gaseous methanol at 3 h after leaf damage in comparison to the methanol emission by intact control leaves (Figure 1D). Second, methanol generated by de novo synthesized PME is released into the air but does not accumulate in leaf tissue or sap (Figure 1E). Third, gaseous methanol upregulates methanol-inducible genes (MIGs) in the leaves of neighboring plants (Figures 6,7). Fourth, methanol induces antibacterial resistance (Figures 2,5). Fifth, although virus entry per se induces PME mRNA accumulation (Figure 1A), gaseous methanol drastically increases the TMV sensitivity of non-wounded leaves (Figure 11).
We suggest the following model to explain the mechanism of the observed phenomenon (Figure 12). Microdamage (Figure 12, step 1) to the leaf caused by wind-induced leaf rubbing, human handling or insect attack, results in the upregulation of the PME gene (Figure 12, step 2). Upregulation of the PME gene leads to at least three events. First, PME triggers defense reactions that provide resistance against bacteria and viruses; i.e., wound mediated PME mRNA accumulation may promote the defense reactions described earlier [52]. It is worth to emphasize that a model for a mechanical damage – transgenic tobacco overexpressing PME – is resistant to R. solanacearum (see Figure S6 and Table S2). Second, PME enzymatic activity increased 2.5-fold in N. benthamiana leaves at 3 h after wounding (139±9.2 vs. 360±0.068 nkat/mg). Third, PME catalyzes the production of gaseous methanol (Figure 12, step 3), which induces the MIG mRNA accumulation (Figure 12, step 4). Gaseous methanol may provide a feedback loop and suppress PME transcription (Figure 12, step 5) such that the leaf returns to its pre-wounding methanol production state. PMEi is likely to take part in this process by suppressing PME enzymatic activity [75]. MIGs are responsible for TMV spreading/reproduction and resistance to R. solanacearum (Figure 12, step 6).
It was previously shown that transgenic tobacco with elevated PME synthesis is resistant to TMV [52]. This strain exhibits increased methanol emission levels and MIG expression but is not susceptible to TMV. We can consider the effects in PME-transgenic plants to be a consequence of long-term (even “lifelong”) MIG induction, which clearly differs from the effects of short-term methanol treatment. These cases are thus examples of ‘chronic’ and ‘acute’ situations, respectively. The patterns of MIG expression in these two cases are similar (i.e., activated) but still very different (compare Figures 6 and 8). Methanol treatment elicits a MIG “wave” that eventually fades, while MIG expression in PME-transgenic plants is always slightly elevated, which might lead to some secondary effects. Moreover, the expression of PME is much higher in PME-transgenic plants than in methanol-treated plants. We believe that this increased PME expression, which is absent in methanol-treated plants, makes the PME-transgenic plant resistant to TMV.
Methanol is not a plant poison. Treatment of plants with high-concentration methanol solutions (5–50%) revealed that foliar sprays of aqueous methanol, even at a concentration of 50%, led to increased growth and development in C3 crop plants in arid environments [86]. This is likely to be the result of more effective utilization of light energy during photosynthesis [87].
Previously, foliar sprays of a 10% methanol solution were used to identify methanol-sensitive genes in Arabidopsis thaliana [62]. Methanol affected the expression of hundreds of genes, and multiple detoxification and signaling pathways were activated. We used gaseous methanol at physiological concentrations, which were likely 10,000 times lower than those used by Downie et al. [62]. This difference in methanol concentration may explain why we observed the upregulation of only a few previously identified genes (see Table S1). Most of the MIGs identified here (167 ESTs) were classified as stress response genes. The majority of these (117 clones) represented 6 of the most up-regulated SSH-identified genes: BG, PI-II, MIG-21, PMEi, elicitor inducible protein and 1-aminocyclopropane-1-carboxylic acid oxidase, the latter of which is involved in ethylene biosynthesis [88]. The NCAPP transcript was represented by only three clones (Table S1); however, qRT-PCR verification (Table 2) showed that the NCAPP gene was highly inducible, the second most inducible after BG. The SSH approach did not identify the LOX, PR-3, PR-4, FPS or PAL genes, which are induced by other VOCs (Figure S5).
Pathogen attack and plant damage accompanied by the emission of VOCs, including ethylene [17], methyl salicylate [18], methyl jasmonate [19], [20], nitric oxide [21], [22] and cis-3-hexen-1-ol [23], leads to the upregulation of different PR genes [14], [23], [24]. In addition to methanol, we detected the emission of ethylene and GLV. Ethylene is a simple gaseous hormone that integrates external signals with internal processes. Wound-induced ethylene production has been studied thoroughly [89]. The two-step ethylene biosynthesis, i.e., the conversion of S-adenosyl-L-methionine to 1-aminocyclopropane-1-carboxylic acid (ACC) and its subsequent oxidation to ethylene, is regulated by ACC synthase (ACS) and ACC oxidase (ACO), respectively. ACS and ACO are encoded by members of multi-gene families [90]–[92]. Ethylene production is regulated by different isoforms of ACO and ACS in response to different stresses [93]. For example, the accumulation of the transcripts of 3 out of 4 members of the ACO gene family has been examined in tomato, and only ACO1 was wound-responsive [90]. Our SSH approach revealed 6 ACO clones in leaves treated with methanol. We also showed that leaf wounding or PME overexpression (Figure 3) did not increase ethylene emission as a secondary response to methanol. This contradiction might be explained by previous data indicating that ACC synthase, but not ACO, is rate-limiting in ethylene biosynthesis [94]. We have not detected ACC synthase gene upregulation (Table S1). Moreover, it has been shown that increased ACO activity does not always immediately lead to parallel changes in ethylene production [95], e.g., in stress response (methylviologen, oxidative stress inductor, or methyl jasmonate). We hypothesize that methanol might be a similar stimulus, affecting ACO but not ethylene synthesis when applied at physiological concentrations. On the other hand, ethylene biosynthesis is regulated by different isoforms of ACO in response to particular stress cues [93]. Finally, the antibacterial effects of methanol were demonstrated not only in sealed desiccators but also in a flow-through system (Figure 5) in which methanol was blown out. Therefore, the effects of virtual ethylene were excluded or at least significantly diminished.
We also detected cis-3-hexen-1-ol as a representative of GLV emission after plant wounding (Figure 3). The suppression of R. solanacearum growth observed in the “receiver” plants could be caused by gaseous methanol and GLV. This was confirmed in experiments in which cis-3-hexen-1-ol evaporated in the desiccator resulted in decreased bacterial growth in target plants (Figure 2B, diagram bar #5). In an attempt to elucidate the mechanism underlying this phenomenon, we discovered that GLVs rapidly released from wounded leaves stimulate PME mRNA accumulation and therefore PME-generated methanol emission. In our experiments with detached N. benthamiana leaves incubated for 3 h in a 300-ml sealed container with cis-3-hexen-1-ol (0.36 µg), the level of PME mRNA increased by more than two times (2.41±0.37) in comparison to water control (1.00±0.25). Taking into account the connection between cis-3-hexen-1-ol, PME and methanol emission, we believe that the effect of cis-3-hexen-1-ol on bacterial growth is indirect.
Antibacterial resistance accompanied by MIG upregulation is likely to be related to PI-II gene transcription induction. Type I proteinase inhibitors are powerful inhibitors of serine endopeptidases in animals and microorganisms [96]. The PI-II gene is not expressed in the leaves of healthy plants, but it is induced in leaves that have been subjected to different types of stress, including wounding and bacterial infection [76]. R. solanacearum encodes several secreted proteases [97], [98], including a type III effector, PopP2, which mimics a plant transcriptional activator and manipulates the plant transcriptome [99], [100]. PME-transgenic tobacco with high levels of PI-II expression (Figure 8) demonstrated high resistance to R. solanacearum (Figure S6 and Table S2). This finding supports the role of PI-II in the suppression of bacterial proteases.
To determine whether BG, MIG-21 and NCAPP could enhance cell-to-cell communication, we used Agrobacterium to mediate the delivery of GFP- and MIG-expressing vectors. Although methanol treatment induced resistance against bacteria (Figure 10B) and therefore decreased the number of detected crTMV:GFP foci, we found that these foci increased in size (Figure 10 C,D). Methanol changes Pd SEL and upregulates MIGs (BG and NCAPP); therefore, it is likely to promote cell-to-cell trafficking and TMV reproduction. The participation of MIG-21 in cell-to-cell trafficking is unconfirmed, but the role of BG and NCAPP in Pd dilation has been described previously [74], [101], [102]. However, there is no data explaining the correlation between methanol-mediated BG, NCAPP, MIG-21 upregulation and antibacterial resistance. A recently revealed link between nuclear transport and cell-to-cell movement [103] suggests that there may be competition between methanol-mediated cell-to-cell transport and R. solanacearum type III effector nuclear traffic. We cannot exclude the possibility that airborne signals from wounded leaves may also facilitate TMV spreading/reproduction in neighbors as an unintended consequence of the acquired antibacterial resistance of the plants. Interestingly, it has been suggested that the conditions generated by agriculture during the Holocene period may have promoted viral spreading in plants [104].
Further research is required to elucidate the mechanisms of the reactions triggered by methanol in plants. How methanol is regulated during wound stress conditions remains unclear, as do the identities of possible factors involved in this process. The involvement of MIGs in viral spreading has been clearly demonstrated. However, the underlying cellular mechanisms controlling the targeting of BG, NCAPP and MIG-21 to the Pd is still unknown. Finally, the factors that coordinate the spatiotemporal correlation of MIGs with bacterial resistance and viral cell-to-cell spreading and reproduction have yet to be determined.
N. benthamiana and N. tabacum plants were grown in soil in a controlled environment chamber in a 16 h/8 h day/night cycle.
Full-length BG (β-1,3-glucanase), MIG-21 and NCAPP cDNAs were obtained by PCR using the primer pairs 5′GAGCTCATGTCTACCTCACATAAACATAATAC3′/5′AAGCAGTGGTAACAACGCAGAGTACtttttttttttttttttttttttttttttt3′, 5′GAGCTCATGGCATCACTTCAGTGCC3′/5′CTGCAGTCAGCAGCTCCCTCTATTC3′ and 5′GAGCTCATGTCTTCAAAGATTGGTCTG3′/5′CTGCAGCTATTTCTTGATAGAAAACGTG3′, respectively, with total N. benthamiana cDNA as the template.
The viral vector crTMV:GFP (pICH4351) has been described previously [105]. To synthesize the 35S-based binary vectors pBG, pMIG-21 and pNCAPP, PCR-amplified cDNA was inserted into the XbaI, EcoRI (pBG) or SacI, PstI (pMIG-21 and pNCAPP) sites of pBin19.
The methanol, cis-3-hexen-1-ol and ethylene contents were determined by GC on a capillary FFAP column (50 m×0.32 mm; Varian Inc., Lake Forest, CA, USA) in a Kristall 2000 gas chromatograph (Eridan, Russia). The methanol and cis-3-hexen-1-ol in water/decane samples were measured under the following operating conditions: carrier gas – nitrogen, nitrogen flow – 30 ml/min; air flow – 400 ml/min; hydrogen flow – 40 ml/min; injected volume – 1 µl; injector temperature – 160°C; column temperature – 75°C, increased to 150°C at a rate of 15°C/min; retention time – 6.5 min (methanol) or 17 min (cis-3-hexen-1-ol); and flame ionization detector temperature – 240°C. Ethylene content in air samples was analyzed under the following operating conditions: carrier gas – nitrogen; nitrogen flow – 30 ml/min; air flow - 400 ml/min; hydrogen flow – 40 ml/min; injected volume – 1 ml of vapor phase; injector temperature – 130°C; column temperature – 45°C; retention time – 4.5 min; and flame ionization detector temperature – 240°C.
Methanol treatment was executed by exposing plants to methanol vapors on filter paper in a sealed desiccator. The effects of plant VOCs were measured in either a single hermetically sealed 20-l desiccator or a flow-through set-up involving two attached 20-l desiccators (the first for the “emitter” plants and the second for the “receiver” plants) supplied with filtered air at a rate of 0.15 l/min. Intact and wounded N. tabacum or PME-transgenic tobacco plants were used as “emitter” plants, whereas N. benthamiana plants were used as “receivers”. Pots (width, 9.5 cm; depth, 9.5 cm) containing plants (10.0±1.0 g) and soil (198.0±20.0 g) were placed into desiccators and maintained for 3 h or 18 h at a constant temperature of 24°C with a 16 h/8 h light/dark photoperiod. Then, “receiver” plants were withdrawn from the desiccator and tested for MIG RNA accumulation and bacterial and TMV resistance. In experiments assessing the decay of MIG mRNA accumulation, the plants withdrawn from the desiccator after methanol treatment were kept at 24°C with a 16 h/8 h light/dark photoperiod for leaf RNA isolation.
The tobacco strain R. solanacearum was grown under routine conditions on yeast–peptone–glucose (YPG) agar containing the following (per liter): 5 g yeast extract, 10 g peptone, 5 g glucose and 15 g agar. The incubation temperature was 28°C. Overnight cultures of R. solanacearum at the indicated concentrations in 10 mM MES (pH 5.5) buffer supplemented with 10 mM MgCl2 were injected into fully developed leaves by syringe. At four days post inoculation (dpi), bacterial growth was measured by macerating five leaf discs of 1 cm2 from the inoculated tissue of each sample in 10 mM MgCl2, plating the serial dilutions on nutrient agar plates, and counting the colony-forming units (cfu).
Agrobacterium tumefaciens strain GV3101 was transformed with individual binary constructs and grown at 28°C in LB medium supplemented with 50 mg/l rifampicin, 25 mg/l gentamycin and 50 mg/l carbenicillin/kanamycin. Agrobacterium cells from an overnight culture (5 ml) were collected by centrifugation (10 min, 4,500× g), resuspended in 10 mM MES (pH 5.5) buffer supplemented with 10 mM MgSO4 and adjusted to a final OD600 of 0.2 for TMV-directed GFP accumulation or 0.001 for cell-to-cell movement assays. Agroinjection was performed on almost fully expanded N. benthamiana leaves that were still attached to the intact plant. A bacterial suspension was infiltrated into the leaf tissue using a 2 ml syringe, after which the plants were grown under greenhouse conditions at 24°C with a 16 h/8 h light/dark photoperiod. In the cell-to-cell-movement assay, N. benthamiana plant leaves were agroinjected with 2×GFP and were stored for 6 h in a plant growth chamber at 24°C with light; these plants were then loaded into the desiccator. Then, methanol (160 mg) was added, and the desiccator was sealed. After a 3-h exposure to methanol vapors, the plants were withdrawn, and fluorescent cells were counted after 21 h of storage in a growth chamber. In the viral focal growth experiments assessing TMV cell-to-cell spreading, N. benthamiana plant leaves were agroinjected with crTMV:GFP, stored for 6 h in a plant growth chamber, and then loaded in the desiccator. Subsequently, methanol (160 mg) was added, and the desiccator was sealed. After a 3-h exposure to the methanol vapors, the plants were withdrawn. Fluorescent cells were counted after 4 days of storage in a growth chamber at 24°C with a 16 h/8 h light/dark photoperiod.
GFP fluorescence in the inoculated leaves was monitored by illumination with a handheld UV source (DESAGA). At higher magnifications, GFP fluorescence was detected using a dissecting microscope (Opton IIIRS) equipped with an epifluorescence module. Unless otherwise indicated, the lower epidermal cells of injected leaves were observed at 24 or 72 h after agroinfiltration.
50 mg of leaf tissue from infiltrated areas were ground in the 1.5 ml tubes in 200 µl of GFP-extraction buffer (150 mM NaCl, 10 mM Tris-HCl, pH 8.0). Then the samples were centrifuged 16 000× g 10 min and 1 ml of GFP-extraction buffer was added to the supernatant. The fluorescence was measured using Quantech fluorometer (ThermoScientific, USA).
Plant material was ground to a fine powder in liquid nitrogen using a mortar and pestle. Total RNA was extracted from leaves using TRIzol reagent (Invitrogen). Approximately 5 µg of total nucleic acid isolated from mock-treated or virus infected leaves was denatured, separated in 1.5% agarose gels containing 10% formaldehyde in MOPS buffer, pH 7.0, and transferred to a nylon membrane (Hybond-N+, Amersham). Membranes were incubated in a pre-hybridization solution containing 6× SSC, 0.5% SDS, 5× Denhardt's reagent and 200 µg/ml tRNA for 4 h at 68°C and probed with a denatured DNA fragment containing the PME coding sequence. Probes were labeled with [α32P]-dATP (3000 Ci/mmole) in a PCR reaction.
N. benthamiana plants withdrawn from the desiccator after exposure to methanol were mechanically inoculated with TMV virions (100 µg/ml) in 50 mM sodium phosphate buffer, pH 7.0, in the presence of Celite, as described previously [106].
Concentrations were determined using a Nanodrop ND-1000 spectrophotometer (Isogen Life Sciences). All RNA samples had a 260∶280 absorbance ratio between 1.9 and 2.1.
After DNAse treatment (Fermentas), 2 µg of denatured total RNA was annealed with 0.1 µg of random hexamers and 0.1 µg of Oligo-dT and incubated with 200 units of Superscript II reverse-transcriptase (Invitrogen, USA) for 50 min at 43°C to generate cDNA. Real-time qPCR was carried out using the iCycler iQ real-time PCR detection system (Bio-Rad, Hercules, CA, USA). Target genes were detected using Eva Green master mix (Syntol, Russia) according to the manufacturer's instructions. The thermal profile for EVA Green real-time qPCR included an initial heat-denaturing step at 95°C for 3 min and 45 cycles with a denaturation step at 95°C for 15 s, an annealing step (amplicon-specific temperatures provided in Table S4) for 30 s and an elongation step at 72°C for 30 s coupled with fluorescence measurement. Following amplification, the melting curves of the PCR products were monitored from 55 to 95°C to determine the specificity of amplification. Each sample was run in triplicate, and a non-template control was added to each run. Target gene mRNA levels were calculated according to the equation proposed by Pfaffl [111]: EtargetΔCt target (sample-reference). PCR efficiency (E) was calculated according to the equation E = 10(−1/slope) by performing the standard curves. Target gene mRNA levels were normalized to the corresponding reference genes (18S and ef-2ά for N. tabacum).
Student's t-tests were performed using Excel (Microsoft, Redmond, WA). ANOVA tests were performed using SPSS v.18 (IBM Corporation, Somers, NY). P-values<0.05 were considered significant.
|
10.1371/journal.ppat.1004033 | An Immunomics Approach to Schistosome Antigen Discovery: Antibody Signatures of Naturally Resistant and Chronically Infected Individuals from Endemic Areas | Schistosomiasis is a neglected tropical disease that is responsible for almost 300,000 deaths annually. Mass drug administration (MDA) is used worldwide for the control of schistosomiasis, but chemotherapy fails to prevent reinfection with schistosomes, so MDA alone is not sufficient to eliminate the disease, and a prophylactic vaccine is required. Herein, we take advantage of recent advances in systems biology and longitudinal studies in schistosomiasis endemic areas in Brazil to pilot an immunomics approach to the discovery of schistosomiasis vaccine antigens. We selected mostly surface-derived proteins, produced them using an in vitro rapid translation system and then printed them to generate the first protein microarray for a multi-cellular pathogen. Using well-established Brazilian cohorts of putatively resistant (PR) and chronically infected (CI) individuals stratified by the intensity of their S. mansoni infection, we probed arrays for IgG subclass and IgE responses to these antigens to detect antibody signatures that were reflective of protective vs. non-protective immune responses. Moreover, probing for IgE responses allowed us to identify antigens that might induce potentially deleterious hypersensitivity responses if used as subunit vaccines in endemic populations. Using multi-dimensional cluster analysis we showed that PR individuals mounted a distinct and robust IgG1 response to a small set of newly discovered and well-characterized surface (tegument) antigens in contrast to CI individuals who mounted strong IgE and IgG4 responses to many antigens. Herein, we show the utility of a vaccinomics approach that profiles antibody responses of resistant individuals in a high-throughput multiplex approach for the identification of several potentially protective and safe schistosomiasis vaccine antigens.
| Schistosomiasis is a neglected tropical disease that kills as many as 300,000 people each year. Mass drug administration is widely used to control schistosomiasis, but fails to prevent rapid reinfection in endemic areas. There is a desperate need for a prophylactic vaccine; however, very few candidates have been developed. Herein, we take advantage of recent advances in systems biology and longitudinal studies in schistosomiasis endemic areas to pilot an immunomics approach to the discovery of vaccine antigens. The emerging field of immunomics enables the determination of an “antibody signature” to a pathogen proteome for both resistant and susceptible individuals. We constructed the first protein microarray for a multi-cellular pathogen and probed it with sera from naturally resistant vs. susceptible individuals from a high transmission area in Northeastern Brazil. Using multi-dimensional cluster analysis, we showed that resistant individuals mounted a distinct and robust IgG1 antibody signature to a small set of newly discovered and well-characterized surface antigens in contrast to infected individuals. This antigen discovery strategy can lead to identification of several potentially protective and safe schistosomiasis vaccine antigens.
| Schistosomiasis is a chronic, often debilitating, parasitic disease affecting over 200 million people worldwide and killing at least 300,000 people annually [1]. The disability adjusted life years (DALYs) lost to schistosomiasis are potentially as high as 70 million [2], [3]. Adult flukes live in the portal and mesenteric veins (Schistosoma mansoni and S. japonicum) or in the veins of the bladder (S. haematobium), as male/female pairs, and survive for many years producing hundreds of fertilized eggs per day. Severe morbidity results from the host immune responses to eggs that become trapped in the tissues, including periportal fibrosis, portal hypertension, urinary obstruction and bladder carcinoma [4].
Currently, chemotherapy with praziquantel (PZQ) is the standard treatment for schistosomiasis. Control programs based on mass drug administration (MDA) with PZQ have been complicated by rapid and frequent re-infection of treated individuals, and the difficulties and expense of maintaining continuous MDA over the long term [5]. Additionally, resistance to PZQ can be induced in the laboratory [6], and field isolates displaying reduced susceptibility to the drug have been reported (reviewed in [7]). Despite recent large-scale MDA efforts [8], integrated control programs aimed at limiting schistosomiasis by improving education and sanitation, molluscicide treatment programs to reduce the population of the intermediate snail host, and chemotherapy have had limited success [5], [9]. A vaccine that induces long-term immunity to schistosomiasis is therefore necessary to reach our goals of elimination.
The high prevalence of chronic schistosomiasis in endemic populations suggests that sterile immunity is rarely generated. However, the decline in infection intensity at an earlier age in populations with high infection intensity [10], and more rapid development of resistance to re-infection after several rounds of PZQ treatment (drug-induced resistance) [11], indicates that non-sterilizing immunity, though slow to develop, can occur. Despite the slow acquisition of non-sterile immunity over time, there is still an urgent need for a prophylactic vaccine, particularly one that targets children, who represent the most at-risk population.
There are two major obstacles to the development of an efficacious schistosomiasis vaccine. The first is the ability of schistosomes to employ a range of strategies for evasion of the host immune response. Central to the parasite's ability to evade immune clearance is its unique host-interactive outer surface, or tegument, consisting of a single, contiguous, double-bilayered membrane that covers the entire worm [12]. At this interface essential functional interactions with the human host occur, such as nutrient uptake and environmental sensing. The tegument is also the primary site where the parasite defends itself against immune recognition. The host-interactive surface is indeed the target of the few successful examples of metazoan parasite vaccines, such as those targeting the cattle tick Boophilus microplus [13], the gastrointestinal nematode Haemonchus contortus [14] and the cestodes, Taenia ovis [15] and Echinococcus granulosus [16]. The second major obstacle to the development of a schistosomiasis vaccine resides in the historic approach to antigen discovery for this pathogen. To date, only one schistosomiasis vaccine, rSh28GST from S. haematobium, is currently in phase I clinical trials, where it was shown to be safe and immunogenic [17]. Other vaccine antigens for S. mansoni are in pre-clinical and clinical development [18], [19], with safety and immunogenicity results yet to be reported. We [19]–[21] and others [22], [23] have advocated for the utility of tegument proteins as a basis for subunit vaccines against schistosomiasis. Three of the current lead candidate antigens are located in the tegument and are exposed on the surface of the parasite [24]–[26]. The genomes for the three major human schistosomes have been sequenced [27]–[29], and coupled with proteomic studies that characterised the surface proteomes of S. mansoni [30] and S. japonicum [31], have provided researchers with a catalogue of proteins for discovery and development of a new panel of vaccine antigens.
To best mine this extensive proteomic data and identify antigens that are preferentially recognised by antibodies from naturally resistant individuals resident in areas of high transmission for schistosomiasis, we have utilized a clinical cohort of individuals referred to as Putative Resistants (PRs). As part of a ten year longitudinal study of individuals from high S. mansoni transmission areas of Brazil, we identified a cohort of individuals who were constantly exposed to S. mansoni infection as determined by extensive water contact and epidemiological studies, but remained egg-negative over the course of the study [32]–[34]. In addition to this unique epidemiological profile, these individuals mounted an immune response that displayed a markedly different phenotype from that of chronically infected (CI) individuals [35]–[37]. Indeed, two of the current antigens in pre-clinical development - Sm-TSP-2 [26] and Sm29 [24] – were discovered as a result of their selective recognition by PR subjects, highlighting their utility as a tool for discovery of protective vaccine antigens.
Herein, we describe the screening of the first protein microarray for a human helminth parasite, and only the second such array for a eukaryotic parasite other than Plasmodium sp. We developed a targeted array consisting primarily of tegument derived proteins from both S. mansoni and S. japonicum [38]and screened the array with sera from PR and CI individuals with low, medium and high intensity infections, and then compared and contrasted antibody/antigen recognition profiles to determine antibody signatures that characterised natural resistance or susceptibility to infection. We assessed IgG subclass and IgE responses such that potential vaccine antigens could be assessed for their protective properties as well as their safety profiles in terms of exacerbating allergic IgE responses [39]. We showed that individuals with medium and heavy intensity infections generally recognized more antigens and with higher magnitude than did PR individuals and those with low infection intensities. Moreover, we found that PR individuals did not mount an intense IgE response to these antigens compared to CI individuals, but instead produced IgG1/3 (cytophilic) antibody responses to only a few membrane bound antigens. We successfully utilized this approach to identify new, and confirm existing, vaccine antigens via their selective IgG1/IgG3 recognition profiles by PR individuals in the absence of a potentially deleterious IgE response.
Details of the microarray production and associated QC have been described elsewhere [38] and are shown in Figure S1 and Table S1, however this is the first report that describes probing of the microarray with human sera. We included on the array two dilutions of the ubiquitously immunoreactive Epstein Barr virus protein EBNA-1 at two concentrations, 0.1 and 0.3 mg/ml, as a non-schistosome control for the rapid translation system (RTS) used to express schistosome recombinant proteins. Both antigens were consistently recognized by IgG1 antibodies from all individuals tested (Figure S2), indicating that sera from all cohorts were of sufficient integrity for further analyses.
Given the distinct roles of different immunoglobulin isotypes and IgG subclasses in chronic helminth infections, and to gain a comprehensive picture of antibody reactivity from PR versus CI individuals, we analyzed IgG1, IgG3, IgG4 and IgE responses to soluble worm antigen preparation (SWAP) and a panel of schistosome antigens. PR subjects mounted the strongest anti-SWAP IgG1 and IgG3 responses whereas the moderate and heavily infected groups mounted the strongest IgG3 responses to SWAP (Figure S3). One hundred and sixteen (116) from a total of 215 (54%) RTS proteins spotted were recognized by at least one antibody isotype/subclass from at least one cohort of exposed individuals (reactive proteins) (Table S2 and Table S3). Individuals with medium and heavy intensity infections generally recognized more antigens and with stronger SI than did PR individuals and those with low infection intensities (Figures 1A-D). CI-Mod and CI-Heavy cohorts had significantly higher IgG4, IgG3 and IgE responses than PR and CI-Light cohorts (Figures 1A-C) (*p<0.05; **p<0.01, ***p<0.001, ****p<0.0001). In contrast, the PR cohort had significantly higher IgG1 (Figure 1D) responses than CI-Mod and CI-Heavy cohorts, although the total number of antigens above the cut-off was lower for this antibody subclass. In general, there was a strong correlation between infection intensity and the number of antigens recognized by combinations of IgG3, IgG4 and IgE from infected individuals (Table S4 and Table S5). Of the 116 reactive proteins (RTS and recombinants) 41 were recognized by just a single antibody isotype/subclass: 8 proteins were recognized by only IgG1, 24 proteins were recognized by only IgG3, one protein was recognized by only IgG4 and 10 proteins were recognized by only IgE. Eleven proteins were recognized by all antibody isotypes/subclasses (Figure 1E) (Table S2).
IgE responses were detected to 79 different antigens (Figure 2, Table S2) and most of these were restricted to the CI-Mod and CI-Heavy groups. Significant differences (P≤0.05) between mean antibody responses from 2 or more of the endemic groups were detected to all 79 proteins (Tables S2–4). The only purified recombinant protein (non-RTS) that was the target of an IgE response was Sm29. Antigens for which the strongest IgE responses were detected included proteins that were predicted and/or proven to be located on the tegument membrane (including tetraspanins, Ly6/CD59-like proteins such as Sm29, and glucose transporters) and predicted intracellular proteins including mitochondrial enzymes, chaperones and glycolytic enzymes such as triose phosphate isomerase (Table S1 and Table S2).
IgG4 responses were detected to 21 proteins (Figure 3) – 20 RTS proteins and purified recombinant Sm29 expressed in E. coli. Significantly different IgG4 responses were detected for all reactive antigens between at least two of the endemic cohorts (Table S3, Table S4). Sm29 was recognized weakly by IgG4 but was considered a cross-reactive protein because the US non-endemic control group had a low level IgG4 response against this protein (Figure 3, Table S2).
IgG3 responses were detected to 96 proteins, 95 of which were RTS and 1 E. coli-derived purified recombinant proteins (Figure 4, Table S2). Of the 96 reactive proteins, only 3 displayed no significant differences between the cohorts (Figure 4, Table S4).
IgG1 responses were detected to 43 proteins (Figure 5, Table S2), including purified recombinant Sm29 and Sm-TSP-2. Significantly different IgG1 responses between endemic cohorts were detected for 31 of these reactive antigens (Table S4). Twenty-two proteins were the targets of an IgG1 response in the PR cohort that was significantly different to at least one of the CI groups (Table S4). The most robust of these PR-specific IgG1 responses were aimed at the two positive control recombinant proteins, Sm-TSP-2 and Sm29, and the RTS protein Smp_139970, a calmodulin-3 like protein that we have termed Sm-CAM-3. Sm-CAM-3 shared 46% and 23% amino acid identities with its closest S. mansoni and primate (macaque) homologues respectively (Figure S4).
Correlations between different isotype responses to the same proteins were calculated (Table S5). The strongest correlations detected (r2>0.9) were between IgG4/IgE responses in all the schistosome-exposed cohorts (P<0.0001, Figure S5) and IgG3/IgG4 and IgG3/IgE responses in the CI-Mod and CI-Heavy cohorts.
All the 215 proteins printed on the array were subjected to cluster analysis to identify proteins with similar reactivity profiles. Two different methods of unsupervised clustering were applied: partitional and hierarchical clustering. Considering all of the possible combinations of antibody reactivity patterns, we used classical multidimensional scaling (MDS) cluster analysis to generate clusters of proteins. In this analysis, the reactivity of each protein was described with the average SI for each cohort. Proteins with an average SI below the cut-off in the evaluated group were considered to be zero and only proteins with an average signal intensity above the cut-off for at least one isotype/subclass were considered for clustering. For partitional clustering, working with 4 antibody isotypes/subclasses and 5 cohorts, proteins fell into one of 7 clusters determined by K-means methodology [40]. To facilitate visualization of the process (and avoid superimposing data points in a 2-dimensional format), we compressed the 20 dimensions into just 2 dimensions (Figure 6A). For hierarchical clustering, a dendrogram was designed using complete linkage [41] to combine two proteins and color coded to match the k-means clusters; identities of the proteins within each cluster can be found in Table S2 and Figure S5. There was good correlation between partitional and hierarchical clustering, indicating that the division of proteins in these groups was robust. To further enhance the visualization process, clustered proteins were distributed in 2 dimensions based on isotype/subclass specific responses of each cohort to each individual protein (Figure 6B).
A number of clusters of interest for vaccine development were observed. Clusters 4 and 5 are characterized by proteins that are moderate to strong targets of IgE and IgG3 or IgG1 responses, respectively, particularly in the CI-Mod and CI-Heavy groups. Cluster 7 predominantly consists of non-reactive proteins and a small handful of proteins that were exclusively targeted by IgG1 responses of the individual sera in the PR group but not the CI or non-endemic control groups. Of the strongly reactive PR IgG1 proteins, Sm29 was also recognized by all IgG subclasses as well as IgE and belonged to cluster 2; Sm-TSP-2 and Sm-CAM-3 on the other hand were uniquely targeted by PR IgG1 and not other isotypes or subclasses and belonged to cluster 7 (Figure 6, Figure S6 and Table S2). Other cluster 7 proteins that were uniquely recognized by PR IgG1 responses, albeit relatively weak responses, included Smp_124240 (Na/K transporting ATPase beta subunit) and Sj_AY915291 (fatty acid CoA synthetase). Both of these proteins have multiple predicted membrane spanning domains (not shown).
We examined the antibody recognition profiles of individuals within the cohorts to some of the current antigens that are under various stages of pre-clinical development as human schistosomiasis vaccines, including Sm-TSP-2, Smp80 (calpain) and Sm14, and bovine vaccines to interrupt zoonotic transmission (Sj23). We compared the responses of these known vaccine antigens with selected RTS proteins including the PR IgG1-specific target Smp_139970 (Sm-CAM-3) and 2 proteins that were significant targets of IgE and/or IgG4 in CI-Mod and CI-Heavy cohorts, Smp_050270 and Smp_008310 (Figure 7). Different antigens displayed distinct IgE and IgG subclass profiles. Sm-TSP-2 was the target of a strong IgG1 response that was unique to the PR group, in agreement with the published literature [26]. Smp_139970 (Sm-CAM-3) showed the same recognition profile as that targeting Sm-TSP-2. Mean SI values for IgE were below the cut-offs for all of the vaccine antigens, however varying numbers of individuals in the CI-Moderate and CI-Heavy groups were positive for some of the antigens, although SI values were weak compared with other RTS proteins such as Smp_008310. Similarly, IgG4 responses were mostly below the cut-off for the established vaccine antigens. The mean IgG3 responses were mostly negative but weakly positive for Smp80 and Sm-CAM-3 in the CI-Mod cohort. IgG1 responses to the known (and potentially new) vaccine antigens were the most noteworthy in terms of unique recognition by the PR cohort: strong IgG1 responses to both Sm-TSP-2 and Sm-CAM-3 were detected in just the PR group and none of the CI groups (Figure 7).
Herein we describe the first immunomics-based approach to study the humoral immune response to a multi-cellular pathogen. The “immunome” can be defined as the entire set of antigens or epitopes that interface with the host immune system [42]. Recent advances in high order multiplexing, or megaplexing, such as the protein microarray discussed below, provide a practical, high-throughput and affordable approach to estimating the immunomic profiles of humans or animals to a pathogen [43], [44]. This approach permits investigators to assess the repertoire of antibodies created in response to infections or vaccinations from large collections of individual sera. Further, it can be used to perform large-scale sero-epidemiological, longitudinal and sero-surveillance analyses not possible with other technologies.
Numerous passive transfer studies [45], [46] support the critical role of antibodies in immunity to S. mansoni infection in rodent models. Perhaps the most compelling evidence that the humoral immune response targets the tegument and can kill parasites comes from studies with rats, which are semi-permissive to S. mansoni [47], [48]. Resistance to schistosomiasis can be passively transferred via serum from resistant rats, and protective antibodies can be removed by adsorption on the surface of schistosomes [49]. Indeed, two of the recombinant antigens used on our array - Sm-TSP-2 and Sm29 - have proven efficacious in a mouse challenge model and were the major targets of single chain antibodies from resistant rats adsorbed from the surface of live schistosomes [50].
The role of antibodies in protective immunity against schistosomiasis in humans is, however, somewhat contentious. Unlike experimentally infected rats, protective immunity to schistosomes in humans develops slowly (over many years) and is rarely sterilizing in nature. Distinct molecular mechanisms are thought be critical in the acquisition of immunity in different transmission scenarios. For example, some individuals can successfully mount a protective antibody-mediated response that targets adult S. mansoni antigens after repeated rounds of PZQ therapy [51], [52] – this drug-induced resistance is mediated by IgE and T helper type 2 (Th2) cytokines, and can be accelerated and augmented by repeated drug treatment [11], [53], [54]. We show here that CI individuals make robust IgE responses to many antigens, and the number of antigens recognized increases with increasing intensity of infection as measured by eggs per gram of feces. This would appear to contrast with the protective role that is often associated with IgE in helminth infections, including schistosomiasis [55].
In contrast to drug-induced resistance to schistosomiasis, naturally acquired resistance has been reported in a subset of people who have constant exposure to schistosomes but have never been treated with PZQ (exemplified by the PR cohort in our study) – these individuals generate robust T cell responses against the surface of the larval schistosomulum and are characterized by elevated levels of IFN-γ [35], [56]–[58]. We show here that PR individuals, despite constant exposure to S. mansoni, do not appear to mount a strong IgE response to the proteins on the array. Unlike CI individuals, PR subjects are repeatedly negative for schistosome eggs in the feces, and are therefore unlikely to receive the IgE-inducing stimulus of eggs trapped in the bowel wall and the subsequent hepato-portal inflammation that typifies chronic schistosomiasis. PR individuals are likely to kill juvenile schistosomes before they reach sexual maturity, either in the skin or the lungs. The strong recognition of just a handful of tegument antigens by IgG1 from PR individuals therefore implies a protective role for IgG antibodies (and/or T cells) targeting these proteins.
A major outcome of this study is the development of a tool by which the immunogenicity and probable safety profile (i.e. IgE recognition) of an antigen can be rapidly assessed, and a putative association of that antigen-antibody interaction with resistance or susceptibility to infection inferred. In terms of vaccine antigen discovery, we employed the following principles to a given antigen: (1) up-selection for further evaluation based on preferential recognition by IgG1 and/or IgG3 from the PR but not the CI cohorts; (2) down-selection based on recognition by IgE from either PR or CI individuals. Our rationale for down-selection of IgE-inducing antigens is primarily due to safety concerns [19]. In a recent phase I clinical trial of the of the Na-ASP-2 hookworm vaccine in a Necator americanus-endemic area of Brazil, hookworm exposed individuals were vaccinated with a recombinant protein which, despite proving safe and immunogenic in helminth-naïve individuals in the USA [59], induced an immediate hypersensitivity (urticarial) response in vaccinees, resulting in suspension of the clinical trial and further development of this antigen as a vaccine [39]. The allergenicity of the hookworm vaccine was linked to pre-existing IgE to parasite-derived Na-ASP-2 in the circulation of exposed individuals. Rather than completely excluding antigens that are the targets of pre-existing IgE responses from further development as vaccine antigens, we suggest that IgE antigens that associate with resistance might be carefully progressed towards studies that assess their anaphylactogenic potential, particularly given the large body of data that shows a protective role for IgE in resistance to human helminths. For a vaccine targeting infants prior to natural exposure to schistosome-infected water, a vaccine that incorporates antigens which are the targets of IgE in older individuals is feasible, and might even harness the putative protective capacity of this immunoglobulin isotype as children first become exposed to the parasite and undergo boosting and isotype class switching.
Using these criteria described above to set the parameters for multi-dimensional clustering, a small number of antigens that made up cluster 7 are noteworthy as targets of IgG1 from the PR cohort but not IgE from any cohort. One of these antigens, Sm-TSP-2, was already known to be a selective target of PR IgG responses [26], and its recognition profile on the microarrays served to confirm its potential as a vaccine antigen, as well as the utility of our approach for identifying protective antigens. At least three RTS proteins from cluster 7 were noteworthy as targets of PR IgG1 but not IgE from any exposed cohort. Of these antigens, the strongest IgG1 response was aimed at Smp_139970 (Sm-CAM-3). Sm-cam-3 mRNA (and its S. japonicum ortholog, contig 8758) is upregulated in cercariae [60], [61] (Figure S4A) and encodes a member of the calmodulin family of calcium-sensing proteins. Calmodulins respond to changes in calcium ion concentrations by undergoing a conformational change upon binding, which in turn, facilitates interactions with other signaling proteins. Using gene silencing and quantitative parasite motility assays, a schistosome calmodulin dependant kinase (CamKII) was shown to minimise the impact of PZQ treatment against adult S. japonicum [62]. Sm-CAM-3 is a small (∼8 kDa) protein that contains a single Ca2+ binding EF-hand motif and shares moderate homology with mammalian and other parasite calmodulins, the majority of which are larger than Sm-CAM-3 and contain multiple EF-hands. Of the four calmodulin family members present in S. mansoni, two have been characterized (SmCaM1 and SmCaM2) and were detected in the tegument of adult worms [30], [63] and the epidermal and tegumental layers of larval stages in the snail host [64]. Indeed RNAi-mediated silencing of these genes resulted in stunted larval development [64]. Generic calmodulin antagonists have been shown to inhibit the in vitro growth and egg-hatching ability of schistosomes [64], [65] and the growth of Plasmodium [66] but have limited use as anti-parasitic interventions due to the highly conserved nature of calmodulins across species. It is therefore noteworthy that Sm-CAM-3 shares only ∼20% identity with primate calmodulins (Figure S4B), supporting its development as a safe schistosomiasis vaccine that is unlikely to induce antibodies which cross-react with homologous human proteins. Although Sm-CAM-3 is immunogenic in PR individuals, it is a small protein and might not be overly immunogenic as a subunit vaccine. An ideal schistosomiasis vaccine might therefore incorporate Sm-CAM-3 as part of a larger chimeric construct with other vaccine antigens such as Smp80-calpain or Sm-TSP-2, a strategy that was recently shown to boost efficacy in a mouse challenge model [67]. Two other cluster 7 RTS proteins were unique targets of just IgG1 in the PR cohort, albeit with relatively weak responses – Smp_124240 (Na/K transporting ATPase beta subunit) and Sj_AY915291 (fatty acid CoA synthetase). Both proteins have multiple predicted membrane spanning domains, and warrant expression of defined extracellular domains in a cell-based system for further investigation as vaccine antigens.
In addition to Sm-TSP-2, two other high profile vaccine antigens, Smp80-calpain (Smp_137410) and Sm14 (Smp_095360), were contained within cluster 7. Although some individuals mounted IgG and to a lesser extent IgE responses to these RTS products, the mean SI values for any isotype/subclass to either antigen were below the cut-offs. The apparent absence of a positive mean SI (indicative of a strong antibody response) to an RTS protein on the array should be treated with due caution – the benefit of RTS protein production is its inherent high-throughput nature and its suitability for printing onto arrays in nanoliter quantities. One of the major limitations of RTS protein production however is the absence of complex secretory machinery and the dependence on secretory pathways for correct folding and processing of some proteins. While there are RTS systems that contain eukaryotic microsomal membranes, these systems are not widely used for protein microarray production, and would add substantial cost to the production of large arrays. Sm-TSP-2 was available to us in recombinant, cell-derived form, and was not successfully translated in RTS form. Smp80-calpain and Sm14 were successfully produced in RTS form (Figure S1), but we cannot guarantee faithful replication of all the native epitopes. We therefore urge caution in the interpretation of a “negative” result using this microarray approach, but we have confidence in assigning a “positive” antibody response.
Proteome microarrays have been used to identify candidate vaccine antigens for a number of infectious diseases of viral and bacterial origin [43]. To date, Plasmodium is the only eukaryotic parasite for which proteome microarrays have been described [68]. Screening of P. falciparum proteome arrays with sera from well-defined clinical cohorts resident in malaria-endemic areas [68]–[70] and recipients of radiation attenuated vaccines [71] has resulted in the identification of a suite of new vaccine antigens, some of which have proven efficacious in mouse models of malaria (DLD, unpublished observations).
Our proteome microarray included a small number of carefully selected S. mansoni and S. japonicum proteins, working on the assumption that tegument surface proteins are most likely to be the targets of PR protective immune responses. Based on our findings here, notably the large number of immunogenic proteins, a second generation array that consists of a much larger number of S. mansoni proteins, both extracellular and intracellular, would likely yield many more immunogenic proteins, including potential vaccine and diagnostic antigens. This is particularly relevant in the context of antigen discovery using sera from individuals who have developed resistance to schistosomiasis after repeated rounds of PZQ therapy (drug induced resistance - DIR), where the immune response is primarily aimed at intracellular molecules released by dying worms [72] or other means of protein export such as exosomes [73]. We are now screening our array (and subsequent generation arrays) with sera from DIR individuals. Indeed, if a schistosomiasis vaccine is developed, it is likely to be incorporated into an integrated control program that couples chemotherapy with vaccination [19], so a comprehensive assessment of the targets of DIR immunity will prove to be an essential component of future schistosomiasis vaccine antigen discovery.
We have shown here that proteome microarrays provide an ideal means by which to explore humoral immunity and vaccine antigen discovery for parasitic helminth infections. The approach is less labor intensive and more sensitive than traditional immunoproteomics based approaches that employ 2D Western blots followed by protein extraction from SDS gels [72]. Moreover, antigens can be readily up-selected for their protective properties and down-selected for potentially deleterious allergic properties, in a high-throughput fashion with large numbers of sera. The power of this technology lies with the nature of the assembled cohorts – whether they are well-characterized groups of naturally resistant and susceptible individuals, or animals that have been experimentally rendered resistant by vaccination (e.g. irradiated schistosome cercariae [74], [75]). With the recent sequencing of the S. haematobium genome [27], and the enormous burden of disease that is attributed to urogenital schistosomiasis in Sub-Saharan Africa [76], it is now essential to apply a systems vaccinology approach to the integrated control of all the major schistosomes infecting humans. Future efforts will explore the replication of conformational epitopes in prokaryotic (as done herein) versus eukaryotic (eg. insect cell lysates) RTS systems, and larger protein microarrays containing the entire parasite secretome will be produced to allow a more comprehensive screen and ensure that a pipeline of schistosomiasis vaccine antigens is generated for progression towards clinical trials.
All subjects provided written informed consent using forms approved by the Ethics Committee of Centro de Pesquisa René Rachou (reference number Fev/04), the Federal Institutional Review Board of Brazil or CONEP (25000.029.297/2004-58) and the George Washington University School of Medicine Institutional Review Board (GWUMC IRB# 100310).
Individuals aged 18–60 (inclusive) from a S. mansoni endemic areas in Minas Gerais State, Brazil were followed longitudinally. Individuals were determined to be Putative Resistant (PR) if they had regular contact with infected water as determined by water contact studies and surveys for infected snails [26], and no S. mansoni eggs in their feces after three days of examination of fecal samples by ether sedimentation and Kato Katz fecal thick smears (2 slides per fecal sample) over a period of 10 years of investigation (n = 20). The PR group were matched with sera taken from individuals deemed to be chronically infected (CI) with S. mansoni as determined by the fecal exam methods described above and stratified in the following groups by the intensity of S. mansoni infection as expressed in eggs per gram of feces (epg) by Kato Katz fecal thick smear: CI-Light (n = 30; epg <100), CI-Moderate (n = 20; epg = 101–400) and CI-Heavy (n = 20; epg>401). Individuals in each CI intensity strata were age, sex, and water contact matched with a PR individual. The PR and some of the CI sera were the same as those used in Tran et al. [26]. We included negative control groups from non-endemic areas of both Brazil (Belo Horizonte, Minas Gerais; n = 12) and the U.S. (n = 12). Donor sera from the U.S. were taken at the George Washington University under an IRB approved protocol. Table S6 contains the demographic information on the different cohorts utilized in this study.
A subset of potentially immunogenic open reading frames (ORFs) were selected for expression and printed from publically available coding sequences for S. mansoni (n = 63) and S. japonicum (n = 214) [38]. Most of these sequences were selected based on bioinformatic, proteomic and transcriptomic data using the following criteria: high sequence homology among the two schistosome species; expression in the immunologically vulnerable schistosomulum stage; predicted or known to be localized on or in the parasite tegument; and limited sequence similarity with mammalian homologs. Primer design and PCR amplification from S. mansoni and S. japonicum cDNA libraries were performed as described [38]. Amplicons were cloned into the custom made pXi T7 vector containing N-terminal 10-His and C-terminal HA tags by homologous recombination, as described previously [77]. Of the sequences selected, 88% (n = 244) were successfully amplified and the resultant plasmids purified, and the inserts were verified by PCR and sequencing [38] (Table S1). A total of 217 high-yielding plasmids (45 from S. mansoni and 172 from S. japonicum) with correct inserts were expressed in an in vitro cell-free system based on Escherichia coli ribosomes (Roche RTS 100), and the protein extracts were contact-printed without purification onto nitrocellulose glass ONCYTE slides. As controls, the following purified recombinant antigens expressed in yeast or E. coli were printed onto the array in two dilutions (0.1 and 0.3 mg/ml): Sm-TSP-2 (Smp_181530), Sm29 (Smp_072190), Sj-TSP-2/238 (NCBI ABQ44513) and Sj silencer (NCBI AAP06461). Non-schistosome control proteins/RTS reactions were also spotted onto the microarray as described [38], and included Epstein Barr virus protein EBNA-1 as well as purified human immunoglobulins. The printed in vitro expressed proteins were quality checked using antibodies against incorporated N-terminal poly-histidine (His) and C-terminal hemagluttinin (HA) tags. The efficiency for in vitro expression was higher than 95%, where positive features were considered to have detectable His or HA tags (Figure S1).
Sera were pre-adsorbed for anti-E. coli antibodies by rocking for 30 min at RT with E. coli lysate before probing of arrays. Protein arrays were blocked in blocking solution (Maine Manufacturing) for 2 hours at RT prior to probing with human sera (diluted 1∶50) at 4°C overnight with gentle constant rocking [71]. Arrays were washed 3 times for 5 min with TBS/0.05% Tween 20 (TTBS) then isotype and subclass specific responses were detected using biotinylated monoclonal antibodies against human IgG1, IgG3, IgG4 (Sigma) and IgE (Hybridoma Reagent Laboratory, Baltimore, MD) diluted 1∶100 for 2 h at RT. Arrays were washed again then incubated for 2 h in streptavidin Cy-5 diluted 1∶400 and washed with TTBS followed by TBS then MQ water, 3×5 min in each solution. Air-dried slides were scanned on a Genepix 4200AL scanner (Molecular Devices) and signal intensities (SI) quantified using the ScanArray Express Microarray Analysis System Version 3.0 (Molecular Devices). Raw SI were corrected for spot-specific background using the Axon GenePix Pro 7 software. Data were analyzed using the “group average” method [78] whereby the mean SI was considered for analysis. Briefly, the SI for negative control spots (empty vector) was calculated for each serum and each antibody type. This value was considered as the background and was subtracted from the SI of each protein spot. To determine if an antigen was recognized a cut-off for each antibody type was defined. The cut offs were defined as one standard deviation above the average of the negative control spots for all groups of sera (Negative BZL, Negative USA, PR, CI-Light, CI-Mod, CI-Heavy) after correcting for spot-specific background. The cut offs were- IgE: 2828.8; IgG4: 2767.3; IgG3: 1557; and IgG1: 316.4.
All 217 RTS proteins as well as purified E. coli-derived Sm29 and yeast-derived Sm-TSP-2 were used to conduct a spatial analysis. When mean SI was below the cut-off for a given antibody isotype/subclass, the SI value was adjusted to zero for this analysis only. To identify clusters containing proteins with the same antibody reactivity profiles we generated a distance matrix estimated from the pairwise Euclidian distance of log transformed SI for each antigen based on the cut-off values for each antibody isotype/subclass in the different cohorts. Complete linkage clustering methodology was used to create a dendrogram analysis of pairwise Euclidian distances for each protein according to the equation below:
Where D(i,j) represents the distance between the proteins i and j, E is the antibody isotypes (IgG1, IgG3, IgG4 and IgE) and S is the set of individual subjects; f iap is the fluorescence signal relative to antibody isotype present in the serum of subject p reactive against protein i; f jpa is the fluorescence signal for the same sample against protein j. To provide a visual representation of each distance matrix, we used a multidimensional scaling (MDS) plot with two dimensions (2D). The unsupervised methodology k-means algorithm with 1,000 interactions [40] was used to define seven clusters. Clusters were validated using clValid, a R software package for cluster validation [79]. The distance matrix, MDS, clustering and graphing were performed using the R software platform (www.r-project.org) [80]. Graphics representing specific relativities that characterized each cluster were designed using GraphPad Prism 5.0. The raw SI values grouped into clusters are provided in Table S2.
Kruskal-Wallis with Dunn's multiple comparison test was used to compare more than two independent samples (Figure 1) and to calculate the statistical differences between the groups if the protein was classified as reactive (Figures 2–5). Correlations were calculated using the Spearman Test. Statistical analyses were performed with GraphPad Prism 5.0.
|
10.1371/journal.ppat.1002141 | Thy1+ Nk Cells from Vaccinia Virus-Primed Mice Confer Protection against Vaccinia Virus Challenge in the Absence of Adaptive Lymphocytes | While immunological memory has long been considered the province of T- and B- lymphocytes, it has recently been reported that innate cell populations are capable of mediating memory responses. We now show that an innate memory immune response is generated in mice following infection with vaccinia virus, a poxvirus for which no cognate germline-encoded receptor has been identified. This immune response results in viral clearance in the absence of classical adaptive T and B lymphocyte populations, and is mediated by a Thy1+ subset of natural killer (NK) cells. We demonstrate that immune protection against infection from a lethal dose of virus can be adoptively transferred with memory hepatic Thy1+ NK cells that were primed with live virus. Our results also indicate that, like classical immunological memory, stronger innate memory responses form in response to priming with live virus than a highly attenuated vector. These results demonstrate that a defined innate memory cell population alone can provide host protection against a lethal systemic infection through viral clearance.
| Immunological memory is a hallmark of adaptive immunity and provides the basis for our ability to become ‘immune’ to pathogens to which we have previously been exposed, and provides the basis for vaccination. For decades, the paradigm held that only the classical adaptive lymphocytes were capable of forming and maintaining protective immunological memory. Recently, several papers have shown the capacity of an innate cell population, a subset of natural killer (NK) cells, to exhibit certain aspects of immunological memory. Here we show that innate memory forms in response to infection with vaccinia virus and resides in a discrete subset of NK cells. We further demonstrate that this innate memory provides significant host protection against a subsequent systemic infection with a lethal dose of vaccinia virus, in some cases resulting in the complete clearance of detectable virus. We also demonstrate that priming with live, replicating virus stimulates innate memory more robustly than a highly attenuated vector. These findings shed new light on this emergent area of immunology, and hold significant implications for harnessing innate memory as part of creating novel vaccination strategies.
| Immunological memory allows the immune system to provide enhanced host protection upon secondary exposure to an infectious pathogen. Memory has long been considered the sole province of adaptive lymphocytes. Lymphocytes recognize pathogens via unique somatically-rearranged antigen receptors, expand clonally upon activation, and eventually give rise to a population of long-lived progeny. These progeny cells maintain their antigenic specificity and exhibit enhanced functional activity upon secondary exposure to a priming pathogen.
Recent studies have suggested that a reconsideration of the classical paradigm of immunological memory is warranted. These studies have shown that innate cell populations have the capacity to generate enhanced responses upon secondary exposure to the priming immunogen[1], [2], [3]. O'Leary, et al., demonstrated that NK cell-mediated delayed-type hypersensitivity (DTH) responses[1] can be generated upon secondary exposure to sensitizing compounds. Further, they showed that these compound-specific DTH responses were mediated by a liver-resident NK cell population expressing the Thy1 (CD90) molecule. Recently, Sun, et al. demonstrated that an innate memory response forms to MCMV and is mediated by a population of NK cells expressing Ly49H. In mice containing the B6 haplotype NK complex, NK cells expressing the germline-encoded NK receptor Ly49H—an activating NK receptor identified that is capable of specifically recognizing a virally-encoded product (the MCMV protein m157)[4], [5]—are the predominant contributors to the innate response to a primary MCMV infection[2], [6], [7], [8], [9], [10], [11]. Sun et al., made use of the cognate recognition of m157 by Ly49H+ to establish the specificity of the enhanced host protection provided by memory Ly49H+ NK cells upon re-exposure to MCMV[2].
Recently, a report from the von Andrian laboratory [3] showed that CXCR6+ NK cells primed with virus-like particles (VLPs) expressing viral transgenes were capable of mediating antigen-specific contact hypersensitivity (CH) in response to antigens not known to be recognized by germline-encoded receptors. These memory NK cells were also capable of providing partial host protection from infection with live virus expressing the priming antigens, as measured by a delay in mortality upon lethal viral challenge and the capacity to resolve localized, lower dose viral infection. Further, they demonstrated that signaling through CXCR6, a chemokine receptor that binds to CXCL16 (primarily expressed by hepatic sinusoidal epithelium), is critical for the maintenance of this hepatic NK memory population of cells.
The present studies were initiated to determine whether innate memory could contribute to the control of viral pathogens for which no cognate germline-encoded receptor has been identified. Furthermore, we wanted to assess whether innate memory cells alone can provide protection against a viral challenge in the absence of classical adaptive immunity. In this series of experiments, we used live vaccinia virus to prime and challenge the memory NK cell population and demonstrate that memory NK cells alone are capable of providing sterilizing protection against a systemic challenge with a fatal dose of the priming pathogen. This protection was mediated by a Thy1+ subset of hepatic NK cells primed with live vaccinia virus; hepatic Thy1+ NK cells primed with a highly attenuated strain of vaccinia (Modified Vaccinia Ankara (MVA)) were unable to confer this protection upon adoptive transfer into naïve, immunodeficient hosts. This suggests that innate memory, like classical adaptive memory, is primed more efficiently with live, replication-competent organisms. We demonstrate that innate memory to vaccinia virus is extremely durable, as the memory NK cells that mediated clearance of the lethal pathogen were active greater than 6 months after priming with the live pathogen. We also show that on a cellular level, enhanced activation of hepatic NK cells from virally primed mice can be demonstrated not just in response to the pathogen, but in response to more generalized stimuli.
We chose vaccinia virus (VV) as a pathogen for these studies because immunocompetent mice resolve vaccinia virus infection through a combination of humoral and cell-mediated adaptive immune responses[12]. Moreover, the virus is highly pathogenic in naïve mice that lack classical adaptive immune function. By infecting mice with a recombinant vaccinia virus expressing the transgene firefly luciferase (rVV-luc), we were able to monitor the infection in vivo by periodically injecting luciferin-D (the luminescent substrate for firefly luciferase) into the mice and using the IVIS imaging system to monitor viral burdens and virus localization.
First, we infected mice in which the IgHμ constant region has been disrupted (the μMT mouse; IgHko)[13] which results in a consequent failure to develop mature B cells. The genetic block in B cell development is extremely stringent when in C57Bl/6 (B6) mice[13], [14] (data not shown). We used the IgHko B6 mice for these studies to eliminate the possibility that a humoral immune response might develop during a primary infection with vaccinia virus that could confer protection against a secondary exposure to virus. We also reasoned that T lymphocytes present in these mice would clear a primary vaccinia virus infection efficiently; RAG1ko mice (mice that lack both T and B cells) succumb to primary infection with vaccinia virus at very low doses. In using the IgHko B6 mouse model, we could prime mice in the presence of T cells, thereby allowing the IgHko mice to survive the primary infection, and then eliminate cell subpopulations in vivo by administration of monoclonal antibodies prior to secondary challenge. Antibody-mediated depletion of T cells could be maintained indefinitely in the IgHko mice by repeated administration of depleting antibodies since these mice cannot generate anti-Ig antibody responses that might interfere with the infused antibodies.
IgHko mice received an inoculation of either PBS (control naïve IgHko), or 1×107 plaque forming units (pfu) of rVV-luc intraperitoneally (ip) (primed IgHko)). We then rested the mice for a minimum of six months after clearance of the rVV-luc (as determined by IVIS imaging; Fig. S1) to ensure that any effects we observed during a secondary challenge would reflect the influence of a durable, long-lived memory immune response rather than the activity of residual effector cell populations that may have persisted following clearance of the virus. Ten days prior to a secondary vaccinia virus challenge we began administering isotype control antibodies or a cocktail of T cell- depleting monoclonal antibodies (clones GK1.5 (αCD4)[15], H57–597 (αTCRß)[16], and UC7-13D5 (αTCRγδ)[17]) ip to groups of naïve and primed animals. We did not include a CD8α-depleting monoclonal antibody in this cocktail because we did not want the depletion of CD8α+ innate populations of cells to potentially confound the outcome of the experiments. The monoclonal antibody mixtures were administered every two weeks throughout the life of the experiment to maintain T cell depletion. The efficiency of T cell depletion was monitored by flow cytometric analysis of peripheral blood stained with a panel of monoclonal antibodies, including antibodies specific for CD3, TCRβ, TCRγδ, and NK1.1. Representative flow cytometry plots from control and T cell-depleted mice 1 week post-challenge showed that the efficiency of T cell depletion was extremely high, greater than 99% as measured by percentage of CD3+ within the lymphocyte gate (Fig. 1a). Very few of those cells that fall within the CD3+ gate in the depleted mice showed staining for TCRβ, TCRγδ, or CD8α; there are no masking antibodies to CD3 or CD8α present in the cocktail. Moreover, pilot studies performed in control B6 mice showed that the extent of T cell depletion in the peripheral blood following administration of the depleting antibodies reflected the T cell depletion in spleen and lymph nodes (data not shown).
In repeated experiments, we observed that both primed (data not shown) and naïve IgHko mice treated with isotype control antibodies were capable of clearing vaccinia virus infection. However, in the T cell-depleted groups of mice, only the primed mice (primed, T-depleted IgHko) were capable of resolving the infection (Fig 1, b and c). Representative IVIS images show the kinetics of vaccinia virus clearance in each of the groups of mice (Fig. 1b). The luminescence measurements on all mice demonstrated that while naïve, T-depleted IgHko were incapable of controlling vaccinia virus after challenge, the primed T-depleted IgHko mice resolved the infection (Fig. 1c). The clearance of vaccinia virus infection occurred within 2–3 weeks in primed, T cell-depleted mice (Fig. 1, b and c), more slowly than virus clearance occurred in primed, control IgHko mice (clearance in 1 week-data not shown) or in naïve, control IgHko mice (clearance in 10–14 days; Fig. 1, b and c). Once the primed, T cell-depleted mice had resolved the vaccinia virus infection, there was no evidence of a later recrudescence of viral replication. In mice that succumbed to vaccinia (naïve, T cell-depleted IgHko), we observed spread of vaccinia virus from the peritoneal cavity into hotspots located on the tails, feet, and mouths of infected mice. This distribution of vaccinia virus was consistent with previous reports[12] and was likely the result of transfer of infection to satellite sites via saliva. Together, these data suggest that the long-term resolution of vaccinia virus infections can occur in the absence of classical adaptive immune effector mechanisms.
We next sought to identify the cell population(s) that mediated this non-T and non-B cell protective memory response in vivo. Since Thy1+ lymphokine-activated killer (LAK) cells have been implicated in protection against infection with vaccinia virus[18], [19], [20], [21], [22], and O'Leary, et al., have shown that the multiple compound-specific DTH responses they observed are mediated by a liver-resident Thy1+ NK cell population[1], we hypothesized that Thy1+ NK cells may be the effector population providing protection against secondary vaccinia virus infection. To explore this possibility, we inoculated groups of IgHko mice with either 1×107 pfu rVVluc (primed IgHko) or PBS (naïve IgHko) ip. Eight months later, these mice were treated either with isotype control antibodies, the T cell-depleting antibody cocktail (300 μg each of clones GK1.5, H57–597, and UC7-13D5), or a T- and Thy1-depleting antibody cocktail (300 μg each of clones GK1.5, H57–597, UC7-13D5, and the Thy1.2 (CD90.2)-specific clone 30H12[23]); the monoclonal antibody mixtures were administered every 2 weeks throughout the course of the experiment to maintain T and Thy1+ cell depletion. One week after initiating monoclonal antibody administration, all groups of mice were challenged ip with 1×106 pfu rVV-luc. Representative flow cytometric plots of peripheral blood cells from groups of naïve and primed mice 2 weeks after challenge demonstrate that the depletion of the target cell populations by monoclonal antibody administration was extremely efficient (>99% T cell depletion in all groups at 2 weeks post-challenge; Fig. 2a). Depletion of T cells had no significant impact on the number of NK cells present in the peripheral blood. We did observe a significant shift in the representation of Thy1 (CD90)+ NK cells in mice that had been depleted of T cells. Nevertheless, the proportion of Thy1+ NK cells was comparable in the naïve and primed groups (Fig. 2b), suggesting that differences in clearance and survival between the naïve and primed T-depleted groups is not a consequence of differential effects of T cell depletion on the NK cell populations. As anticipated, the addition of the Thy1.2 (CD90.2)-depleting monoclonal antibody 30H12 in the T&Thy1-depleting cocktail efficiently eliminated all Thy1+ NK cells in IgHko recipients (Fig. 2b). Both the naïve and primed IgHko control (T cell-competent) groups of mice cleared the infection with kinetics consistent with primary (naïve IgHko) and secondary (primed IgHko) adaptive immune responses, respectively (Fig. 2, c and d). As observed in the initial series of experiments, the naïve, T-cell-depleted IgHko mice were unable to control the challenge infection while the primed, T cell-depleted IgHko mice resolved the vaccinia virus infection within 2–3 weeks. However, depletion of both T cells and Thy1+ non-T cells abrogated the protection that was evident in the primed T-cell-depleted IgHko group (Fig. 2c and d). The naïve T- and Thy1-depleted mice also succumbed to infection. These data suggested that the cell population(s) mediating protection against secondary vaccinia virus challenge was a Thy1+, non-T, non-B cell population.
To examine the comparability of data generated using the luciferase/IVIS system and data generated using a traditional strategy for measuring vaccinia virus infection in vivo, we quantified vaccinia virus titers in ovaries of primed T-depleted and primed T- and Thy1-depleted groups of mice. The results established that the clearance of luciferase activity in primed T-depleted animals as visualized using the IVIS technology correlated with an inability to detect vaccinia virus in ovaries of these animals (Fig. S2 and data not shown). This observation indicates that viral clearance as measured by IVIS is consistent with traditional methods of measuring viral clearance.
One of the fundamental characteristics of a classical memory response is that memory cells exhibit enhanced activation upon secondary stimulation. Therefore, we wanted to determine if functional differences existed between NK cells from naïve and vaccinia virus-primed mice. To do so, we performed in vitro activation assays using isolated liver mononuclear cell preparations. We pooled liver mononuclear cells from groups of 8–12 age-matched naïve mice and mice that had been VV-primed 6 months earlier to ensure that we could isolate sufficient numbers of cells for stimulation and analysis. These cell preparations were isolated from enzymatically-dissociated livers via percoll gradient separation and 5×105 cells/well were aliquoted into 24 well plates. The cells were then incubated for 6 hours in the presence of fluorescently-labeled anti-CD107a/b antibodies in wells containing either no stimulus, PMA and ionomycin, 200 micrograms of plate-bound isotype control antibody (clone CI.8), 200 micrograms of plate-bound anti-NK1.1 antibody (clone PK-136), or either 2×106 pfu of vaccinia virus or 2×106 viral particles (vp) of recombinant adenovirus that had been treated with 2% paraformaldehyde. The recovered cells were washed, stained for CD69 expression, and fixed prior to flow cytometric analysis. In each condition, analysis on gated NK cells (NK1.1+CD3−) showed that the NK cells from vaccinia virus-primed animals showed higher activation, as determined by CD69 (activation) and CD107a/b (degranulation) staining (Fig. 3). Primed NK cells showed significantly higher levels of activation and degranulation than naïve NK cells when stimulated with PMA and ionomycin (24.1% CD69+CD107+ for primed NK vs. 14.3% for unprimed; specific activation above unstimulated 13.5% for primed vs. 7.1% for unprimed NK), and plate-bound anti-NK1.1 antibody (40.8% primed vs. 22.1% unprimed; specific activation above isotype-stimulated 29.3% primed vs. 14.3% unprimed). This activation occurred in response to the indicated stimuli, as activation was well above control levels, and by stimuli that were not dependent on a presenting cell population or the presence of vaccinia virus antigens. The cells also showed an enhanced response to fixed, plate-bound vaccinia virus when compared to the unstimulated control cells (16.5% primed vs. 9.9% unprimed; specific activation above unstimulated 5.9% primed vs. 2.7% unprimed), although the response to vaccinia virus was only marginally higher than the response to an unrelated virus (16.5% primed vs. 9.9% unprimed responded to vaccinia virus, while responses to adenovirus were 14.2% primed vs. 8.6 unprimed; vaccinia-induced activation relative to adenovirus-induced activation 2.3% primed vs. 1.3% unprimed). These results showed that, as with the non-specific stimuli, primed NK cells exhibited significantly enhanced activation relative to unprimed NK cells in response to direct exposure to viral particles. However, the relatively small increases in activation seen in both primed and unprimed cells in response to vaccinia virus as compared to an unrelated virus (adenovirus) suggest that direct and specific recognition of vaccinia virus antigens plays a limited role in stimulating both vaccinia –primed and –unprimed NK cells. Collectively, these results show that NK cells from vaccinia virus-primed mice were more responsive than NK cells from naïve mice to a wide range of activating stimuli.
We next characterized the dynamics of NK cell populations throughout the course of vaccinia virus infection in unmanipulated, wild type mice. B6 mice were administered PBS (control) or 1×107 pfu of rVV (challenged) ip. At various time points after infection, groups of control and challenged mice were sacrificed, and cells from the peripheral blood, spleen, and liver were isolated, counted, and analyzed by monoclonal antibody staining and flow cytometry. A rapid expansion was seen in the absolute number of NK cells in the spleen (data not shown) and liver (Fig. 4a) during the first week following challenge, followed by a contraction of this population as the infection resolved. By 7 days following infection, the absolute number of liver NK cells (CD3−NK1.1+) had increased 5-fold. There was a preferential expansion of the Thy1+ NK cell population in the liver of infected mice, with the representation of Thy1+ cells increasing in the liver both as a percentage of total NK cells (from 46% to 64% of total liver NK cells by day 4 post-challenge) and in their absolute number (a 10-fold increase over baseline by 7 days after infection). A preferential expansion of splenic Thy1+ NK cells was also observed, albeit to a lesser degree (approximately 2–fold expansion in the total number of splenic NK cells, and a 5-fold expansion in the number of Thy1+ splenic NK cells; the percentage of total splenic NK cells that were Thy1+ rose from a baseline of 32% to a peak of 57% by 4 days post-infection; data not shown).
We also undertook a detailed phenotypic analysis of the liver NK cells following vaccinia virus infection. These cells were stained with antibodies specific for NK-associated cell surface molecules, including CD94, NKG2D, CD43, Ly49H, Ly6C, KLRG1, CD27, CD11b, and Thy1 (Fig. 4, b and c). During the first week after vaccinia virus challenge, the absolute number of all NK subsets we analyzed had increased. However, we were able to identify several NK cell subsets that preferentially increased in number and percentage: the total Thy1+ NK cell population, as described above, and a subpopulation of Thy1+ NK cells that were also KLRG1hi. Most of these cells were phenotypically NKG2DbrightLy6ChiCD27loCD11bhiCD43hi (Fig. 4, b and c, and data not shown). The representation of this KLRG1hi population rose from approximately 5% to a peak of 20–25% of the liver Thy1+ NK cell population by days 4–7 after infection, and then contracted to baseline levels by 21 days after challenge (Fig. 4b and c). However, by 118 days after challenge there were no phenotypically discrete subsets within the Thy1+ NK cell subset that we could identify as potential memory populations.
We wanted to determine whether the expansion of Thy1+ NK cells observed in the liver and spleen during primary infection were a consequence of proliferation or an alteration in NK cell trafficking. Groups of B6 mice were infected with 1×107 pfu rVV ip , and at various time points administered the thymidine analogue ethynyl deoxyuridine (EdU) 12 hours prior to sacrifice. We designed the experiment to allow the assessment of the proliferative activity within the NK cell population at various time points rather than following continuous EdU administration. As shown in Fig. 5, extensive proliferation within the NK cell population occurred early in infection (25% of total NK cells proliferating between days 3–4 post-challenge), and returned to baseline levels by 7 days post-challenge (Fig. 5a). The preferential expansion of Thy1+ NK cells was seen at both day 4 and 7 post-challenge (where the percentage of EdU+ NK cells that were also Thy1+ reached levels of 74.6% and 82.9% respectively, compared to 49% in naïve controls; Fig. 5b). These results are consistent with the population dynamics observed in the liver NK cell population (Fig. 4a), and suggest that the large increases in NK cell number and Thy1+ NK cell percentage in liver are driven by in situ proliferation early in infection. In support of the mechanism of in situ proliferation, plaque assays performed on de novo infected animals showed that low levels of vaccinia virus could be recovered from the livers of animals receiving vaccinia virus ip beginning at 4 hours post-infection and through day 4 post-infection (ranging from 4×102–5×104 pfu; data not shown). This finding established that pathogen is also present in situ and might directly stimulate the expansion of Thy1+ NK cells.
Our initial experiments in the IgHko model system indicated that a VV-primed Thy1+, non-T- non B- cell population was capable of providing host protection upon secondary exposure to the virus. However, despite the extreme efficiency of depletion (>99%; Fig. 1a and Fig. 2, and b) achieved in the IgHko system, in vivo depletion via monoclonal antibody administration was not 100% efficient. Therefore, it remained a formal possibility that the protection we observed might be mediated by residual T cells. To address the possibility of residual T cell contamination, and confirm that a population of memory Thy1+ NK cells could confer protection against vaccinia virus, we evaluated the anti-viral activity of these cells in an adoptive transfer system. First, we wanted to establish that any protection we might observe in mice receiving these transferred cells was mediated by the transferred NK cell populations and not by contaminating T cells. To determine if we could adequately control for contaminating T lymphocytes in the transfers, we isolated NK cell-depleted liver mononuclear cell preparations from enzymatically-dissociated livers of naïve and vaccinia virus-primed mice by percoll gradient separation followed by magnetic separation of non-NK cells. We then transferred 2×106 of these NK-depleted mononuclear cells (a mixture of T cells, B cells, macrophages, DCs, and granulocytes) into naïve RAG1ko mice; one group received naïve NK-depleted cells, one group received VV-primed non-NK cells, and a third group received VV-primed non-NK cells in conjunction with the administration of the T cell-depleting monoclonal antibody cocktail utilized in the IgHko depletion experiments described above. Five days after the transfer of these cells, we challenged the recipients with 1×105 pfu of rVV-luc ip and monitored their ability to control infection using the IVIS system. Mice that received naïve or VV-primed NK-depleted cells cleared the infection with kinetics consistent with a primary (naïve recipients) or secondary (primed recipients) T cell response (Fig. 6). However, the mice that received the T cell-depleting monoclonal antibody with the VV-primed non-NK cells were unable to control infection and had to be sacrificed at d34 post-challenge. These results indicate that even when transferring large numbers of primed T lymphocytes into naïve RAG1ko hosts, administration of the T cell-depleting monoclonal antibody cocktail abrogated the protection afforded by the transferred cells.
We then isolated NK cells from the collagenase-treated livers of B6 or B6 congenic mice a minimum of 6 months after these mice had received either PBS (naïve) or 1×107 pfu rVVluc (primed) ip. We utilized two congenic mouse models in these studies; B6.PL mice, which express the CD90.1 isoform of CD90, or B6.SJL mice, which express the CD45.1 isoform of CD45. The B6, IgHko, and RAGko mouse models utilized in these studies express CD90.2 and CD45.2. Blood mononuclear cells were isolated from livers after enzymatic digestion followed by Percoll gradient centrifugation, and NK cells were isolated from this cell fraction using the MACS NK cell isolation kit and autoMACS instrument. Purified NK cells were then separated into Thy1− and Thy1+ populations. Three distinct cell fractions were adoptively transferred into naïve RAG1ko mice: 5×106 NK-depleted cells (a mixture of T cells, B cells, macrophages, DCs, and granulocytes), 1×105 Thy1− NK cells, and 1×105 Thy1+ NK cells. The groups of mice receiving the NK cell populations were treated with the T cell-depleting monoclonal antibody cocktail at the time of transfer to ensure that any contaminating T cells were eliminated. Thy1+ NK cells, but not T cells could be detected in the spleen or peripheral blood of RAG1ko hosts that received purified liver Thy1+ NK cell transfers 4 weeks after transfer and challenge (Fig. S3); as expected, donor T cells were readily apparent in the peripheral blood of mice receiving NK-depleted liver cell transfers (Fig. S3), but were not detectable in mice receiving NK cell transfers.
Five days after cell transfers (and concurrent T cell-depleting antibody administration for recipients of purified NK cell population transfers), we challenged these RAG1ko mice with 1×105 pfu rVV-luc ip and monitored the mice for viral burdens using the IVIS system. Representative IVIS images of individual mice from each of the experimental groups are shown in Fig. 7a, and the persistence or clearance of virus in each mouse from each group in 4 separate experiments are summarized in Fig. 7, b and c. As predicted, the RAG1ko mice that received naïve, NK cell-depleted liver lymphocytes (15/16 mice; 94%) resolved the vaccinia virus challenge within 2 weeks, and the RAG1ko mice that received the VV-primed NK cell-depleted liver lymphocytes (15/15; 100%) resolved the vaccinia virus challenge within 1 week (Figs. 7b and c, and data not shown). However, the groups of RAG1ko mice that received naive Thy1− (1/16; 6.25%) or Thy1+ (0/17; 0%) NK cell populations, or VV-primed Thy1− (0/16; 0%) NK cells were unable to resolve vaccinia virus infections. Strikingly, a significant proportion of RAG1ko mice that received VV-primed Thy1+ NK cell transfers (9/22; 41%) were able to clear the challenge vaccinia virus infection by three weeks after challenge, with viral clearance provided by transferred VV-primed Thy1+ NK cells significantly enhanced over mice receiving VV-primed Thy1− NK cells (two-tailed t-test; p<0.01), naïve Thy1−NK cells (p<0.05), and naïve Thy1+ NK cells (p<0.01) transfers (Fig. 7b and c). Survival in these groups mirrored the clearance we observed by IVIS (Fig. 7d). All mice that received NK cell-depleted cells in transfer (cells that contained T lymphocytes) survived; however, of NK cell transfer recipients, only those that received the VV-primed Thy1+ NK cells were afforded significant protection, as shown by a survival rate of greater than 40% (9/22 recipients). We also wanted to determine if a highly attenuated poxvirus strain, MVA, was as capable of priming for protective NK cell memory as replication competent VV (Western Reserve strain). We transferred 1x105 Thy1− or Thy1+ NK cells isolated from the livers of naïve or MVA-primed RAG1ko mice into naïve RAG1ko recipients, then challenged the recipients with 1×105 pfu rVV-luc IP 5 days after NK transfer. As shown in Fig. 7e, no significant protection against a replication-competent VV was observed in any of the recipients. These data indicate that VV-primed, but not MVA-primed, Thy1+ NK cells can clear a lethal systemic challenge with VV.
To verify that the protection observed in these studies is afforded by the transferred Thy1 (CD90)+ NK cells and not contaminating T lymphocytes, we isolated either NK-depleted liver lymphocytes (2×106/transfer) or liver Thy1+NK cells (1×105/transfer) from B6.SJL (CD45.1+CD90.2+) mice and transferred them into naïve RAG1KO mice (CD45.2+CD90.2+) with simultaneous administration of the T depleting monoclonal antibody cocktail to the mice receiving the transferred NK cells. Five days after transfer, recipients were infected with 1×105 pfu rVV-luc ip The course of the vaccinia virus infection in these animals is indicated by both IVIS images and classical pfu assay (Fig. 8). We were able to establish that the control (d7 post-challenge; Fig. 8a) and resolution (d14 post-challenge; Fig. 8b) of challenge vaccinia virus infection observed in animals receiving primed B6.SJL Thy1+ (CD45.1+CD90.2+) NK cell transfers occurred in the absence of contaminating T cells (CD45.1+CD3+) and in the presence of the transferred Thy1+ NK cells (CD3−NK1.1+CD45.1+CD90+). The protection and clearance of virus mediated in vivo by adoptively transferred B6.SJL NK-depleted liver lymphocytes (including T cells; CD3+CD45.1+CD90.2+) at d14 post-challenge is also shown (Fig. 8b). These results indicate that the protection observed against vaccinia virus challenges in recipients of primed Thy1+ NK cells is mediated by the transferred NK cells and is not a consequence of contaminating T lymphocytes in the transferred cells.
In these studies, we show that innate memory develops in response to a viral infection in the absence of an identified cognate receptor-virus interaction. The enhanced control of vaccinia virus in primed, T-depleted IgHko mice was apparent by 4 days post-challenge, indicating that the innate responses were both potent and rapid. That we observed this innate protection in animals that had resolved primary infection greater than 6 months prior to the secondary challenge has two important implications. First, it suggests that the protection represents a ‘memory’ response, rather than the persistence of a residual population of activated effectors. Second, the innate memory that formed following exposure to vaccinia virus is extremely durable. We further established that the complete protection we observed in the primed, T- depleted IgHko mice was abrogated with the addition of the Thy1-depleting antibody to the T cell depletion cocktail, indicating that innate memory to vaccinia virus resides within a Thy1+ non-T and non-B cell population. Perhaps the most striking result we observed in these experiments was that the memory provided by this innate Thy1+ cell population was manifested not only in enhanced control of vaccinia virus infection during the early stages of infection, but proved to be sufficiently potent to protect the infected hosts against a systemic lethal dose of virus in the absence of classical adaptive immune cells.
We performed experiments to determine if the memory Thy1+ NK cell that responded to vaccinia virus had a unique, persistent phenotype. In vitro stimulation assays established that the NK cells from primed animals responded more vigorously than NK cells from naïve animals, with increased CD69 upregulation and degranulation upon exposure to both priming antigens (plate-bound vaccinia virus) and a non- antigen-specific stimulus (plate-bound anti-NK1.1). These results indicate that primed hepatic NK cells develop and maintain an enhanced capacity for activation, not just enhanced pathogen-specific recall. These results also suggest that VV-specific memory NK cells may recognize subsequent exposure to vaccinia virus through a combination of signaling related to an induced self profile and direct recognition of virally-encoded proteins. That we observed only small differences in activation between NK cells stimulated with plate-bound vaccinia virus and the unrelated virus adenovirus suggests that the activation of NK cells is not driven predominantly by pathogen-encoded antigens, although it remains a formal possibility that formaldehyde fixation of the virus in these studies may have diminished the capacity of NK cells to interact directly with and recognize immunogenic portions of the virus.
We further sought to define a durable cell surface marker expression profile acquired after a primary infection that might identify a memory cell subpopulation within the liver-resident Thy1+ NK cell compartment. We observed significant increases in total NK cell numbers in spleen (data not shown) and liver, consistent with previous reports of global expansions of NK cell populations after viral infections[7]. We also observed a preferential expansion of the Thy1+ NK cell population, as determined both by the percentage and the absolute numbers of Thy1+ NK cells. Our EdU incorporation experiments (Fig. 5) established that the expansion in total NK cell numbers, and Thy1+ NK cells in particular, is driven by in situ proliferation of Thy1+ NK cells during the first week of infection, not the accumulation of effector cells migrating from peripheral sites. This expansion of Thy1+ NK cells was accompanied by a preferential increase in the representation of a KLRG1+ subset of cells that was predominantly CD27loCD11bhiLy6chi(Fig. 4, b and c), a profile consistent with these cells being mature, activated NK cell effectors[24], [25]. Importantly, in accordance with an earlier report[7], we observed no preferential expansion of the Ly49H+ NK cell subset that mediates innate memory against MCMV[2]. These data suggest that innate memory to vaccinia virus is mediated by a subset of NK cells that is distinct from those that form innate memory to MCMV. However, by 4 months post-challenge, the phenotypic differences between the Thy1+ liver NK cell population in control and vaccinia virus-infected mice observed during the acute phase of infection was no longer apparent. This observation is consistent with a recent report that previously stimulated NK cells retain enhanced functional capacity in the absence of a phenotypic profile that distinguishes these cells from naïve NK cells[26].
A significant concern that must be addressed when assessing the protection against infection described in the present studies is whether any residual functional T cell populations might be present that would be capable of mediating the protection. In the IgHko system, we were able to achieve extremely efficient depletion of T cells as measured by anti-CD3 staining (approximately 99% at 1 week post-challenge). However, while this T cell depletion in the naïve and primed IgHko groups of mice was efficient, complete depletion of all T cells was not possible. Despite the fact that T cell depletion completely abrogated protection in naïve IgHko recipients, it therefore remained a formal possibility that residual T cells in the primed mice could contribute to host protection. We therefore addressed this issue by adoptively transferring highly purified NK cell subsets into naïve RAG1ko hosts and concurrently administering the same T cell-depleting cocktail (Figs. 6-8; Fig. S3).
We were able to demonstrate that protection against vaccinia virus could be transmitted to naïve RAG1ko hosts by adoptive transfer of purified vaccinia virus-primed Thy1+ liver NK cells. Of the naive RAG1ko mice that received 1×105 primed Thy1+ liver NK cells, 41% (9/22) were able to resolve a systemic, lethal vaccinia virus challenge infection, while only 2% (1/49) of RAG1ko mice that received other purified NK cell subpopulations were able to resolve and survive infection (Fig. 7). Our inability to achieve protection in 100% of RAG1ko mice by transfer of primed Thy1+ liver NK cells may be a consequence of limiting numbers of Thy1+ liver NK cells and the inefficiency associated with the cell transfers.
We administered the T cell-depleting monoclonal antibody cocktail to NK cell transfer recipients at the time of transfer to prevent contaminating T cells from contributing to the protection of these recipients. Indeed, as shown in Fig. 6, even when transferring 20-fold more cells —cell preparations that contained a significant proportion of primed T cells, in contrast to the highly purified NK cell populations transferred in these experiments — we observed that the T cell depleting cocktail abolished protection (as measured by both IVIS and survival). Further, when we looked for the presence of contaminating T cells at d7, d14, and d28 post-challenge (d12, d19, and d33 post-transfer and antibody administration, respectively), we were able to identify transferred Thy1+ NK cells but were unable to detect T cells in the peripheral blood, livers, or spleens of NK cell transfer recipients (Fig.8; Fig. S3). These data indicate that the protection we observed in mice that received VV-primed Thy1+ liver NK cell transfers was indeed conferred by this population of cells (Figs. 7 and 8).
We also investigated whether MVA-primed Thy1+ NK cells from RAG1ko hosts were capable of conferring the same protection upon transfer into naïve immunodeficient hosts and determined that priming with the highly attenuated MVA virus was not sufficiently robust to generate protective memory comparable to that induced by priming with live, replication competent VV. There are several possible explanations for why MVA priming did not engender the same degree of protective innate memory. One possibility is that the viral product(s) necessary for NK memory recognition of VV were lost in the attenuation process. A recent study [3] was able to demonstrate antigen-specific memory NK cells induced by a range of virus-like particles (VLPs), including VLPs containing antigens derived from pathogens not endemic to the mouse population (like HIV-1), suggesting that this mechanism is unlikely. Even the highly attenuated MVA strain has a rather large genome encoding an extremely diverse range of potential antigens, including a wide range of antigens shared with the vaccinia virus strain used in the challenges. A second possibility is that NK memory formation to VV requires conditioning from adaptive cell populations at some point during the priming response. However, the ability of RAG1ko hosts to generate protective, pathogen-specific NK memory upon exposure to a variety of VLPs also suggests that the presence of classical adaptive lymphocytes are not necessary during the priming phase for effective innate memory formation. A third possibility is that, as has been observed for classical memory [27], innate memory is generated more potently in response to a live, replication-competent agent. If an ‘active’ infection stimulates a more robust response, it may do so is by stimulating enhanced expression of self-molecules in cells under the stress of active infection and replication. In such a scenario, NK cells might well represent the population best equipped to respond to those stressed-self indicators through their array of germline-encoded receptors.
These data demonstrate that VV infection generates a liver-resident Thy1+ NK cell population capable of mediating a protective innate memory response. It is not clear how primed Thy1+ liver NK cells provide this protection against VV infection. It will be important to determine whether innate memory NK cells respond to pathogens more effectively than naïve memory NK cells because they have enhanced lytic capacity, increased stores of premade effector molecules (such as cytokines, granzyme, and/or perforin), superior proliferative capacity, or an as yet undefined property that facilitates a particularly robust effector response. While the in vitro stimulation assay shown in this study indicates that the primed NK cells do possess enhanced effector capacity (Fig. 3), the precise mechanism by which this is mediated remains unclear.
The mechanism by which primed Thy1+ liver NK cells recognize vaccinia virus upon secondary exposure is of great importance. There are several hypotheses, not mutually exclusive, that could account for pathogen-specific recognition of a previously encountered pathogen. First, there may exist a whole class of previously unidentified receptors expressed by innate cell populations that, like Ly49H, have co-evolved to specifically recognize the various pathogens endemic to the host population. Such molecules could act as highly pathogen-specific pattern recognition PRRs. Indeed, the murine homologue of the human NKp46 receptor (also known as Ncr1) is critical for host defense against influenza[28], and human NKp46 has been shown to bind viral hemagglutinins[29], [30], [31]. Moreover, previous studies have shown that genes within the NK complex on chromosome 6 in B6 mice contribute to NK cell-mediated immune responses to ectromelia[32], [33], an orthopoxvirus closely related to vaccinia virus. This finding suggests that an as-yet-unidentified receptor capable of recognizing a poxviral product may be encoded within this region.
A second potential mechanism by which pathogen-specific recognition could be mediated through germline-encoded receptors could reflect alterations in the ability of NK receptors to interact with MHC class I and MHC class I-like ligands when binding and/or presenting viral epitopes. Alternatively, infection by pathogen could induce a state within the cell that results in the display of an ‘infected-self’- or ‘stressed-self’- associated profile of protein expression that can then be recognized by receptors on innate memory NK cells. One consequence of such an ‘infected-self’ mechanism of recognition would be that the innate memory formed in response to one pathogen might provide protection against challenge with heterologous pathogens that induce a similar molecular self-profile in infected cells. The phenomenon of NK cell recognition and response to induced-self molecules has previously been implicated in both infections and oncogenesis[34], [35], [36], [37], where ligands MULT1[38], H60[39], and Rae-1[39], [40]) for the activating receptor NKG2D are preferentially induced upon infection or cellular transformation[36], [37], [41]. Indeed, activation through NKG2D has been established as a critical component of the NK cell response to ectromelia virus infection[42]. That protective NK cell responses to poxvirus infections have been tied to both an as-yet unidentified gene in the B6 NK receptor complex[32], [33] and to NKG2D activity[42] suggests that both the pathogen-specific PRR and ‘infected-self’ mechanisms may contribute to NK cell responses to poxviruses. Indeed, our in vitro assays showing enhanced responsiveness of vaccinia virus-primed NK cells to stimulation through the NK1.1 receptor pathway as well as direct exposure to vaccinia virus would be consistent with both mechanisms playing a significant role in NK memory cell recognition of poxviruses.
These studies have established a model of innate memory to a pathogen in a system in which there is no known pathogen-specific receptor expressed by innate cell populations. This system will allow us to explore models of innate memory recognition further. Understanding the nature of that recognition will be critical for harnessing innate memory for prophylactic or therapeutic use. The stimulation of innate memory targeted to specific pathogens might represent a novel and powerful approach for the design of future vaccination strategies.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All animals are treated humanely and in accordance with the policies of the Beth Israel Deaconess Medical Center (BIDMC), the regulations of the Animal Welfare Act, and other laws and policies of the federal government and other agencies. All mice were maintained under specific-pathogen-free conditions and research on mice was approved by the BIDMC Institutional Animal Care and Use Committee (IACUC) under protocol #095-2009. All efforts were made to minimize suffering of animals in this study.
Age-matched adult female C57Bl/6, B6.PL (Thy1.1 congenic), B6.SJL (CD45.1 congenic), IgHμko (μMT[13], [14]), and RAG1ko were obtained from Jackson Laboratories (Bar Harbor, ME). All mice were maintained in the Harvard Institute of Medicine and Beth Israel Deaconess Medical Center (BIDMC) Animal Research facilities and used in accordance with protocols approved by the Institutional Animal Care and Use Committees of BIDMC, Harvard Institutes of Medicine, and Harvard Medical School.
Stocks of recombinant vaccinia were prepared as previously described[43]. Briefly, seed stocks of recombinant vaccinia virus (Western Reserve strain) expressing firefly luciferase[44] (rVV-luc; kindly provided by Michael Seaman (BIDMC; Boston, MA USA) were added to HeLa cell monolayers at an MOI of 1. Seeded monolayers were harvested and washed into 10 mM Tris pH 9.0 48–72 hours after seeding, and cell-associated virus was released from cells by 3 freeze-thaw-sonication cycles followed by centrifugation at 850 g for 10 min at 4°C. Supernatants (containing free virus) were then layered over a 36% sucrose cushion and centrifuged for 2 hours at 27,000 rpm (1.33×105 g) at 4°C. After centrifugation, pellets containing purified cell-free virus were resuspended in 10 mM Tris pH 9.0, aliquoted, and titered using a standard in vitro plaque assay using CV-1 cells[43]. Recombinant Modified Vaccinia Ankara virus expressing firefly luciferase [27] (rMVA-luc; were provided by Michael Seaman (BIDMC; Boston, MA USA).
Sterile, certified low endotoxin preparations of monoclonal antibodies H57–597 (hamster IgG anti-mouse TCRβ)[16], UC7-13D5 (hamster IgG3 anti-mouse TCRγδ)[17], 30H12 (rat IgG2b anti-mouse Thy1.2)[23], GK1.5 (rat IgG2b anti-mouse CD4)[15], LTF-2 (rat IgG2b anti-Keyhole Limpet Hemagglutinin (KLH); used as an isotype control), as well as a polyclonal preparation of purified low endotoxin hamster IgG, were purchased from Bio-X-cell cell culture services (West Lebanon, NH), diluted in sterile PBS to the desired concentration, and administered intraperitoneally (ip). Depleting or isotype control antibodies were administered intraperitoneally to groups of naïve and VV-primed groups of IgHko mice every 2 weeks starting at least 1 week prior to secondary challenge. For adoptive transfer experiments, appropriate depleting or isotype control antibodies were administered intraperitoneally concurrently with transfer of the indicated cell populations.
Fluorescently conjugated antibodies specific for mouse CD3ε (clones 145-2C11, 17A2, and/or 500A2), CD11b (clone M1/70), CD16/32 (clone 24G2; for blocking Fc receptor-mediated binding of fluorescent labeled antibodies), CD49b (clone DX5), CD27 (clone LG.7F9), CD43 (clone eBioR2/60), CD45.1 (clone A20), CD45.2 (clone 104), CD62L (clone MEL-14), CD69 (clone H1.2F3), CD90.1 (clone HIS51), CD90.2 (clone 53-2.1), CD94 (clone 18d3), CD107a (clone 1D4B), CD107b (clone ABL-93), CD127 (clone A7R34), CD244.2 (B6 2b4 alloantigen; clone ebio244F4), KLRG1/MAFA (clone 2F1), Ly6c (clone AL-21) Ly49C/I/F/H (clone 14B11), Ly49H (clone 3D10), NKG2D (clone CX5), NK1.1 (clone PK136), TCRß (clone H57–597), and TCRγδ (clone GL-3) were obtained from either BD Biosciences (San Jose, CA USA) or eBioscience, Inc. (San Diego, CA USA).
Livers were removed from naïve and vaccinia virus-primed B6, B6.SJL (CD45.2 congenic), or B6.PL (CD90.1 congenic) mice and diced in calcium-magnesium- free HBSS + 2% FBS prior to enzymatic dissociation. Diced livers were resuspended in a digestion buffer consisting of 500 µg/ml Collagenase D (>0.15 U/mg; Roche Applied Science) and 10 µg/ml DNAse I (2500 U/mg; Roche Applied Science) in HBSS + 2% FBS and agitated for 1 hour at 37°C. The resulting cell suspension was passed through a 70 micron filter and added to the cell suspensions released by dicing the livers harvested prior to digestion. The remaining pieces were washed two additional times, and the resulting cell suspensions were passed through a 70 micron filter and added to the previously pooled cell suspensions. The pooled cell suspensions were then spun at 500 g for 10 minutes at 20°C. The supernatant was discarded and pelleted cells were resuspended and washed 2 more times in HBSS + 2% FBS. The pelleted cells were then resuspended in 40% Percoll in 1x HBSS, then underlaid with an equal volume of 67% Percoll in 1x HBSS prior to centrifugation at 900 g for 30 minutes at 20°C. Cells at the interface between the two Percoll layers were saved and washed twice in MACS buffer (PBS +0.5% BSA +2.5 mM EDTA). Cells for phenotypic analysis were counted using Guava ExpressPlus software on a Guava easyCyte instrument (Millipore; Billerica, MA) prior to staining with fluorescently-conjugated monoclonal antibodies for analysis on an LSR II instrument in our laboratory. For preparation of adoptive cell transfers, two consecutive rounds of purification were performed: first the mouse NK cell isolation kit (cat # 130-090-864) was used in accordance with the manufacturer's protocols (miltenyi biotec; Bergisch Gladbach, Germany) to deplete non-NK cells on an autoMACS instrument. The enriched NK population from B6 or B6.SJL mice was then washed and separated via the AutoMACS (program ‘possel’; non-labeled fraction is Thy1− NK, labeled fraction is Thy1+ NK) using MACS beads specific for CD90.2 (cat # 130-049-101, miltenyi biotec). The enriched NK population from B6.PL (CD90.1+CD90.2−) livers was incubated with antibodies specific for CD3ε, NK1.1, and Thy1.1 prior to flow cytometric sorting into Thy1− NK (Thy1−CD3−NK1.1+) and Thy1+ (Thy1+CD3−NK1.1+) populations on a BD Vantage instrument in our laboratory.
Purified populations of NK cell-depleted blood mononuclear cells (control; 2–5×106 cells/mouse), Thy1− (1×105 cells/mouse), or Thy1+ NK cells (1×105 cells/mouse) in 1x PBS from naïve and vaccinia virus-primed mice, or naïve RAG1ko and MVA-primed RAG1ko mice, were transferred into age-matched female RAG1ko recipients via tail vein injection using a 25 G needle. Within 5 days post-transfer, recipients were challenged with 1×105 pfu of rVV-luc ip and monitored over time using the IVIS Illumina-II imaging system (Xenogen, inc.; Alameda, CA USA).
At numerous time points following challenge, rVV-luc burdens were monitored in vivo using the IVIS imaging system. Briefly, rVV-luc –infected and/or uninfected control animals were injected ip with 100 µl of a 30 mg/ml solution of firefly luciferin-D (Caliper Life Sciences; Hopkinton, MA, USA) in PBS, and 100 µl of a 20 mg/ml ketamine and 1.72 mg/ml xylazine mixture and imaged 14–16 minutes later in the IVIS series 100 imager (IgHko experiments) or IVIS- lumina II instrument (adoptive transfer experiments). Typical exposure times were 60 seconds (IVIS 100 series) or 30 seconds (IVIS-Lumina II); however, when higher viral burdens were present exposure times were shortened to ensure that images were not oversaturated, and all measurements were normalized for one minute exposure. Overlay images and luminescence measurements were made using Living Image software (version 2.50.1; Xenogen).
Tissue culture plates were treated with either carbonate binding buffer alone (unstimulated and PMA/ionomycin wells), 1 mg/ml control Ig (clone CI.8; Bio-X-cell) in binding buffer, 1 mg/ml anti-NK1.1 (clone PK-136; Bio-X-cell), 2×106 pfu/ml vaccinia virus, or 2×106 vp adenovirus in binding buffer overnight at 4°C. Wells were washed three times with HBSS + 2% FBS. Wells coated with vaccinia virus or adenovirus were treated with 2% formaldehyde in PBS for 20 minutes at room temperature, washed 4 additional times with complete DMEM + 10% FBS, and all wells were blocked for 2 hr at 4°C with cDMEM + 10% FBS. 5×105 liver mononuclear cell preparations from groups of 8–12 mice were incubated in DMEM alone (unstimulated, antibody coated, and vaccinia coated wells) or DMEM with PMA and ionomycin (1 µg/ml and 5 µg/ml respectively) for 6 hours at 37°C. Cell suspensions were then recovered from the plate and processed for flow cytometric analysis.
Groups of B6 mice were infected with 1×107 pfu rVV ip 12 hours prior to sacrifice, infected mice and a matching group of naïve mice were administered 250 µg of the thymidine analog 5-ethynyl-2′-deoxyuridine (EdU) in PBS ip. At sacrifice, mononuclear cells were isolated from livers and spleens, stained with monoclonal antibodies to NK1.1, CD3ε, CD90.2, and fixed. Fixed cells were then washed in a permeabilization buffer containing saponin, treated with Click-IT EdU staining buffer containing Alexa 647 azide to stain incorporated EdU, washed, and analyzed by flow cytometry.
Ovaries were harvested in 10 mM Tris pH 9.0 and snap frozen on a dry ice-ethanol bath. Samples were then thawed, homogenized, and snap frozen again. The samples were then thawed, vortexed, and serially diluted stepwise 1∶10 in duplicate in cDMEM + 10% FCS. 500 µl of each dilution was placed on a confluent layer of CV-1 cells in 6 well plates and incubated at 37°C for one hour, prior to aspiration and addition of 3 ml cDMEM + 10%FCS to each well. Plates were incubated for an additional 48 hours at 37°C prior to aspiration and staining and fixing with 500 µl of 0.1% crystal violet in 20% ethanol for 5 min. After removing the staining solution, the plates were air-dried and then counted. Viral loads in the ovaries were calculated based on the number of plaques, the dilution factor, and the volume of homogenate.
|
10.1371/journal.pgen.1000895 | Candidate Causal Regulatory Effects by Integration of Expression QTLs with Complex Trait Genetic Associations | The recent success of genome-wide association studies (GWAS) is now followed by the challenge to determine how the reported susceptibility variants mediate complex traits and diseases. Expression quantitative trait loci (eQTLs) have been implicated in disease associations through overlaps between eQTLs and GWAS signals. However, the abundance of eQTLs and the strong correlation structure (LD) in the genome make it likely that some of these overlaps are coincidental and not driven by the same functional variants. In the present study, we propose an empirical methodology, which we call Regulatory Trait Concordance (RTC) that accounts for local LD structure and integrates eQTLs and GWAS results in order to reveal the subset of association signals that are due to cis eQTLs. We simulate genomic regions of various LD patterns with both a single or two causal variants and show that our score outperforms SNP correlation metrics, be they statistical (r2) or historical (D'). Following the observation of a significant abundance of regulatory signals among currently published GWAS loci, we apply our method with the goal to prioritize relevant genes for each of the respective complex traits. We detect several potential disease-causing regulatory effects, with a strong enrichment for immunity-related conditions, consistent with the nature of the cell line tested (LCLs). Furthermore, we present an extension of the method in trans, where interrogating the whole genome for downstream effects of the disease variant can be informative regarding its unknown primary biological effect. We conclude that integrating cellular phenotype associations with organismal complex traits will facilitate the biological interpretation of the genetic effects on these traits.
| Genome-wide association studies have led to the identification of susceptibility loci for a variety of human complex traits. What is still largely missing, however, is the understanding of the biological context in which these candidate variants act and of how they determine each trait. Given the localization of many GWAS loci outside coding regions and the important role of regulatory variation in shaping phenotypic variance, gene expression has been proposed as a plausible informative intermediate phenotype. Here we show that for a subset of the currently published GWAS this is indeed the case, by observing a significant excess of regulatory variants among disease loci. We propose an empirical methodology (regulatory trait concordance—RTC) able to integrate expression and disease data in order to detect causal regulatory effects. We show that the RTC outperforms simple correlation metrics under various simulated linkage disequilibrium (LD) scenarios. Our method is able to recover previously suspected causal regulatory effects from the literature and, as expected given the nature of the tested tissue, an overrepresentation of immunity-related candidates is observed. As the number of available tissues will increase, this prioritization approach will become even more useful in understanding the implication of regulatory variants in disease etiology.
| The biological interpretation of genome-wide association study (GWAS) signals [1]–[5] is very challenging since most candidate loci fall either in gene deserts or in regions with many equally plausible causative genes. Following the concurrent progress in understanding the genetic basis of regulatory variation [6]–[9], differential gene expression has been proposed as a promising intermediate layer of information [10] to aid this interpretation [11]. Most commonly, interrogating the GWAS SNPs themselves for significant associations with gene expression [12]–[13] has been employed to explain some of the GWAS results. However, the ubiquity of regulatory variation throughout the human genome [6],[14] makes coincidental overlaps of eQTLs and complex trait loci very likely. This likelihood is a direct consequence of the correlation structure in the genome (linkage disequilibrium - LD), which makes functionally unrelated variants statistically correlated.
As sample sizes increase, allowing the discovery of larger numbers of eQTLs of smaller effect size and as the expression experiments will be performed in a larger variety of tissues, we can envisage that almost every gene will have an associated eQTL under a certain condition. Consequently, the probability that any of these will map to a genomic region where a GWAS SNP also resides is very high. Therefore, it is important to emphasize that while it is very tempting to infer potential causal mechanisms based on such overlaps, this would be a naïve inference in the absence of additional supporting evidence for causality. In the long run, this will not only be an issue for gene expression, but also for any other cellular phenotype. Association studies for intermediate phenotypes with possible relevance to complex traits are underway and their results will overlap some of the GWAS signals. The biological meaning of these overlaps will again need to be evaluated in the context of the genome's correlation structure.
It is not evident though how to model each genomic region with overlapping association signals in the absence of information about the history of the region. Accounting for the historical parameters of a region under the coalescent, while desirable, is computationally and practically not feasible since the human population history is too complex to properly model and small deviations or slightly incorrect assumptions could create false signals or reduce power.
In order to distinguish such accidental colocalizations [15]–[16] from true sharing of causal variants, we propose here an empirical methodology instead. This directly combines eQTL and GWAS data while accounting for the LD of the region harbouring the GWAS SNP. We demonstrate the value of the approach by predicting the regulatory impact of several GWAS variants in cis and trans and we also show that the correlation strength (r2, D') between the GWAS SNP and the eQTL is not a sufficient predictor of regulatory mediated disease effects.
To identify likely causal effects (not variants since we do not have full sequencing data) associated with complex traits and diseases we took advantage of published association data catalogued in the NHGRI [17] database and gene expression data generated in LCLs derived from HapMap 3 individuals (see Methods). In this study, we limited the expression analysis to the 109 CEU individuals, as they are the closest in ancestry to the majority of individuals in published GWAS studies. We used the NHGRI database (accessed 02.03.09) to extract 976 GWAS SNPs with minor allele frequency (MAF) >5% that were also genotyped in the HapMap 3 CEU, thus allowing to test the exact GWAS SNPs for associations with differential gene expression in LCLs. In total we examined 17673 genes. In order to discover eQTLs, we used Spearman Rank Correlation (SRC). This method [14] captures the vast majority of associations discovered with standard linear regression (LR) models, with the additional advantage that it's not affected by outliers and hence has more power and allows direct comparison of nominal P-values. We looked for both proximal (cis) and distal (trans) effects as follows: variants within 1Mb on either side of the transcription start site (TSS) of a gene are considered to be acting in cis, while those at least 5 Mb downstream or upstream of the TSS or on a different chromosome are considered to be acting in trans.
In order to assess the overall impact of the currently known GWAS SNPs on expression, we contrasted their cis and trans effects to those of a random set of SNPs, representing the null. In a QQ plot (Figure 1), we compare the distributions of the best cis and trans association p-values per SNP for the 976 GWAS SNPs (observed) to 1000 sets of most significant p-values of 976 random SNPs each (expected). The 1000 random sets of 976 SNPs were sampled to have identical MAF distribution to the GWAS SNPs. In cis, we observe a much stronger regulatory signal in the GWAS data compared to random (Figure 1). The significant difference between the two becomes apparent above a −log10(P-value) = 4. In trans, we also detect a more significant regulatory signal for GWAS SNPs compared to random, however not as strong as in cis. This is to be expected given that the much greater statistical space we're exploring in trans limits the power to detect such effects. Nevertheless, despite their confinement to one tissue type - LCLs, these comparisons support the overall explanatory potential of regulatory variation for the biological effects of GWAS variants. As expected given the tissue nature, the phenotypes responsible for this enrichment are immunity related (Figure 2).
To identify the subset of causal effects from the regulatory enrichment observed, we focused only on the genomic regions harbouring either cis or trans eQTLs. We split the genome into recombination hotspot intervals based on genome-wide estimates of hotspot coordinates from McVean et.al. [18] Limiting the search space for causal effects to these intervals is a reasonable conventional approach, as few or no recombination events are expected between the reported associated SNPs and the functional variants they are tagging.
Given the abundance of cis eQTLs in the genome, mere interval overlap in not sufficient to claim that a colocalized cis eQTL and a GWAS SNP are tagging the same functional variant. However, if the GWAS SNP and the eQTL do tag the same causal SNP, we expect that removing the genetic effect of the GWAS SNP will have a marked consequence on the eQTL association. Starting from this hypothesis, we developed an empirical method to uncover regulatory mediated associations with complex traits. For all genes with a significant cis eQTL (0.05 permutation threshold as defined in Stranger et.al. 2007, see Methods) in a given interval, we create corrected phenotypes from the residuals of the standard LR of the GWAS SNP against normalized expression values of the gene for which we have an eQTL. The residuals capture the remaining unexplained expression variance after the removal of the GWAS SNP effect. We redo the SRC analysis with the pseudo phenotype and retain the adjusted association P-value.
Depending on the internal LD structure of the hotspot interval, the correlation between the GWAS SNP and the eQTL will vary, hence so will the P-values after and before correction. One way to assess the relevance of the GWAS SNP to the eQTL is to compare its correction impact to that of all other SNPs in the interval. For this purpose, we define a Regulatory Trait Concordance (RTC) Score for each gene-GWAS SNP combination as follows, taking into account the ranking of the correction with respect to all SNPs in the interval (RankGWAS SNP) and the total number of tested SNPs (NSNPs).
The rank denotes the number of SNPs which when used to correct the expression data, have a higher impact on the eQTL (smaller adjusted P-value) than the GWAS SNP (i.e. RankGWAS SNP = 0 if the GWAS SNP is the same as the eQTL SNP, RankGWAS SNP = 1 if of all the SNPs in the interval, the GWAS SNP has the largest impact on the eQTL). Given this, the RTC Score will always be in the range (0,1], with values close to 1 indicating that the GWAS effect is the same as the eQTL effect.
We investigated the properties and robustness of the RTC score under the null hypothesis (H0: eQTL and GWAS are tagging two different causal SNPs) and the alternative hypothesis (H1: same causal SNP). For this purpose, we have simulated causal SNPs (cSNP), eQTLs and dSNPs (see Methods) varying the LD levels between them as well as the LD pattern of the hotspot interval where they reside. We have then masked the cSNPs and calculated the RTC score under these different LD scenarios for both hypotheses.
The RTC score is uniformly distributed under the null, when the simulated causal eQTL SNP (c-eQTL) and the causal disease SNP (c-dSNP) are different (Figure 3, left panel).
Under the H1 on the other hand, the RTC score is right skewed, with a clear enrichment for values close to 1 recovering the single causal SNP effect (Figure 3, middle panel).
The simulations show that the complexity and variability of the LD structure in the genome impede the simple use of correlation metrics to infer shared causal effects.
The statistical correlation (r2) between the eQTL and the dSNP is not on its own sufficient to predict whether they tag the same cSNP (Figure 4). The RTC outperforms r2 as it is able to recover causal effects even for low correlated pairs. The historical correlation metric between eQTLs and dSNPs (D') is also not fully predictive of high RTC scores (Figure 5). We observe from the H0 simulation results that D' is not correlated with RTC, meaning that when the eQTL and dSNP tag different functional variants, the RTC score is not high just because D' is high. In addition, while high RTC scoring cases cluster much tighter around high D' values under the H1 compared to r2 previously, a high D' is not sufficient to predict causal effects. That is because it would be impossible to distinguish causal from coincidental effects given a perfect historical correlation scenario.
Finally, we investigated the effect of the overall LD pattern in a region of interest on the RTC. For this purpose, we calculated the median r2 of each hotspot interval and checked its relationship to the RTC score under the null and alternative hypothesis. It is expected that RTC will perform better in intervals with overall low LD, where the correlation between the eQTL and other non-disease SNPs will decay much faster, making the correction for the dSNP stand out. However, we confirm that the LD of the region does not determine high scores by itself. Intervals of low LD where different c-eQTLs and c-dSNPs reside have a uniform distribution of RTC scores (Figure 6, left panel). As expected, we do observe from the H1 simulations that we have most power in intervals with low median r2 (Figure 6, right panel).
As a positive control, we tested the method first on intervals harbouring already identified regulatory associations. We used published cis eQTLs (10−3 permutation threshold) discovered in the same tissue as the HapMap CEU eQTLs (LCLs) but derived from an independent set of samples: 75 individuals of Western European origin from the GenCord project [19]. In this experiment, we considered the GenCord eQTLs as the equivalent of GWAS SNPs and we limited our analysis to intervals with cis eQTLs in both datasets. Furthermore, we conditioned the associated genes for the same interval to be identical in the two expression datasets, expecting thus a common functional variant. As a result of this filtering, we tested SNPs in 157 hotspot intervals, associated with differential expression levels of 154 genes. As expected from the H1 simulations, the RTC Score distribution after correcting for the GenCord eQTLs is right-skewed (Figure 3, right panel), suggesting that the scoring method is sensitive to associations tagging the same functional variant. We detect 33 SNP-probe pairs with an RTC Score of 1 out of the total 185 tested pairs. Given the marked difference in genotyping density between HapMap and GenCord (∼1.2 millionSNPs versus ∼400,000 SNPs respectively) and our hypothesis that the 157 overlapping intervals share the same functional variant, we expect approximately 3 times more perfect scoring cases (99 pairs with RTC Score = 1) than what we observe, had individuals from both datasets been equally densely genotyped. We use the degree of sharing between the eQTLs in the two datasets to derive a reasonable, yet conservative threshold: currently, 105 SNP-probe pairs pass the 0.9 RTC threshold, making it thus a suitable stringent cut-off for calling significant discoveries.
We then applied the scoring method on the NHGRI GWAS SNPs. The 976 common GWAS SNPs map to 784 hotspot intervals. Of these, we focused the cis analysis on GWAS intervals (N = 130) where at least one significant cis eQTL at a 0.05 permutation P-value threshold also resides (Dataset S1). For the trans analysis, we ordered all 784 GWAS intervals by their most significant trans eQTL and kept the topmost 50 intervals for further examination (Dataset S2). Table 1 summarizes our most confident cis results ordered by RTC Score. We detect SNP-gene combinations passing the 0.9 threshold for 28 intervals out of the 130, twice as many than expected by chance (13 expected top 10% scoring intervals under the uniform distribution). Our method confirms prior results in the literature suggestive of disease effects mediated through expression (ORMDL3 for asthma risk [13], C8orf13 locus for lupus risk [20], SLC22A5 for Crohn's disease [12],[21]). In addition, we detect several other yet unknown candidate genes for a variety of conditions.
An interesting example of a novel cis regulatory mediated effect is the one for Crohn's disease with gene SLC38A3, member 3 of the solute carrier family 38. Independent studies detected significant Crohn's associations of two SNPs in the same hotspot interval on chromosome 3 (rs3197999 [12], a non-synonymous SNP in gene MST1 and rs9858542 [1],[22], a synonymous SNP in nearby gene BSN). Suggestive literature evidence in addition to the disease associated non-synonymous SNP made MST1 the most attractive candidate gene out of the many present in that region [23]. However, our data supports an additional regulatory component underlying the susceptibility locus. For both GWAS SNPs, SLC38A3 is the highest scoring candidate in the region (RTC Score: 0.92). Interestingly, this is functionally similar to another Crohn's susceptibility gene SLC22A5 confirmed with our method (RTC Score: 1.0) and also encoding a sodium dependent multi-pass membrane protein (solute carrier family protein). The observed direction of effect is the same for both genes (eQTLs associate with low expression levels) as in previous expression datasets [12] and suggests a possible involvement of this gene family in the disease. This is in agreement with recent studies reporting that disease causative genes are functionally more closely related [24].
The tissue under investigation is LCLs so we expect GWAS signals of immunity related traits (comprising here autoimmune disorders and diseases of the immune system e.g. AIDS progression) to more likely show an overlap with eQTLs. In order to evaluate the relevance of our results, we analyzed the distributions of the best RTC Scores per GWAS SNP stratified by the immunity relatedness of the complex trait they associate with (Figure 7). We observe a significant overrepresentation of high-scoring genes (> = 0.9) for immunity related traits compared to non-immunity related ones (Fisher's Exact Test, P-value = 0.0125) [25]. This suggests that the scoring scheme predicts regulatory effects of the relevant phenotypes. In addition, we observed that for GWAS signals with RTC score >0.9, only 10% of the nearest gene to the GWAS SNP was also the eQTL gene. These however, correspond as expected to instances when the eQTL gene is also the nearest gene to the eQTL itself. If that is not the case, the inference of relevance of a gene simply based on its proximity to the GWAS SNP is not informative.
Even if the causal SNP is not cis-regulatory, using gene expression to determine its downstream targets, coupled with information about the biological pathways these targets act in could help interpret the primary GWAS effect. We investigated this hypothesis in the topmost 50 GWAS intervals ordered by their trans eQTL significance. For each interval, we apply the RTC Scoring scheme on the subset of genes in the whole genome with a notable effect in trans (SRC nominal P-value <10−5). These signals amount to a total of 552 genes. We obtain SNP-gene combinations passing the 0.9 Score threshold for 24 of the 50 tested intervals (corresponding to a total of 85 genes). Six of these intervals contain GWAS SNPs associated with immunity related traits (Table 2). While not statistically significant - unsurprisingly given that we're only testing a small subset of the total GWAS intervals - these examples support the usefulness of the trans approach. As hypothesized, for the same complex trait associated SNP we can discover several potential candidate genes in trans, throughout the genome. Some of these are biologically plausible results and merit further investigation. However, many trans candidates are hard to interpret at this stage given their incomplete annotation and further functional studies will need to be performed for validation.
The power to detect significant associations between genotyped SNP proxies and a phenotype depends on the correlation between those proxies and the functional variant [26]. Just like for the simulated data, we tested whether the correlation between a GWAS SNP and its colocalizing eQTL is sufficient for predicting a shared causal effect. For both the cis and the trans analysis, we observe that the r2 between the eQTL and the disease SNP is not a direct predictor of the RTC Score, and in several cases we predict that even pairs with low r2 are likely tagging the same functional effect (Figure 8, top panel). The reason for this is that many of the high scoring pairs with poor statistical correlation (low r2) are actually historically correlated (D' = 1). Nevertheless, D' is not very informative either (Figure 8, bottom panel), the main problem here being that in regions with generally high D' among many SNPs, one cannot determine which of the pairs actually represents a common functional variant.
Another metric of potential predictive value is the fraction of eQTL variance explained by the dSNP. Figure 9 indicates the relationship between the RTC score and the fraction of explained variance at the eQTL left unexplained after the dSNP correction (ratio of linear regression adjusted R∧2 after and before correction). As expected given the definition of the RTC, the highest density of good scoring results is registered for dSNPs that explain most of the eQTL variance. However, RTC outperforms the variance metric, scoring high even when that's not the case and thus making the setting of a threshold on the explained variance not sufficiently informative either.
To aid the functional interpretation of complex trait association signals, we describe here an empirical methodology that directly integrates eQTL and GWAS data while correcting for the local correlation structure in the human genome. As regulatory variants are pervasive throughout the genome, coincidental overlaps of eQTLs and GWAS SNPs are very likely. Hence, current methods that limit themselves to asking whether disease intervals also harbour eQTLs are unreliable for distinguishing trait relevant regulatory effects from other eQTLs. Our methodology addresses and helps resolve this issue.
This approach is not limited to gene expression, but could be generalized to any other phenotype. As new methods are developed and larger cohorts become available, various intermediate cellular phenotypes are interrogated via association studies with the hope to find explanatory links between genotypic variation and complex trait predisposition. However, the biological interpretation of these discoveries will also be hardened by the presence of tight LD. It is therefore necessary to evaluate them in a conservative manner, correcting for the local correlation structure in each genomic interval with overlapping association signals.
In this paper, we discover causal regulatory effects and their affected candidate genes in cis and to some extent in trans by assessing the impact on the expression phenotype of the removal of the GWAS SNP effect. We compute a score (RTC) for each individual genomic interval that assesses the likelihood that the eQTL and the GWAS SNP are tagging the same functional variant. By ranking the effect of the removal of the GWAS SNP in comparison to the outcome for any other SNP in the region and by accounting for the number of SNPs tested, we produce a score comparable across intervals. We evaluate the performance of the score in various simulated LD scenarios and we present its robustness by its expected uniform distribution when the eQTL and GWAS SNP are tagging different functional variants. In comparison, we investigate how well do current SNP correlation metrics (r2, D') perform on their own. We show that the LD between the GWAS SNP and its colocalizing eQTL is not a good predictor of a shared functional effect. This is very important especially since most of the current replication and follow-up studies only focus on variants highly correlated (r2>0.8) with the initial discoveries. It is important to stress at this point that neither the eQTL nor the GWAS SNP is likely the causal variant. Therefore, what really matters is not the statistical correlation between two proxies but the correlation between each of the proxies and the causal variant, whose frequency is unknown. In any case, no obvious combination of LD measures can substitute the RTC scoring scheme and we thus conclude that many interesting candidate genes would be missed if one were to rely solely on correlation-based approaches.
In this paper, we also explore the explanatory potential of regulatory variation given the currently published GWAS data. We observe a significant overrepresentation of eQTLs among GWAS SNPs, especially affecting genes in cis. Long-range trans effects are also present but less prevalent, possibly due to lower power to detect such associations. As expected given the tissue the expression data was measured in (LCLs), we observe a significant abundance of cis regulatory causal effects for immunity related traits. Our result reinforces the necessity to expand the tissue diversity [27]–[28] of genome-wide expression studies in order to facilitate such discoveries for a wider range of human conditions.
By applying the RTC method on the NHGRI GWAS SNPs, we are able to confirm previously suspected regulatory mediated disease effects and discover novel candidate genes affected by GWAS SNPs. We provide a list of follow-up candidate genes affected in cis and in addition, we show the utility of genome-wide expression data irrespective of the nature of the primary SNP effect by predicting clusters of genes affected in trans. The individual examination of the candidates prioritized with our approach will undoubtedly assist the biological interpretation of the ever-increasing list of GWAS signals. As associations with more intermediate cellular phenotypes will be reported, the integration of all these signals will be crucial for understanding the biology of complex traits.
RNA levels were measured in lymphoblastoid cell lines (LCLs) derived from the HapMap 3 individuals using a whole-genome expression array (Illumina Sentrix WG-6, Version 2) as previously described [14]. Each sample had two technical replicates. We analyzed here only expression data from the CEU, a HapMap 3 population of 109 unrelated individuals of Northern European ancestry. The mapping of Illumina probes to unique Ensembl gene IDs resulted in 21,811 probes corresponding to 17,673 autosomal genes available for association analysis. 1,186,075 SNPs (MAF >5%) genotyped in the same individuals were used in the eQTL analysis.
The log2 transformed raw intensity values were normalized as follows: quantile normalization of sample replicates (two intensity values per Illumina probe) followed by median normalization across all individuals.
All SNPs from the catalogue of genome-wide association studies maintained by the National Human Genome Research Institute (NHGRI www.genome.gov/26525384) and published by 02.03.2009 were downloaded. Of these, only the 976 unique common variants (MAF >5%) genotyped in the HapMap 3 CEU samples were kept for analysis.
Associations between SNP genotypes and normalized expression values were conducted using Spearman Rank Correlation (SRC). For the cis analysis, we considered only SNPs within a 1MB window from the TSS of genes, while in trans we test all SNPs further than 5MB away from the gene's TSS and all SNP-gene pairs on different chromosomes. We assess the statistical significance of the cis associations using permutations as previously described [7],[14]. We call a cis eQTL significant if the nominal association P-value is greater than the 0.01 tail of the minimal P-value distribution resulting from the SNP's associations with 10,000 permuted sets of expression values for each gene.
We mapped all common autosomal CEU HapMap 3 SNPs (1,186,075 SNPs) to recombination hotspot intervals as defined by McVean et.al. [18] For the cis analysis we selected the 130 hotspot intervals where at least one significant cis eQTL and a GWAS SNP colocalize while for the trans, we analyzed a subset of 50 of the total 784 unique intervals (where the 976 GWAS SNPs map to). These are the topmost intervals ordered by their most significant trans eQTL (nominal SRC P-value).
For both the cis and trans GWAS analysis, the best P-value associations per SNP were stored. The set of the most significant P-values of the 976 GWAS SNPs was compared to 1000 sets of most significant P-values of 976 random SNPs. The 1000 random sets of 976 SNPs each were conditioned to have the same MAF distribution as the 976 GWAS set.
The QQ plot showing the abundance of regulatory signal in GWAS data is the median QQ plot of 1000 (GWAS, random SNPs) comparisons. It shows the distribution of the −log10 quantile values of the GWAS best associations (observed) versus the median of the corresponding 1000 −log10 quantile values from each of the 1000 random SNP sets (expected). In order to assess the significance of the observed versus expected median QQ plot, we superimpose the upper limit of the 95% confidence interval. This is calculated from the sorted 0.95 quantiles of 10000 pairs of 976 random SNPs each.
We assess the likelihood of a shared functional effect between a GWAS SNP and an eQTL by quantifying the change in the statistical significance of the eQTL after correcting for the genetic effect of the GWAS SNP. We redo the SRC association of the eQTL genotype with the residuals from the standard LR of the “corrected-for” SNP against normalized expression values. We account for the LD structure in each hotspot interval separately by ranking (RankGWAS SNP) the impact on the eQTL (quantified by the adjusted association P-value after correction) of the GWAS SNP correction to that of correcting for all other SNPs in the same interval. By taking into account the total number of SNPs in the interval (NSNPs), we can compare this ranking across different genes and intervals. For this purpose we define the regulatory trait concordance (RTC) Score ranked below ranging from 0 to 1, with values closer to 1 indicating causal regulatory effects.
We investigate the properties of the RTC score with respect to different correlation metrics under the null hypothesis (H0: eQTL and dSNP tag different functional variants) and the alternative hypothesis (H1: eQTL and dSNP tag the same functional variant).
We use the HapMap3 CEU cis eQTLs (315 genes at 10−3 permutation threshold) to create a list of causal SNPs (cSNP). For the H0, we call these cSNPs causal eQTL SNPs (c-eQTL). For each c-eQTL, we sample a different causal disease SNP (c-dSNP) from the same interval, with the requirement that its MAF comes from a distribution identical to that of the 976 NHGRI GWAS SNPs. Subsequently, we sample up to five eQTL-dSNP pairs per interval where the eQTLs and dSNPs are the topmost correlated (r2) SNPs with the c-eQTL and the c-dSNP respectively. After sampling, we exclude cases where the eQTL and dSNP are identical, as these contradict the H0..c-eQTL-c-dSNP-eQTL-dSNP quartets mapping to 287 unique hotspot intervals were sampled and tested under H0.
Under the H1, we sample up to five eQTL-dSNP pairs for each hotspot interval harbouring a cSNP as follows: the eQTLs are chosen as the top most significant SNPs per eQTL gene - excluding the cSNP; the dSNPs are randomly sampled from the same hotspot interval such that the r2 between each of them and the cSNP is in the range [0.5,0.9]. At any stage of the 5-step iteration per cSNP, the dSNP must be different from the cSNP and the eQTLs sampled up to that point. cSNP-eQTL-dSNP trios mapping to 290 unique hotspot intervals throughout the genome were sampled and tested under the H1.
We use the LD values (r2) of all pairwise SNP combinations per interval to calculate the median r2, an estimate of the LD extent per region.
To perform a control experiment where the trait is gene expression, we used cis eQTLs (10−3 P-value permutation threshold) detected in LCLs derived from 75 unrelated individuals of Western European origin from the GenCord project [19]. Hotspot intervals (N = 157) where both a HapMap and a GenCord eQTL associating with the same Ensembl gene reside were analysed with the RTC Scoring scheme.
|
10.1371/journal.pbio.1001287 | NUP-1 Is a Large Coiled-Coil Nucleoskeletal Protein in Trypanosomes with Lamin-Like Functions | A unifying feature of eukaryotic nuclear organization is genome segregation into transcriptionally active euchromatin and transcriptionally repressed heterochromatin. In metazoa, lamin proteins preserve nuclear integrity and higher order heterochromatin organization at the nuclear periphery, but no non-metazoan lamin orthologues have been identified, despite the likely presence of nucleoskeletal elements in many lineages. This suggests a metazoan-specific origin for lamins, and therefore that distinct protein elements must compose the nucleoskeleton in other lineages. The trypanosomatids are highly divergent organisms and possess well-documented but remarkably distinct mechanisms for control of gene expression, including polycistronic transcription and trans-splicing. NUP-1 is a large protein localizing to the nuclear periphery of Trypanosoma brucei and a candidate nucleoskeletal component. We sought to determine if NUP-1 mediates heterochromatin organization and gene regulation at the nuclear periphery by examining the influence of NUP-1 knockdown on morphology, chromatin positioning, and transcription. We demonstrate that NUP-1 is essential and part of a stable network at the inner face of the trypanosome nuclear envelope, since knockdown cells have abnormally shaped nuclei with compromised structural integrity. NUP-1 knockdown also disrupts organization of nuclear pore complexes and chromosomes. Most significantly, we find that NUP-1 is required to maintain the silenced state of developmentally regulated genes at the nuclear periphery; NUP-1 knockdown results in highly specific mis-regulation of telomere-proximal silenced variant surface glycoprotein (VSG) expression sites and procyclin loci, indicating a disruption to normal chromatin organization essential to life-cycle progression. Further, NUP-1 depletion leads to increased VSG switching and therefore appears to have a role in control of antigenic variation. Thus, analogous to vertebrate lamins, NUP-1 is a major component of the nucleoskeleton with key roles in organization of the nuclear periphery, heterochromatin, and epigenetic control of developmentally regulated loci.
| Eukaryotes—fungi, plants, animals, and many unicellular organisms—are defined by the presence of a cell nucleus that contains the chromosomes and is enveloped by a lipid membrane lined on the inner face with a protein network called the lamina. Among other functions, the lamina serves as an anchorage site for the ends of chromosomes. In multicellular animals (metazoa), the lamina comprises a few related proteins called lamins, which are very important for many functions related to the nucleus; abnormal lamins result in multiple nuclear defects and diseases, including inappropriate gene expression and premature aging. Until now, however, lamins had been found only in metazoa; no protein of equivalent function had been identified in plants, fungi, or unicellular organisms. Here, we describe a protein from African trypanosomes—the single-cell parasites that cause sleeping sickness—that fulfils many lamin-like roles, including maintaining nuclear structure and organizing the chromosomes of this organism. We show that this protein, which we call NUP-1 for nuclear periphery protein-1, is vital for the antigenic variation mechanisms that allow the parasite to escape the host immune response. We propose that NUP-1 is a lamin analogue that performs similar functions in trypanosomes to those of authentic lamins in metazoa. These findings, we believe, have important implications for understanding the evolution of the nucleus.
| Eukaryotic genomes are primarily organized as linear chromosomes and further segregated into transcriptionally active euchromatin and repressed heterochromatin [1]–[3]. In metazoa such chromatin organization requires the coiled-coil lamins, intermediate filament proteins that form a stable meshwork between the nuclear envelope (NE) and nuclear matrix, physically associating with peripheral heterochromatin [2],[4],[5]. Lamins directly participate in nuclear pore complex (NPC) positioning, maintenance of nuclear structure, spindle assembly, and control of developmental gene expression programs [6]–[9]. Lamins also function in positioning of the nucleus within the cell, and nuclear reassembly following mitotic NE vesiculation in open mitosis [10]–[13]. In humans, aberrant lamin protein structure or expression can lead to irregular nuclei and inappropriate gene expression, manifesting as pathological laminopathies, including progeria and muscular dystrophies [14]. Remarkably, no lamin orthologues had been identified in non-metazoa [15]. Moreover in yeasts, lamins and a major nucleoskeleton are clearly absent, despite the presence of apparent heterochromatin [16]. Together, this implies that the lamin-dependent mechanisms of heterochromatin organization in metazoan cells are a lineage-specific feature, and have evolved relatively recently, following the split between the animals and fungi [17].
Nevertheless, structures morphologically resembling nuclear peripheral heterochromatin and a lamina have been described in several divergent eukaryotic lineages, but their molecular basis has remained elusive [18]–[20]. For plants, for example, equivocal evidence is suggestive of the presence of a lamina-like nucleoskeletal structure (discussed in [21]). Further, heterochromatin is also tethered to the nuclear envelope of plants, and at least one candidate nucleoskeletal protein, NMP-1, has been identified. NMP-1 is a 36 kDa predominantly alpha-helical protein that associates with the nuclear matrix, but remains functionally uncharacterized, and is also likely plant specific [22].
Trypanosomatids are highly divergent organisms, whose origins may even lie close to the Eukaryotic root. Their mode of transcriptional control is highly unusual and, for the most part, independent of conventional promoter control, relying instead on polycistronic transcription and trans-splicing. Trypanosomes have structures reminiscent of a peripheral nucleoskeleton, while they also possess prominent heterochromatin-like material at the nuclear periphery that is implicated in control of gene expression [23]. The African trypanosome T. brucei is an obligate parasite living primarily in the blood, lymphatics, and cerebrospinal fluid when in the mammalian host (bloodstream form; BSF) and in the midgut and salivary glands in the Tsetse fly (procyclic form; PCF). The very different environments encountered by the parasite between these two hosts demand rapid and complex transcriptional changes. Further, the trypanosome cell cycle is highly coordinated, with precisely ordered division of organelles and suborganelles [24] and nuclear DNA elements. In trypanosomes these consist of 11 pairs of conventional megabase chromosomes harbouring the majority of protein coding genes plus dozens of unusual lower molecular weight minichromosomes containing mainly VSG genes. These two classes of chromosome segregate during mitosis with differential kinetics, location, and possibly also mechanism [25].
The BSF has a sophisticated system for immune evasion based on antigenic variation via expression at high copy number of a single variant surface glycoprotein (VSG). Periodic switching of the active VSG gene prevents elimination of the entire parasite population by allowing a subpopulation to escape the host immune response. VSG expression is tightly controlled to ensure monoallelic expression and takes place exclusively from telomere-proximal expression sites (ESs). An ES is present at many of the megabase chromosome telomeric regions. Further, VSG is developmentally regulated; the procyclic stage expresses only a second dominant surface protein, named procyclin. In contrast to higher eukaryotes, with largely promoter-controlled gene systems, trypanosome megabase chromosomes are organized into extensive polycistronic units, and mRNA levels are chiefly regulated by post-transcriptional mechanisms [26],[27]. However, trypanosomatids do possess chromatin subcompartments implicated in the control of gene expression. In T. brucei, electron dense heterochromatin encompassing telomeric regions is largely restricted to the nuclear periphery. Critically, in BSFs an RNA polymerase I–containing extranucleolar expression site body, located within the nuclear interior, provides an environment permissive for VSG transcription [28]–[31]. As telomeres carry multiple repressed VSG genes, expression site translocation between peripheral heterochromatin and the expression site body likely mediates antigenic variation [29], which therefore depends on chromatin organization and involves epigenetic mechanisms.
While several chromatin-remodelling and histone-modifying enzymes that act at telomeric regions have been described, no nucleoskeletal components acting specifically on trypanosome chromatin organization, with functions analogous to metazoan lamins, are known [32]–[34]. However, one candidate is NUP-1, a large coiled-coil protein that coenriches with the nuclear envelope and appears to be a component of fibrous material subtending the inner NE. We sought to determine if NUP-1 functions in heterochromatin organization and gene regulation at the nuclear periphery in T. brucei, which would have implications for the evolutionary origins of such mechanisms [23]. Using multiple approaches, we show that NUP-1 has several functions that are clearly analogous to those of metazoan lamins, including mediation of nuclear structural integrity, epigenetic chromatin organization, and maintenance of developmentally regulated gene expression. We propose that such mechanisms are likely widespread amongst eukaryotes, and hence an ancient and fundamental feature of gene control, rather than a lineage-specific aspect of metazoan cells.
NUP-1 is a large protein (pI 5.07, MW>400 kDa) associated with the nuclear periphery [23],[35] that was identified as a major component of the T. brucei NE proteome [36]. As predicted by both COILS and CCHMM_PROF, NUP-1 is almost exclusively coiled-coil with a central region of 17 near-perfect repeats of 144 amino acids (Figure 1A) [37],[38]. No trans-membrane helices were predicted. Moreover, the N- and C-terminal regions are also predicted to be predominantly alpha-helical and extended (>90% confidence, http://www.sbg.bio.ic.ac.uk/phyre2/); given that the rise per residue is ∼1.5 Å, this predicts that were the structure fully extended, the NUP-1 monomer would extend nearly to 400 nm, which is ∼25% of the diameter of the trypanosome nucleus.
A single NUP-1 syntenic orthologue was present in each trypanosomatid genome examined (Figure 1, Table S1), each exhibiting the same structure as NUP-1 but varying in size and number of repeats; the repeats within each NUP-1 orthologue are nearly identical but diverge significantly between species. Further, NUP-1 orthologues in Trypanosoma species diverge significantly from those in Leishmania (Figure 1, Table S2). So far, BLAST has failed to identify sequences with significant similarity to NUP-1 in Phytomonas, Bodo saltans (a free-living kinetoplastid), or Euglena gracilis.
NUP-1 mRNA is expressed at similar levels in BSF and PCF T. brucei, indicating a role throughout the life cycle (Figure S1). We established the location of NUP-1 by C-terminal genomic tagging of one allele with GFP in PCF cells followed by confocal microscopy. Fluorescence was observed at the nuclear periphery, and taken together with previous immunoEM, subcellular fractionation, and monoclonal antibody studies, indicates that NUP-1 is localized to a net-like structure at the nuclear periphery. The location described here has a more net-like distribution compared to the original description, which suggested a punctate nuclear rim localization, and was interpreted as a potential nucleoporin (Figures 2 and 3, Movies S1, S2, S3, and S4) [23],[35]. A similar pattern was seen in BSF cells with rabbit polyclonal antibodies raised to the NUP-1 repeat (Figure 2A, Movie S4), demonstrating that the GFP tag did not affect NUP-1 localization and confirming that the network is likely a more accurate view of NUP-1 distribution and not the puncta seen earlier [35]. We note that the vertex length (i.e., the distance between strongly stained puncta and apparent fibres or fibrils) is somewhat variable, but is similar to the 400 nm of the putative extended NUP-1 protein. While we are unclear as to how many NUP-1 molecules contribute to these structures, this, together with data below (Figure 3), suggests that the protein is highly extended (Figure 2D). It is also unclear at this time if NUP-1 is the sole component of the network or if other proteins contribute. Regardless, these data suggest that NUP-1 is a bona fide component of the nucleoskeletal network and, moreover, that the location is highly distinct from the punctate staining that we have reported previously for over 20 NPC proteins [23],[36]. We propose referring to NUP-1 as nuclear peripheral protein 1 to help differentiate it from nucleoporins.
To determine if NUP-1 was stably associated with the nuclear periphery, we used fluorescence recovery after photobleaching (FRAP). After bleaching, no significant fluorescence recovery was seen during 150 s (Figure 2B, Movie S5), while fluorescence recovery was observed almost immediately with NLS-tagged GFP (unpublished data). This suggests that NUP-1 is part of a comparatively immobile network at the nuclear periphery in interphase and that rapid exchange of NUP-1 subunits does not occur.
In trypanosomes nuclear mitosis is preceded by division of the kinetoplast, the mitochondrial DNA, while cytokinesis lags behind mitosis by a considerable period. This allows early steps in mitosis to be conveniently detected in mounts of log-phase parasites, and also for post-mitotic nuclei to be analysed within the mother cell prior to cell division. Overall the NUP-1-containing network remained in place at all stages of mitosis (Figure 2A). An extension of the NUP-1 network was also clearly present within the midbody between nuclei in anaphase, which suggests that the NUP-1 network retains intimate contact with the nuclear envelope of the midbody. Note that at late anaphase the midbody becomes depleted of DNA, which has accumulated at the distal poles of the daughter nuclei, so that the midbody is no longer stained significantly with DAPI at this stage.
Despite the maintenance of a clear NUP-1 presence at the nuclear envelope throughout mitosis, significantly less NUP-1 was present in the proximal compared to distal portions of each daughter nucleus (Figure 2C). This rearrangement may contribute to mechanical weakening of the NE to facilitate nuclear fission and strengthening of the distal regions where the spindle is attached. We also found that the overall distance between these NUP-1 puncta increased in mitotic cells compared to interphase, which further suggests remodelling of the network in a cell-cycle-dependent manner (Figure 2D). It is also likely significant that the distance between NUP-1 punctate structures is of the order of 400 nm (Figure 2A and D), suggesting that there is a highly organized assembly of the NUP-1 protein in the trypanosome nucleoskeleton.
We also frequently observed a small spot of NUP-1 between post-mitotic nuclei, apparently unassociated with DNA as detected by DAPI (Figure S1, Movie S4). To determine if this was a genuine extranuclear localization, we compared the location of NUP-1 with NLS-tagged GFP, used to mark the nucleoplasm. This revealed that the NUP-1 spot localized to the residual midbody connection between daughter nuclei at terminal mitotic stages as it also contained GFP (Figure S1D,E). This suggests that a nucleus-derived fragment remains between the daughter nuclei following mitosis, which does not contain significant amounts of DNA; this is presumably a result of the fission mechanism, perhaps analogous to generation of an aerosol when water drops from a faucet. The presence of nucleoplasmic-targeted GFP suggests that this structure, which is only seen in post-mitotic cells and therefore probably rapidly degraded, is unlikely to be stably associated with a cytoplasmic structure (e.g., an MTOC).
To test whether NUP-1 is distributed throughout the nucleoskeletal network, we stained cells expressing NUP-1-GFP with antibodies against both the C-terminal GFP tag and the repeat region. Given an optical resolution of ∼200 nm, we reasoned that if NUP-1 were predominantly globular, then complete overlap between the two signals would be observed, while if NUP-1 were predominantly elongated, and given a total overall length for the fully extended protein of ∼400 nm, then there would be partial separation of the two signals (Figure 3A). While there was clearly substantial overlap between the spatial distribution of the repeat and C-terminal stains, partial separation was indeed observed (Figure 3B), with the most likely interpretation that NUP-1 is predominantly extended. To confirm the specificity of the co-stain with anti-GFP and anti-repeat antibodies, we also stained cells expressing NUP98-GFP, a nucleoporin, with the NUP-1 anti-repeat antibody (Figure 3B), which demonstrated no cross-reactivity. By contrast, when the GFP epitope was part of a smaller and more compacted FG-repeat-containing nucleoporin, TbNUP89 (Figure 3C), there was clear coincidence between the GFP-stain and that obtained from a cross-reacting antibody to vertebrate NUP107 [36]. These data suggest that at least some NUP-1 polypeptides have an extended conformation, but substantially more data are required to fully understand the architecture of the NUP-1 network.
NUP-1 has a predicted C-terminal nuclear localization signal (residues 3633–3643 of 3647) (http://cubic.bioc.columbia.edu/services/predictNLS/). We tested if this sequence was functional by eliminating residues 3633–3647 by fusing GFP in situ to the NUP-1 ORF upstream of the putative nuclear localization signal. This resulted in nuclear localization being disrupted, and which was restored by adding back the nuclear localization signal to the in situ construct, indicating that the region 3633–3647 is necessary and sufficient for nuclear targeting (Figure S1). A more extensive truncation, deleting the C-terminal domain but adding a nuclear localization signal at the new C-terminus, correctly targeted NUP-1 to the nucleus, and the localization of this truncation appeared indistinguishable from the full-length protein. This suggests that the C-terminal domain is not essential for incorporation of NUP-1 into the nucleoskeletal network.
We used RNAi-mediated knockdown to suppress expression of NUP-1 mRNA in BSF and PCF trypanosomes. 24 h (BSF) or 48 h (PCF) after induction of dsRNA, NUP-1 mRNA abundance decreased by ∼35%, corresponding to the onset of proliferative defects (Figure 4A, Figure S2A). By 24 h post-induction NUP-1 protein levels in BSF cells were depleted by ∼75% (Figure 4B) at which time gross alterations to the localization of the DNA as stained by DAPI were observed, including nuclear enlargement, abnormal extensions (blebbing), and irregular boundaries (examples in Figure 4C, quantitated in Figure 4D). These resemble the morphological changes observed in numerous laminopathies [8],[9]. We used BSF cells for subsequent analyses except where specified. Residual NUP-1 at 24 h post-induction was collapsed in patches rather than evenly distributed around the nuclear periphery, suggesting compromised organization (Figure 4E).
NUP-1 knockdown led to a decreased proportion of interphase cells (1K1N) and increased proportion of cells entering mitosis/cytokinesis (2K1N, 2K2N) and atypical cells with abnormal nuclei and atypical copy numbers of nuclei or kinetoplasts (Figure S2C “other”). A peak of blebbing structures was followed by a peak of DNA presenting a diffuse boundary (Figure 4D), suggesting that blebbing might allow DNA to spill out of the nuclear remnant. As blebbing was initially observable in mitotic cells, we suggest that NUP-1 depletion results in loss of structural integrity during mitosis and consequent failure to complete mitosis. Interestingly, some distorted nuclei retained the ability to form a spindle as demonstrated by the presence of an intranuclear gamma-tubulin (KMX) rhomboid in a similar proportion of mitotic cells as seen with uninduced cells. This suggests that NUP-1 is unlikely to play a direct role in spindle formation (Figure S2C). Overall, these data indicate a loss of the normal morphology and hence organization of the nucleus upon NUP-1 depletion, which included abnormal midbody organisation in NUP-1 depleted cells that achieved late mitosis. We suggest that failure to complete mitosis is due to rupture of the nuclear envelope (example; left hand procyclic cell Figure 4C).
By transmission electron microscopy, many knockdown cells had irregular and asymmetric nuclei (Figure 5A–C). Significantly the ER, Golgi, flagellum, and kinetoplast appeared unperturbed, indicating specific nuclear defects (unpublished data). We observed portions of the NE that lost sharp definition, likely due to NE crenelation (multiple small invaginations; 43% of cells, n = 28). Such a feature is much less frequently observed in wild type cells (16% of cells, n = 25) (Figure 4E, arrowhead). Some 18% of cells (n = 35; 0% in wild type, n = 25) exhibited quasi-arrays of circular structures that have the same diameter as NPCs (Figure 5D); clustered NPC arrays are also seen with metazoan lamin defects [39],[40]. To verify the roles of NUP-1 in NPC spacing, and also to provide verification that the NPCs had clustered as suggested by EM, we in situ tagged the trypanosome FG-repeat nucleoporin TbNup98 with GFP [36], and monitored the positioning of this protein following NUP-1 depletion. In uninduced cells TbNup98-GFP was observed as regularly spaced puncta surrounding the nucleus (Figure 5F), consistent with our earlier observations and with TbNup98 being a bona fide nucleoporin [36]. In NUP-1 downregulated cells TbNup98-GFP clustered in patches at the nuclear periphery (Figure 5F, Movie S6). We conclude that NUP-1 is required to correctly position and space NPCs at the nuclear envelope, a function performed by lamins in metazoans.
Given the clear role in maintaining nuclear architecture we asked if NUP-1 functions in chromatin organization. T. brucei contains 22 chromosomes of ∼1.1–6 Mbp referred to as megabase chromosomes, and several intermediate-sized chromosomes of ∼150–400 kbp. These chromosomes collectively carry the housekeeping genes, VSG basic copy genes, and ∼20 subtelomeric VSG expression sites [41],[42]. Additionally, there are ∼100 minichromosomes that contain simple sequence repeats and non-transcribed VSG genes [43]–[45].
We performed fluorescence in situ hybridization (FISH) with a telomere probe recognizing all chromosomes. We observed NUP-1 partially juxtaposed with telomeres throughout the cell cycle, suggesting some coordination in their movements (Figure 6A). Following NUP-1 RNAi, however, telomeres became clustered and some became located within the nuclear blebs (Figure 6B, arrows). To discriminate between megabase and minichromosomes, we simultaneously used FISH probes specific to megabase and minichromosomal telomeres and a minichromosome-specific probe. In uninduced cells, although most of the telomere and minichromosome FISH signals colocalize, some of the megabase chromosome telomeres (those that hybridized exclusively with the telomere probe) were further from the nuclear centre than minichromosomes (which hybridize with both probes) when aligned for nuclear division, as previously reported (Figure 6C, control) [35],[45]. In NUP-1 knockdown cells, megabase telomeres were the predominant telomere FISH signal in the nuclear blebs, where the minichromosome FISH probe was absent (Figure 6C, arrows); the majority of the telomere FISH signal, however, colocalized with the minichromosome FISH probe in the body of the nucleus (Figure 6C, yellow). Therefore, NUP-1 depletion had a more prominent impact on megabase telomere chromosome positioning than on the minichromosomes.
As NUP-1 appears to interact with chromatin, based on the location of the NUP-1 protein within a nucleoskeletal network at the nuclear periphery and telomere and NPC positional effects observed by knockdown, we asked if NUP-1 influences telomere-proximal gene transcription. We used several complementary assays to address this issue.
Expression of MVSG genes, a specific subset of VSG genes, is restricted to metacyclic stage T. brucei (the life stage present in the tsetse fly salivary gland and injected into a host), and these genes are transcriptionally silent in procyclics [46],[47]. The MVSG position, directly upstream of the telomeres, makes them an excellent model to investigate positional effects at subtelomeric sites. Hence, we first induced NUP-1 RNAi in the MVSG 1.22 eGFP PCF reporter cell line (Figure 7A). RNA level derived from two metacyclic VSG genes and also the eGFP transgene integrated into an MVSG locus was monitored by qRT-PCR. NUP-1 knockdown led to a notably (8- to 22-fold) increased abundance of all three MVSG locus mRNAs (Figure 7B). To discriminate between non-specific MVSG induction and NUP-1-dependent effects, we individually silenced three unrelated but essential genes: polo-like kinase (PLK) [48], F0–F1 ATPase associated factor (ATPaseAF) [49], and clathrin [50]. RNAi against each resulted in very severe proliferative and/or morphological defects (Figure S3A) but no significant increase to MVSG transcription (Figure 7B), confirming the specificity of the NUP-1 knockdown effect on misregulation of MVSG expression.
Next we asked if NUP-1 regulates gene expression in BSF T. brucei using a microarray to assess global gene expression changes (GEO accession GSE26256) [51]. Of 8,110 genes represented, relative expression levels of 62 genes were upregulated greater than 2-fold, a standard stringent cutoff for significance in microarray experiments (λ>40 cutoff, corresponding to an estimated false discovery rate of 4.93×10−4) and demonstrated specific and not global transcriptional changes (Figure 8A, Figure S3B, Table S3). The upregulated gene cohort contained procyclin and VSG genes together with a number of small or repetitive ORFs; significantly no other developmentally regulated or housekeeping ORFs were found. No genes exhibited significant downregulation within these parameters.
Seven upregulated genes were procyclins or procyclin-associated genes (p value = 3.54×10−12), residing at two unlinked RNA polymerase I (PolI)–transcribed loci (Figure 8A, Figure S3B, Table S3) [52]. As procyclin is the developmentally regulated protein coat expressed exclusively in PCF cells, derepression in BSF T. brucei suggests NUP-1-dependent life-cycle-specific silencing of the procyclin locus [53]. The upregulation of transcription at the procyclin locus was verified by qRT-PCR specific for the two major forms of procyclin, EP and GPEET (Figure 8B). Though GPEET was not identified on the microarray, possibly due to the stringent cutoff, by qRT-PCR, it was also detected as upregulated.
Most significantly, 26 derepressed genes were annotated as VSGs (p value = 2.03×10−25) (Figure 8A, Figure S3B, Table S3). African trypanosomes achieve expression of a single VSG gene, and thus antigenic variation, by silencing all but one VSG coding sequence present in expression sites and by selective incorporation of a single VSG gene within the expression site body [29],[30],[52],[54]. VSG genes not located in an expression site are transcriptionally silent. As oligonucleotides specific for expression site VSGs were not present on the microarray, the VSGs identified by the microarray could correspond to basic copy VSGs. However, the VSG family is highly homologous and 50% of the VSGs identified by the microarray were strongly homologous to one of the 13 (of an estimated 15) sequenced T. brucei expression site VSGs (unpublished data) [42]. To further investigate if expression site VSGs were misregulated by NUP-1 RNAi and if cross-hybridization with expression site VSGs could account for the derepressed VSGs, we performed qRT-PCR for four expression site VSGs. A highly significant increase in VSG mRNA was observed (Figure 8B). Amplified products were sequenced and aligned with the most homologous microarray oligonucleotide. Expression site VSG sequences from NUP-1 RNAi cells were identical to the published sequences, and large regions of identity were also present between the oligonucleotides and the amplified products (Figure S3C). Though we cannot rule out increased transcriptional activity from basic copy VSGs upon NUP-1 knockdown, we do detect mRNA from misregulated expression site VSGs and therefore can conclude that NUP-1 participates in maintaining the inactive transcriptional status at the expression sites in BSF T. brucei.
The expression site comprises a PolI promoter driving a polycistronic transcription unit, at the distal end of which resides the VSG gene. The VSG gene is proximal to the telomere, and in the case of VSG 2, the entire locus is ∼60 kb [55]. We asked if misregulation associated with NUP-1 knockdown was restricted to expression site VSG transcripts or if the entire expression site was affected. We knocked down NUP-1 in cells containing a GFP-neomycin phosphotransferase (NPT) reporter close to the VSG 2 expression site promoter (Figure 9A). Using qRT-PCR the transcriptional levels of the reporter and the VSG were found to increase upon NUP-1 knockdown, although changes in protein levels were not detected by Western blot (Figure S4). This suggests a role for NUP-1-mediated control of expression levels across the entire expression site (Figure 9A). To investigate a NUP-1 role in transcriptional control of all telomeric gene sequences, we depleted NUP-1 in cells containing a NPT reporter upstream of a de novo telomere, under control of a PolI transcribed ribosomal DNA promoter (Figure 9B) [56]. Though VSG expression increased, confirming the earlier result, NPT mRNA expression decreased during the experiment (Figure 9B). Significantly, NUP-1 knockdown also decreased the expression of the active VSG (VSG 2; Figure 9B); the mechanism behind this is presently unclear but may be related to an increased switch frequency (i.e., that a proportion of cells have switched away from expression of VSG 2; see below). As a control for specificity, we also monitored the effect of knockdown of an NPC protein (TbNup98) and analysed the effect on transcriptional misregulation at subtelomeric regions and also the effect on telomere positioning throughout the cell cycle. No major effect was observed for any of these assays (Figure S5), confirming that the effects seen with NUP-1 knockdown are specific and not the result of generic nuclear insult. Therefore, NUP-1 depletion does not induce an increase in transcriptional levels of all telomeric transcripts and demonstrates a specific role for NUP-1 in the transcriptional regulation of developmentally regulated expression site–associated sequences.
The frequency of VSG switching in the Lister 427 cell line is naturally extremely low, with ∼0.1% of cells having switched away from the predominant VSG2. Hence, we used immunofluorescence to determine if the transcriptional decrease in the expressed VSG and increase in mRNA from the previously silent expression site VSGs coincides with protein production and surface antigen switching. NUP-1 depleted cells were analyzed using antibodies against the expressed VSG (VSG2) and an alternate VSG located in a normally silent expression site (VSG 6). A significant, nearly 10-fold increase in the frequency of cells expressing VSG 6 on their surface following NUP-1 depletion was found (Figure 10A). Double positive cells (i.e., cells that expressed both VSG 2 and VSG 6) were also detected at increased frequency, and as expected due to the slow turnover of the original VSG 2 following initiation of expression of VSG 6. We also observed an increased frequency of cells negative for both VSG 2 and VSG 6 suggesting random switching away from VSG 2 to VSGs other than VSG 6 (unpublished data). These data suggest that NUP-1 depletion not only leads to misregulation of the VSG at the mRNA level, but that this encompasses an increase in the frequency of antigenic switching.
This increase in antigenic switching, together with a role in telomeric positioning, may suggest that NUP-1 plays a role in regulating expression site positioning at the nuclear periphery and thus might affect the frequency of translocation of inactive ESs into the nuclear interior and subsequent insertion into the expression site body. The expression site VSG promoter is rapidly repositioned to the nuclear periphery upon developmental differentiation from BSF to PCF, a life stage where no VSGs are expressed [57]. To investigate this issue, NUP-1 was knocked down in BSF cells where the active expression site promoter has been tagged with GFP-LacI, as a marker for the expression site body. These cells were then induced to differentiate into PFs. The active VSG-expression site promoter relocated to the nuclear envelope early during differentiation (5 h) in 63% of control cell nuclei, as expected (Figure 10B). Statistical analysis of the GFP-LacI position in cells depleted for NUP-1 indicated that only 16%±3.2% of the nuclei displayed the GFP-LacI spot at the nuclear periphery (Figure 10B). As efficient incorporation of the previously active VSG expression site into the peripheral region of the nucleus is disrupted in the NUP-1 knockdown, these data suggest that developmental silencing of the active expression site at the nuclear periphery requires NUP-1 function, consistent with a role for NUP-1 in peripheral chromatin organization.
While structures that resemble the nucleoskeletal metazoan lamina and heterochromatin are recognized in many eukaryotic lineages, the lamin proteins themselves are clearly restricted to the metazoa [7],[15], and no lamina components have been unequivocally identified to date in any non-metazoan organism. This, together with a very clear absence of lamins and a morpholgical lamina from S. cerevisiae and other yeasts, suggests that the lamina arose during differentiation of the metazoan lineage [17]. This distinction implies that the mechanisms for organization of peripheral heterochromatin are substantially divergent throughout the Eukaryota and, more significantly, that the mechanisms understood as present in Metazoa are modern and lineage-specific. An alternate model, that the nuclear lamina is an ancient feature of eukaryotes but has been lost in yeasts, is also possible, especially as it is clear that S. cerevisiae has experienced other significant secondary losses. Resolving this issue requires the identification and molecular and functional characterization of a nucleoskeletal structure from a non-metazoan lineage. Here we describe the location and function of one such nucleoskeletal structure from trypanosomes, which are members of the Excavata supergroup, and in evolutionary terms are extremely distant from the Metazoa and also undergo a closed mitosis, probably the dominant form of mitosis in an evolutionary context.
NUP-1 is a predominantly coiled-coil protein that localizes to a lattice-like network at the nuclear periphery. As initially characterised, NUP-1 was suggested to be located as puncta at the nuclear envelope and was proposed to be a nucleoporin (hence the name NUP-1) [35]. By contrast, our earlier work suggested that NUP-1 was associated with fibrillar material on the inner nuclear envelope, which was depleted from regions of the envelope subtending the NPC [23]. Here, for the first time, using in situ GFP tagging and high-resolution microscopy, we have revealed that NUP-1 is in fact associated with a network, reminiscent of the lamin nucleoskeleton. We find that several aspects of nuclear organization are extremely sensitive to decreased expression of NUP-1, extending to proliferative defects, structural abnormalities, and NPC clustering; all of these are analogous to phenotypes observed when lamin expression or functions are perturbed. Most significantly, NUP-1 organizes trypanosome chromatin and is required for controlling developmentally regulated and telomere-proximal gene expression, as are lamins. NUP-1 knockdown leads to misregulation of multiple expression site VSGs, the procyclin loci, and MVSGs, all developmentally regulated genes. It is possible that NUP-1 interacts even more extensively with chromatin, with subtle changes below our current detection level. NUP-1 and lamins do, however, demonstrate unique functional aspects: first, as trypanosomes undergo closed mitosis, NUP-1 can have no role in post-mitotic nuclear envelope reformation, but may act in mitotic scission, and second, NUP-1 does not, from presently available data, appear to participate in spindle formation. Taken together, these data argue strongly that NUP-1 is a component of a nucleoskeletal cage within the trypanosome nucleus that supports many functions equivalent to the lamin network of metazoan cells.
The role of NUP-1 in silencing developmentally regulated genes may also provide insights into the process of antigenic variation in trypanosomes. Of the changes to gene expression during life cycle progression in trypanosomes, a major feature is the switch between the dominant surface antigens VSG and procyclin. There are over 1,000 VSG coding sequences in the trypanosome genome, of which only ∼20 occur in the subtelomeric expression sites. Monoallelic expression is achieved by only one expression site being active, in a transcriptionally permissive expression site body environment. This level of selective expression requires a tremendous degree of epigenetic control. While the expression site body partially explains how a single expression site is active, other regulatory mechanisms must constrain the inactive expression sites, securing them against both expression site body entry and spontaneous transcription. This control breaks down on disruption of NUP-1 expression. Following NUP-1 knockdown, megabase chromosome telomeres reposition, multiple expression site VSGs become active, and the frequency of VSG coat switching increases. Additionally, the active expression site promoter fails to migrate to the nuclear periphery upon differentiation, suggesting a role for NUP-1 in sequestering and silencing inactive expression sites. As NUP-1 also silences MVSG genes in PCF cells, it is likely associated with the formation and maintenance of a repressive heterochromatin environment, paralleling lamin functions [58],[59]. Modulations to the NUP-1 network, involving NUP-1 phosphorylation sites for example, may be a mechanism to release sequestered VSG genes to initiate a VSG switch, which is clearly under epigenetic control as trypanosomes rapidly, but reversibly, reduce switch frequency in culture [60]. Overall, our work shows that this mechanism contributes to the extreme level of developmental control of VSG and procyclin, where expression between life stages varies by several orders of magnitude [61]. Significantly, misregulation of VSG and procyclin genes on NUP-1 knockdown is not as extreme, being induced only up to ∼10-fold. This likely requires translocation to an RNA PolI-rich nuclear subdomain (i.e., the expression site body and the nucleolus for VSG and procyclin, respectively).
The identification here of a component of a nucleoskeletal network in a non-metazoan organism indicates that such structures are not lineage restricted. Taken together with the presence of peripheral heterochromatin and fibrous nucleoskeletons being identifiable in taxa from many lineages, this suggests that the mechanisms for epigenetic control and the tight regulation of specific gene cohorts at the nuclear periphery are likely both a general feature of eukaryotes and an ancient one. This is particularly significant as the mechanisms for controlling the expression of individual mRNAs are highly divergent. For example, in contrast to Metazoa, trypanosomes lack promoter control for the vast majority of genes. Hence, epigenetic control by a peripheral nucleoskeleton, which is entirely absent from prokaryotes, may be among the most fundamental and ancient mechanisms for eukaryotic gene regulation. The question arises as to whether the functional commonality between NUP-1 and lamins has arisen via convergent or divergent evolution. At molecular weights of ∼450 kDa and ∼60 kDa, respectively, these proteins are far from obvious orthologues, while significant sequence relationships are restricted to Metazoa for lamins [15] and trypanosomatids for NUP-1. NUP-1 is apparently absent from the free-living kinetoplastid Bodo saltans, and the distant relationships between the Leishmania and trypanosome NUP-1 orthologues indicates a great deal of evolutionary plasticity even between more closely related species; African trypanosomes are highly tolerant of perfect repeats, which may explain the extreme expansion of the T. brucei form. Conservation of NPCs, nuclear envelope, and heterochromatin clearly argues for a common origin for nuclear organization mechanisms, but many NPC components cannot be identified based on sequence between distant taxa [17],[36],[62], suggesting that overall architecture is more important than primary structure.
A convergent evolution model would imply that coiled-coil proteins organizing peripheral chromatin arose independently in the major eukaryotic lineages, and potentially that the last eukaryotic common ancestor lacked such a feature. Indeed, as lamins are absent from yeast, the understandable hypothesis was that this represented the ancestral state [16]. By contrast a divergent evolution model argues that ancestral coiled-coil proteins diverged into lamins, NUP-1, and presumably similar proteins throughout the eukaryotes (e.g., NMP-1 in plants, with secondary losses in specific lineages such as yeasts). This latter scenario is clearly the most parsimonious when taking the present data into consideration and is in agreement with recent large-scale studies of the evolution of coiled-coil proteins and other protein domains [63]–[65]. Taken together we favour the divergent scenario where the nuclear skeleton evolved along with the nuclear envelope and NPC in the earliest eukaryotes; however, we acknowledge that this issue remains far from settled and will require identification of nucleoskeletal components in additional eukaryotic supergroups for further clarification. A recent report indicates that the Metazoan exclusivity of lamins is incorrect, and a clear highly divergent lamin ortholog is present in Dictyostelium [79]. This finding has profound implications, strongly suggesting that the fungi have lost an ancestral lamin, rather than metazoa acquiring lamins following separation from the common ancestor with fungi.
Finally, as nucleoskeletal heterochromatin organization appears crucial for silencing developmentally regulated genes, we speculate that the likely presence of such a system in the last eukaryotic common ancestor implies that this organism had a complex multiple-stage life cycle, requiring nucleoskeletal silencing at the nuclear periphery; this may go some way to explaining the surprising cellular complexity of LECA based on multiple reconstructions. Improved proteomic techniques may lead to identification of lamina components in additional eukaryotic lineages and help to clarify the likely structures within an ancestral nuclear skeleton and the fundamental and conserved processes governing heterochromatin remodelling.
BSF Trypanosoma brucei brucei MITat 1.2 (M221 strain) and PCF T. b. brucei MITat 1.2 (Lister 427) or T. brucei EATRO 795 (MVSG studies) were grown as previously described [66]–[68]. Single marker BSF (SMB) and PTT PCF lines were used for expression of tetracycline-inducible constructs [67]. Expression of plasmid constructs was maintained using antibiotic selection at the following concentrations: G418 and hygromycin B at 1 µg/ml and phleomycin at 0.1 µg/ml for BSF, G418 at 15 µg/ml, hygromycin B at 25 µg/ml, and phleomycin at 5 µg/ml for PCF.
ORFs were tagged using the pMOTag4G and pMOTag4H vectors [69] as templates. The following primers were used: NUP-1F: ACAAACACAGCGACAGGTACGGCAAGTCATGGACATACGTAGCACAAGGAAAAGGTCTCGTTCAGCCAATGCGGTCTCGGGTACCGGGCCCCCCCTCGAG; NUP-1R: TCTAGGTGCATGTGTAGATGAACTGCACACTTTATGCACTAATAACAGGTTTGAAGTACTTACCTGGCATCTCCTGGCGGCCGCTCTAGAACTAGTGGAT; Tb927.4.2070F: GAGCTGAGAAAGTGTAATGACTTAGTTATAAAGAGACTAGAGGATGAGGTTAAAGCTCTTCGTGAAGAACTGCGTGGAAATGAGGCGGGTACCGGGCCCCCCCTCGAG; Tb927.4.2070R: CCAATAGAAAAAAATGTAAGTAGCAATAATACGTATTTAAAAATGTCAAAATTGTCAGCAACAAAGATGCTTACACGAACAGAAAAAAAAAGATGGCGGCCGCTCTAGAACTAGTGGAT; Tb927.7.3330F: CGCTCTGTTGAGGAAGGTGAAGACGATGAGGACGAGGGCGACGCAACCGGTTGCCCAACAACGCATTTGGGAGGGCCATGGGCGCACCATGGTACCGGGCCCCCCCTCGAG; Tb927.7.3330R: CAAAAATATTCGTTACATTAGACATCATTCATCGACTGTAACCTAGGTAGTGTATGAGATACCGTATCAATTACACACTGAGTGTCATGGCGGCCGCTCTAGAACTAGTGGAT; TB92733180F: TGGGAATGCTTCAGCAAGTGGTGAAAAGAACAATGCTCCACGGAATCCCTTCTCATTTGGTGCCTCTTCTGGGAATGCTGGTACCGGGCCCCCCCTCGAG; TB92733180R: ACTAAAGAAGGGTAGAAAACAAAGAAAACACCAAATAAGGTACCTGACGCAGCGGCAACACCACGTCGACTTGCTGGCGGCCGCTCTAGAACTAGTGGAT.
For truncation in situ tagging, the following primers were used: NUP-1noNLSF: GGTGAGCTTGTCCGTTGAGTCATCACATCATTCCAGAATCACTGAACAAACACAGCGACAGGTACGGCAAGTCATGGACGGTACCGGGCCCCCCCTCGAG; NUP-1noNLSR: GTTTGAAGTACTTACCTGGCATCTCCTCACGAGACCGCATTGGCTGAACGAGACCTTTTCCTTGTGCTACGTATTGGCGGCCGCTCTAGAACTAGTGGAT; TbNUP1NrepeatsTagF: CCGTACAGCAAAGGAGAAGCTGGAGAGGAGTGTTGAGGAAATATCTTTTTTAAAAGATGAAGTTTTGGTTAGTAATCGTATACGTAGCACAAGGAAAAGGTCTCGTTCAGCCGGTACCGGGCCCCCCCTCGAG; TbNUP1NrepeatsTagR: TCAACATCTGCACCAACAGCACCATCACTATCCCCCACTTTACCATTCAAAGAAGAAACACTATCCACAAGCAATGGCGGCCGCTCTAGAACTAGTGGAT; TbNUP1noNLSplusNLSTagF: GGTGAGCTTGTCCGTTGAGTCATCACATCATTCCAGAATCACTGAACAAACACAGCGACAGGTACGGCAAGTCATGGACATACGTAGCACAAGGAAAAGGTCTCGTTCAGCCGGTACCGGGCCCCCCCTCGAG.
Linear PCR products were purified by ethanol precipitation. Electroporation was performed with 10–25 µg of DNA using a Bio-Rad Gene Pulser II (1.5 kV and 25 µF). Positive clones were assayed for correct insertion and expression using PCR and/or Western blotting.
For microscopy, cells were fixed with 3% paraformaldehyde (v/v) for 1 h on ice (PCF) or 15 min at room temperature (BSF) and allowed to settle onto poly-L-lysine coated slides (VWR International) at room temperature. For permeablization, cells were incubated with 0.1% Triton X-100 for 10 min in PBS. Slides were blocked in 20% FBS (Sigma) in PBS for 1 h. Cells were incubated with primary antibody in 20% FBS/PBS for 1 h followed by three 5-min washes in PBS. Cells were incubated with secondary antibody in 20% FBS/PBS followed by three 5-min washes in PBS. Slides were mounted with Vectashield mounting medium plus DAPI (Vectashield Laboratories). Antibodies were used at the following concentrations: rabbit anti-GFP 1∶3,000, goat anti-rabbit IgG Alexa Fluor 488 (Molecular Probes) 1∶1,000, mouse anti-HA (Santa Cruz Biotechnology Inc.) 1∶1,000, goat anti-mouse IgG Alexa Fluor 568 (Molecular Probes) 1∶1,000, polyclonal rabbit anti-NUP-1 (produced by Covalab against the NUP-1 peptide NH2-CLNAAGVRVRTSQSDKD-COOH) 1∶750, and rabbit anti-TbNog1 629L (gift from M. Parsons) 1∶700. VSG antibodies and visualization were performed as previously described [70]. For combination immunofluorescence and FISH, samples were processed for immunofluorescence as above and post-fixed in 3% paraformaldehyde for 1 h and then processed for FISH as described below.
Confocal microscopy images were acquired with a Leica TCS-NT confocal microscope with a 63×/1.4 or 100×/1.4 numerical aperture objective. Images were processed with Huygens deconvolution software (Scientific Volume Imaging) and Adobe Photoshop (Adobe Systems Inc.). Quantitation was on raw images. Fluorescence images were acquired using a Nikon Eclipse E600 epifluorescence microscope and a Hamamatsu ORCA charge-coupled device camera. Images were captured using Metamorph software (Universal Imaging Corp.) and raw images processed using Adobe Photoshop.
107 cells per lane were resolved on 4%–12% SDS–polyacrylamide gels (Invitrogen). Proteins were transferred to polyvinylidene fluoride membranes (Millipore). Detection of HRP-conjugated secondary antibodies was by chemiluminescence with luminol (Sigma). Polyclonal rabbit anti-TbBiP serum (a gift from J. D. Bangs) was used at 1∶5,000, polyclonal mouse anti-NUP-1 (a gift from K. Ersfeld) at 1∶10,000, mouse anti-VSG 2 (a gift from G.A.M. Cross) at 1∶20,000, rabbit anti-NPT (Fitzgerald Industries International, Inc.) at 1∶5,000, and polyclonal anti-TbRab11 was used at 1∶2,000. Incubations with the appropriate commercial secondary anti-IgG horseradish peroxidase (HRP) conjugates (Sigma) were performed at 1∶10,000 in TBST for 45 min. Image J (National Institutes of Health) was used to quantify band intensity.
RNA was purified using the RNeasy mini kit (Qiagen) according to the manufacturer's instructions. RNA concentration was quantified using a ND-1000 spectrophotometer and Nanodrop software (Nanodrop Technologies). cDNA was produced using Superscript III Reverse Transcriptase (Invitrogen) according to the manufacturer's instructions. qRT-PCR using cDNA templates was performed with iQ-SYBRGreen Supermix and MiniOpticon Real-Time PCR Detection System (Bio-Rad) and quantified with Opticon3 software (Bio-Rad). β-Tubulin, which has stable mRNA levels throughout the T. brucei cell and life cycles, was used for normalization [71]. The following primers were used for qRT-PCR: 3330qRTF: GGGTGTTTTCGTTGATGAGGTCT; 3330qRTR: ACTGCAGCGAAGACAAAGAAGAG; 2070qRTF: GGTTCAGGAGGAGACGATGAAGT; 2070qRTR: AGCTTTAACCTCATCCTCTAGTCTCT; Nup1qRTF: CGAGGAGGAGGTTGGAGGAG; Nup1qRTR: GCTGGCACTCCTTCTGCAATTT; TbBetaTubulinF: CAAGATGGCTGTCACCTTCA; TbBetaTubulinR: GCCAGTGTACCAGTGCAAGA; VSG2F: CCAAGTTAACGACTATACTTGCCTATT; VSG2R: CAAGTAGCAAGGAAAATTTTAAAAGG; VSG21F: CGGATGCTCAAATCTATTACACAG; VSG21R: GTCAGAATTCTTAGAATGCAGCC; VSG16F: AGTCGTAGCACTTTTGATTCAGG; VSG16R: TTATGCTAAAAACAAAACCGCA; VSG17F: CACCAACACAGCAGAACGAA; VSG17R: TTATGCTAGAATCAAAAATGCAAGC; EPProcyclinQF: AAGGACCAGAAGACAAGGGTC; GPEETProcyclinQF: ATGGCACCTCGTTCCCTT; ProcyclinQR: TTAGAATGCGGCAACGAGA; BIPQF: GGTGAGCGCTATGGACAAGT; BIPQR: GTCCTCGTCCTCGAACTCTG; NOG1QF: GCTCACTTAGCGTAAACCGC; NOG1QR: GTATGGCACGATCACCCTCT; HistoneH3QF: GACCTGCTGCTACAAAAGGC; HistoneH3QR: AAGCAGCGACACAATGTACG; HistoneH4QF: TATCTACGACGAGGTGCGTG; and HistoneH4QR: AACCGTACAGAATCTTGCCG.
For sequencing qRT-PCR products, the following primers were used: VSG2sequence: TTTGAAGTTTTAACCCAGAAGCA; VSG21sequence: AAAAAAGAAAAACCAAGCAAGTCC; VSG16sequence: AGAAATGTCCAGCGAGCC; and VSG17sequence: AAGAACAGGAAACTACAAAGGTAGACG.
Cells were fixed in 2% glutaraldehyde (v/v), resuspended in 0.1 M PIPES, and processed for electron microscopy by the University of Cambridge multi-imaging centre. Sections were viewed on a Philips CM100 electron microscope (FEI-Philips) operated at 80 kV.
PCR products were cloned into either the p2T7TABlue plasmid (BSF) [72] or the p2T7-177 plasmid (PCF) [73]. RNAi was induced in log-phase cells by the addition of 1 µg/ml tetracycline. The following primers were designed using RNAit [74]: NUP-1RNAiF: ATCGAAACGTGAGGGTGAAC; NUP-1RNAiR: ACCCTTGTCTTGGCATATCG; Tb927.7.3330RNAiF: CCACAGAACACACCGAAATG; Tb927.7.3330RNAiR: CCTTCTCGTCCAACTCAAGC; Tb927.4.2070RNAiF: TGATCCATTCCCTTGAGGAG; Tb927.4.2070RNAiR: GGGGAGGTGTGTGTCACTCT; NUP-1177RNAiF: GCTACTCGAGATCGAAACGTGAGGGTGAAC; and NUP-1177RNAiR: CGTAGGATCCACCCTTGTCTTGGCATATCG.
The telomere PNA FISH kit (DAKO) was used according to the manufacturer's instructions. For combined FISH with the Telomere PNA kit and the Digoxigenin labelled 177 base pair fragment that binds to minichromosomes, cells were pre-treated according to the kit manufacturer's instructions and then prehybridized in hybridization buffer (50% formamide, 2× SSC, 10% dextran sulfate, 50 mM sodium phosphate pH 7) for 45 min. Probe (10 µl kit probe, 5 µl hybridization buffer+minichromosome probe) was added and cells were denatured for 5 min at 88°C. Slides were incubated overnight at 37°C. Slides were washed in 50% formamide/2× SSC for 30 min at 37°C, 2× SSC for 10 min at 50°C, 0.2× SSC for 60 min at 50°C, and 4× SSC for 10 min at room temperature. A rhodamine-conjugated anti-digoxigenin antibody (Fab fragments, Roche) was added at 1∶200 in BMEB (100 mM maleic acid, 150 mM sodium chloride, pH 7.5, 1% blocking reagent; Roche). Slides were washed in 1× TBS+0.05% Tween 20 and imaged as described.
We searched for NUP-1 homologues in representative taxa of major eukaryotic supergroups using BLAST. Homo sapiens data were obtained from NCBI (www.ncbi.nlm.nih.gov). Saccharomyces cerevisiae was at the Saccharomyces genome database (http://www.yeastgenome.org/). Drosophila melanogaster data were at FlyBase (www.flybase.org). Caenorhabditis elegans was obtained from WormBase (www.wormbase.org). Chlamydomonas reinhardtii, Ostreococcus tauri, Thalassiosira pseudonana, Phytophthora ramorum, and Naegleria gruberi data were obtained from the Joint Genome Initiative (genome.jgi-psf.org). Cryptococcus neoformans, Theileria parva, Tetrahymena thermophila, and Trichomonas vaginalis data were from TIGR (www.tigr.org). Dictyostelium discoideum, Entamoeba histolytica, Plasmodium falciparum, Trypanosoma brucei, Trypanosoma congolense, Trypanosoma brucei gambiense, Trypanosoma vivax, Trypanosoma cruzi, Leishmania major, Leishmania infantum, Leishmania braziliensis, and Bodo saltans data were obtained from geneDB (www.genedb.org). More complete Trypanosoma congolense and Trypanosoma vivax sequences were a generous contribution from Andrew Jackson at the Wellcome Trust Sanger Institute. Toxoplasma gondii data were from ToxoDB (www.toxodb.org), Cryptosporidium parvum data from CryptoDB (www.cryptodb.org), and Cyanidioschyzon merolae data were retrieved from the C. merolae genome BLAST server (merolae.biol.s.u-tokyo.ac.jp). Euglena gracilis data were obtained from the Euglena gracilis genome project (http://web.me.com/mfield/Euglena_gracilis/E._gracilis.html).
Labelled cDNA probes were created from 2 µg of total RNA using the SuperScript indirect cDNA labelling kit. Unincorporated nucleotides were removed by QuickClean resin, and the resulting cDNA was ethanol precipitated overnight at −20°C. Control and experimental samples were labelled with Alexa Fluor 555 and 647, respectively, and purified using the cDNA labelling purification module. Equal amounts of labelled control and experimental samples were combined. Expression microarray was performed with dye-swaps and experimental and technical replicates as described previously [75]. The log2 of the average mRNA abundance ratios are reported. Differentially expressed genes were identified by maximum-likelihood analysis (λ>40) [76].
Cells in log phase growth were trapped between a slide and coverslip at 27°C for no longer than 30 min. FRAP experiments were performed on a Zeiss inverted microscope (Axiovert 200) with a 63×/1.4 numerical aperture objective and spinning disk confocal scanning unit (CSU22, Yokogawa). Time-lapse sequences were acquired using Volocity software with an EMCCD camera (C-9100, Hamamatsu) operating in streaming mode. Representative movies are presented. Fluorescence intensity was measured in the region of interest using ImageJ. For the experiments on cells expressing NUP-1-GFP, 28 image sequences were acquired in four independent FRAP experiments. Movies were taken at either maximum speed of acquisition (no time-lapse, 0.221 s between frames, 311 time points) or a time-lapse of 1.98 s between frames (70–140 time points) with an exposure time of 200 ms/frame (binning 1×1 pixels). For the experiments on cells expressing NLS-GFP, 30 image sequences were acquired in five independent FRAP experiments. Movies were taken at maximum speed of acquisition (no time-lapse, 0.068 s between frames, 588–625 time points) with an exposure time of 50 ms/frame (binning 1×1 pixels). To normalize data, background fluorescence was subtracted and overall photobleaching over time normalized using another equivalent source of GFP (a similar nuclear region of a cell that had not been bleached).
RNAi line EATRO 795 2913 MVSG1.22->EGFP was obtained by consecutive transfection of EATRO795 PCF cells with pLew13, pLew29, and with a cassette replacing MVSG 1.22 with eGFP. Full details will be published elsewhere. PLK, ATPaseAF, clathrin, and NUP-1 were depleted by RNAi. Three independent RNAi lines were examined for each gene. To generate PLK:RNAi (Tb927.7.6310) and F0–F1 ATPaseAF:RNAi (Tb10.70.7760) lines, the following primers were used to amplify cDNA fragments: PLKBamHI: GATAGGATCCTTCCCACTGTTTGGGTGACG; PLKXhoI: GAATCTCGAGCAAGCCGTGCAGCAATTTCTCTAG; Tb7760BamHI: GATCGGATCCGAAGCTCAGGACC; and Tb7760XhoI: GATACTCGAGGCAGAAACGCATC.
The corresponding PCR fragments were cloned into the BamHI/XhoI sites of the pZJM plasmid [77] and transfected to the EATRO 795 RNAi line. Clones were selected in the presence of G418 (5 µg/ml), hygromycin B (5 µg/ml), blasticidin S (5 µg/ml), and phleomycin (1.75 µg/ml). Clathrin and NUP-1 RNAi lines were selected in the same way, but DNA fragments were cloned into the p2T7-177 plasmid. For growth rates, T. brucei cultures were diluted to ∼7×105 cells/ml and cell densities of parasite cultures evaluated daily. RNAi induced cultures were grown in the presence of tetracycline (2 µg/ml). Cell cultures were diluted to the same density (6–7×105 cells/ml) when cultures reached 1–1.2×107 cells/ml. For quantitative RT-PCR, samples were collected at different time points after TET induction dependent on RNAi line response. Quantitative PCR was carried out on cDNA templates using Power SYBR Green PCR Master Mix (Applied Biosystem, Invitrogen). The following primers were used for qRT-PCR: 161QF: GGCGGTTTGTCTTTGTTTTTG; 161QR: TGACTCCTCTTTGTTGTCGTCTTC; 164QF: CACAGACCTGCAGATGCACTTTAT; 164QR: TGCCTTTATCTTTGCTAAATTTGCT; GFPQF: CTGCTGCCCGACAACCA; and GFPQR: TGTGATCGCGCTTCTCGTT.
NUP-1 or TbNup98 RNAi fragments were cloned into the pRPaiSL stem loop RNAi plasmid [77] following amplification with 5iNUP (5′-GATCGGGCCCGGTACCATCGAAACGTGAGGGTGAAC) and 3iNUP (5′-GATCTCTAGAGGATCCACCCTTGTCTTGGCATATCG) or 3iNup98 (5′-GATCTCTAGA GGATCCATAACCGTACGCCTTTGTGC) and 5iNup98 (5′-GATCGGGCCCGGG TGGAGCGGCGTAGTAGAAGT) and digestion with KpnI/BamHI or XbaI/BamHI (sense fragment) and Bsp120I/XbaI or Bsp120I/XmaI (antisense fragment). Plasmid was linearized with AscI and transfected into 2T1 strain [78] modified with a sub-telomeric NPT reporter 2 kb upstream of a de novo telomere (2T1.R2; [56]) or a GFP:NPT reporter downstream of the repressed VSG2 expression site promoter (2T1.GFP:NPT; [55]). Three independent RNAi lines were examined for each experiment. RNA was prepared using a Qiagen RNeasy kit following induction of RNAi (1 µg ml−1 tetracycline) for 24 and 48 h. The following primers were used for qRT-PCR: NPTQF: TCTGGATTCATCGACTGTGG; NPTQR: GCGATACCGTAAAGCACGAG; RAB11QF: ATCGGCGTGGAGTTTATGAC; and RAB11QR: GTGGTAAATCGAACGGGAGA.
NUP-1 RNAi was induced for 24 h in a cell line with a GFP-tagged active VSG ES promoter. Cells were differentiated and analyzed as in [56].
|
10.1371/journal.ppat.1006537 | Zika virus preferentially replicates in the female reproductive tract after vaginal inoculation of rhesus macaques | Zika virus (ZIKV) is a mosquito-transmitted virus that can cause severe defects in an infected fetus. ZIKV is also transmitted by sexual contact, although the relative importance of sexual transmission is unclear. To better understand the role of sexual transmission in ZIKV pathogenesis, a nonhuman primate (NHP) model of vaginal transmission was developed. ZIKV was readily transmitted to mature cycling female rhesus macaque (RM) by vaginal inoculation with 104–106 plaque-forming units (PFU). However, there was variability in susceptibility between the individual RM with 1–>8 vaginal inoculations required to establish infection. After treatment with Depoprovera, a widely used contraceptive progestin, two RM that initially resisted 8 vaginal ZIKV inoculations became infected after one ZIKV inoculation. Thus, Depoprovera seemed to enhance susceptibility to vaginal ZIKV transmission. Unexpectedly, the kinetics of virus replication and dissemination after intravaginal ZIKV inoculation were markedly different from RM infected with ZIKV by subcutaneous (SQ) virus inoculation. Several groups have reported that after SQ ZIKV inoculation vRNA is rapidly detected in blood plasma with vRNA less common in urine and saliva and only rarely detected in female reproductive tract (FRT) secretions. In contrast, in vaginally inoculated RM, plasma vRNA is delayed for several days and ZIKV replication in, and vRNA shedding from, the FRT was found in all 6 animals. Further, after intravaginal transmission ZIKV RNA shedding from FRT secretions was detected before or simultaneously with plasma vRNA, and persisted for at least as long. Thus, ZIKV replication in the FRT was independent of, and often preceded virus replication in the tissues contributing to plasma vRNA. These results support the conclusion that ZIKV preferentially replicates in the FRT after vaginal transmission, but not after SQ transmission, and raise the possibility that there is enhanced fetal infection and pathology after vaginal ZIKV transmission compared to a mosquito transmitted ZIKV.
| Zika virus was introduced to Brazil in 2015 and it rapidly spread to all of tropical America. Although Zika virus infection is usually mild in adults, it can cause severe birth defects in the developing fetus that makes it critical to prevent ZIKV infection in women who are pregnant or who could become pregnant. Although Zika virus is transmitted primarily by mosquito bite, it can also be transmitted by sex. To understand the role of sexual transmission in Zika virus disease, we inoculated rhesus monkeys intravaginally with the virus and monitored virus in blood and reproductive tract secretions. ZIKV was detected in the female reproductive tract before it was detected in plasma and replication levels in the female reproductive tract did not reflect ZIKV levels in other parts of the body. Thus ZIKV prefers the reproductive tract after vaginal transmission suggesting that fetal disease could be more common or severe after vaginal ZIKV transmission compared to a mosquito transmitted ZIKV infection.
| Zika virus (ZIKV) was first isolated in the Zika forest of Uganda in 1947 (21, 22, 30) and the first descriptions of human disease were reported a few years later (2, 53). ZIKV has a positive-sense RNA genome and belongs to the genus Flavivirus, which also includes dengue virus (DENV), Yellow Fever virus, Japanese encephalitis virus, and West Nile virus (WNV) (30). In approximately 20% of infected humans, ZIKV causes a febrile illness that can include rash, arthralgia and conjunctivitis. In addition, ZIKV has been associated with the development of microcephaly and lissencephaly and ocular lesions in infants born to women who acquired the infection during early pregnancy. In adults, ZIKV infection has also been associated with Guillan-Barré syndrome and other neurological complications including hearing loss and tinnitus. Although ZIKV is a mosquito-transmitted virus, sexual transmission of ZIKV in humans has been documented in several settings [1–13]. After returning to the U.S. from Africa, a man infected his partner [2] and male-to-female [14], male-to-male [5] and female-to-male [10] sexual transmission of ZIKV have been reported in travelers returning to the U.S. from ZIKV positive regions in the Americas. ZIKV was isolated from semen during the ZIKV outbreak in French Polynesia in 2013 [4] and infectious virus has been isolated from semen up to 24 days after the onset of symptoms [9]. Further, ZIKV RNA has been detected in semen up to 6 months after onset of symptoms [15,16] and in the semen of a vasectomized man up to 96 days after onset of symptoms [6]; however, the infectivity and transmission potential of persistent ZIKV RNA in semen is not known. Of significant concern, a case of male-to-female sexual transmission of ZIKV from an asymptomatic male traveler to a woman with no travel history has been reported [8]. This case suggests that transmission via semen is possible even if a man has minimal or no symptoms.
In 2007, an Asian lineage ZIKV outbreak from mosquito transmission was reported in Yap Island with 185 clinical cases and an estimated 5000 infections (75% of the population) in just 3 months [17,18]. Six years later (in 2013), another ZIKV outbreak involving 28,000 infected people was reported approximately 5000 miles away in French Polynesia (FPY) [19]. The ZIKV strain in the FPY outbreak had 99.9% nucleotide and amino acid identities with the Asian ZIKV strain in the Yap Island outbreak [17,19,20], suggesting that the virus in French Polynesia outbreak was imported from Yap Island. Given the distance between the two locations it is unlikely that mosquitoes introduced ZIKV into FPY; it is more likely that an infected person imported ZIKV to FPY. ZIKV subsequently spread from FPY to other Pacific Islands, and by 2014 imported cases and cases of autochthonous transmission were reported in New Caledonia, Easter Island and the Cook Islands [21,22]. The nucleotide sequence of the ZIKV strain in all these outbreaks was 99.9% identical to the ZIKV strain in the Yap Island and FPY outbreaks. In March 2015, the first cases of autochthonous transmitted ZIKV were reported in Bahia, Brazil with a ZIKV strain that was 99.9% identical (nucleotide and aa sequences) to the ZIKV strain in the Yap Island and FPY outbreaks [23,24]. Based on this chain of events and the similarity of the ZIKV strains involved, it is generally accepted that ZIKV moved from the Pacific Islands to South America. ZIKV mosquito vectors are endemic in the Pacific Islands and Brazil [25,26] and ZIKV is readily transmitted between humans by sexual activity [1–13]. Thus, it is likely that one or more infected individuals imported ZIKV over considerble distances to these widely separated islands and countries and then served as reservoir hosts for mosquito transmission, or transmitted ZIKV by sex, to naïve persons.
The World Health Organization declared the ZIKV pandemic a public health emergency on February 1, 2016, and in November 2016, WHO declared Zika virus endemic in the Americas. As of May 2017, more than 5,109 cases of ZIKV infection have been reported in the United States, excluding those in Puerto Rico, Virgin Islands and Guam. Most infections are in travelers returning from affected areas, but 266 ZIKV infections were acquired in the continental US. Of these US acquired infections, 221 infections (83%) were transmitted through mosquito bites in Florida and Texas, while 45 infections (17%) were sexually transmitted [27]. It is now estimated that 1.6 million people are, or have been, infected with ZIKV in the Americas. Despite these observations, the frequency and efficiency of sexual ZIKV transmission is unclear. To better understand the biology of ZIKV sexual transmission, we developed a RM model of vaginal ZIKV transmission.
The captive-bred mature (> 5year old) parous, cycling female rhesus macaques (Macaca mulatta) used in this study were from the California National Primate Research Center. All animals were negative for antibodies to WNV, HIV-2, SIV, type-D retrovirus, and simian T cell lymphotropic virus type 1 at the time the study was initiated. The animals were housed in accordance with the recommendations of the Association for Assessment and Accreditation of Laboratory Animal Care International Standards and with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The Institutional Animal Use and Care Committee of the University of California, Davis, approved these experiments (Protocol # 19471). When immobilization was necessary, the animals were injected intramuscularly with 10 mg/kg of ketamine HCl (Parke-Davis, Morris Plains N.J.). All efforts were made to minimize suffering. Details of animal welfare and steps taken to ameliorate suffering were in accordance with the recommendations of the Weatherall report, "The use of non-human primates in research". Animals were housed in an air-conditioned facility with an ambient temperature of 21–25°C, a relative humidity of 40%-60% and a 12 h light/dark cycle. Animals were individually housed in suspended stainless steel wire-bottomed cages and provided with a commercial primate diet. Fresh fruit was provided once daily and water was freely available at all times. A variety of environmental enrichment strategies were employed including housing of animals in pairs, providing toys to manipulate and playing entertainment videos in the animal rooms. In addition, the animals were observed twice daily and any signs of disease or discomfort were reported to the veterinary staff for evaluation. The menstrual cycles were assessed on the basis of menstrual bleeding, with the first day of menses designated day 0 of the cycle. For sample collection, animals were anesthetized with 10 mg/kg ketamine HCl (Park-Davis, Morris Plains, NJ, USA) or 0.7mg/kg tiletamine HCl and zolazepan (Telazol, Fort Dodge Animal Health, Fort Dodge, IA) injected intramuscularly. The animals were sacrificed by intravenous administration of barbiturates.
Plasma from a ZIKV infected blood donor was used to produce the ZIKV stock for these studies. The donated blood was collected at the Hematology and Transfusion Center, Hospital of Clinics, Universidade Estadual de Campinas-UNICAMP, Campinas, SP, Brazil, and after it was found to positive for ZIKV by RT-PCR, the fresh frozen plasma was released for research. However, no donor personal identification information accompanied the sample and thus, the donor is anonymous [28] and IRB approval was not needed to isolate virus from the sample.
We produced a high titer ZIKV stock from the plasma of a Brazilian blood donor [28] by short-term culture on Vero cells (ATCC, Manassas,VA). The plasma was an aliquot of the same plasma sample from which strain Zika virus/H.sapiens-tc/BRA/2015/Brazil_SPH2015 was isolated [28]. The ZIKV stock contained approximately 107 PFU/ml of infectious virus when titrated by Vero cell plaque assay and approximately 6x109 vRNA copies/ml by the Taqman RT-PCR described below. The atraumatic virus inoculation procedure consisted of inserting a 1 CC needless tuberculin syringe containing 1 ml of the ZIKV stock into the vagina until the tip touched the cervix. Then the syringe was gently withdrawn while the viral inoculum was expelled. This procedure was repeated weekly until an animal was plasma ZIKV RNA+ on 2 consecutive time points (Fig 1). Animals that remained uninfected after 8 vaginal ZIKV inoculations were treated with Depoprovera using published protocols [29] that have been used to enhance vaginal SIV transmission in RM. Briefly, 4 weeks before, and on the day of, challenge with ZIKV, 30 mg of Depo-Provera [29] was administered by intramuscular injection.
The Zika virus inoculum was sequenced in duplicate using a method adapted from Quick et. al. [30]. Briefly, viral RNA was isolated from 1 ml of cell culture supernatant using the Maxwell 16 Total Viral Nucleic Acid Purification kit. Approximately 1.4x105 viral RNA templates were converted into cDNA using the SuperScript IV Reverse Transcriptase enzyme. The cDNA was then split into two multi-plex PCR reactions using the PCR primers described in Quick et. al with the Q5 High-Fidelity DNA Polymerase enzyme. PCR products were then tagged with the Illumina TruSeq Nano HT kit and sequenced with a 2 x 300 kit on an Illumina MiSeq. Fastq reads were analyzed using a series of custom scripts generated in Python, as follows. First, up to 1000 reads spanning each of 35 amplicons were extracted from the data set. Extracted reads were then mapped to the Zika reference for PRVABC59 and Zika virus (strain Zika virus/H.sapiens-tc/BRA/2015/Brazil_SPH2015). Variant nucleotides were then called using SNPeff, using a 5% cutoff. The output.vcf and.bam files could be interrogated in Geneious and differences between the inocula and reference strains could be determined.
Blood was collected from the femoral vein by venipuncture 3–4 times a week, on the day of ZIKV inoculation and, 2, 4, and often 6, days later. Urine samples were collected from pans placed under the animals’ cages on the days that blood samples were collected. Cervicovaginal lavages (CVL) were also collected on the days blood samples were collected by vigorously infusing 1–2 ml of sterile PBS into the vaginal canal and aspirating as much of the instilled volume as possible. Care was taken to insure that the cervical mucus was included in the lavage fluid and that no trauma to the mucosa occurred during the procedure. One half of the CVL sample was snap frozen on dry ice and stored at −80°C until analysis. The remainder was spun and the resulting cell pellet was used RNA isolation. The supernatant was treated with 10× Protease Inhibitor (Roche/Sigma Aldrich, St Louis Mo) and subsequently used for cytokine and chemokine quantitation. The lavage for sample collection and the preparation procedure resulted in at least a 10-fold dilution of the cervicovaginal secretions. RNA was isolated from 1 ml urine, EDTA blood plasma, or CVL by QIAamp UltraSens Virus Kit (Qiagen, Redwood City CA) following the manufacturer’s protocol.
Genital tract tissues (vulva, vagina, cervix, uterus, ovary) and genital lymph nodes (inguinal, obturator and iliac lymph nodes), gut tissues (duodenum, jejunum, ileum, colon and mesenteric lymph nodes), oral tissues (lip/cheek pouch, tonsil, tongue, parotid salivary gland) distal lymphoid tissues (axillary, bronchial lymph nodes and spleen), urinary tract (bladder, kidney), CNS (Frontal cortex, temporal lobe, eye), cerebrospinal fluid (CSF) and blood were collected at the time of necropsy and analyzed for ZIKV RNA levels. Tissues were stored in RNAlater (Ambion, Austin, TX) and kept at -20°C until preparation of RNA. Tissues stored at -20°C thawed and removed from RNAlater were diced with a razor blade in a sterile petri dish as small as possible. Tissues fragments were then placed into 2.0ml screw cap Sarstedt tubes with 1 x 7mm stainless beads (Qiagen, Redwood City CA) added per tube with 600ul RLT Buffer and shaken 5 mins in bead beater to homogenize. The homogenates were processed using the Qiagen RNeasy Mini Kit (Qiagen, Redwood City CA) to extract total RNA with optional DNase treatment on column per manufacturer’s instructions. Skin and fibrous tissues were treated with additional proteinase K digestion as described in Appendix C of the kit handbook. Brain samples required 5ul of Reagent DX to prevent excessive foaming.
We used monolayers of Vero cells (ATCC, Manassas, VA) to isolate infectious virus from selected tissue samples collected from the ZIKV-inoculated animals at necropsy. Briefly, up to 107,tissue mononuclear cells isolated from tissues were added to a confluent monolayer of Vero cells in 6-well plates (Costar Inc., Cambridge, MA) for tissues yielding <106 cells), or T25 flasks (Costar Inc.) for tissues yielding > 106 cells. The co-cultures were incubated at 37°C and culture supernatants were harvested at 2, 4 and 7 days after initiation. The supernatants were assayed for the presence of ZIKV RNA by qRT-PCR (described below). A sample was considered to be positive for infectious virus if the vRNA levels steadily increased in supernatants of the corresponding co-culture. No effort was made to titer the levels of infectious virus in samples.
For urine, plasma, and CVL samples, 25ul of eluted RNA was converted to cDNA with Superscript III (Thermo Fisher Scientific, Waltham, MA) using random primers in a 60ul reaction and quantified in quadruplicate by qPCR on an Applied Biosystems QuantStudio 6 Flex Real-Time PCR System using 2x Universal Taqman Master Mix (Thermo Fisher Scientific, Waltham, MA) with published primers and probe that target the ZIKV E glycoprotein from Lanciotti et al [17] (forward 5’-CGYTGCCCAACACAAGG-3’, reverse 5’-CACYAAYGTTCTTTTGCABACAT-3’, and probe 5’-6fam AGCCTACCTTGAYAAGCARTCAGACACYCAA-BHQ1-3’). All RNA samples were tested in 4 replicate PCR reactions carried out in 96-well optical plates (Applied Biosystems, Foster City, CA). All PCR reactions included primers and probes for GAPDH to detect problems with the assay or RNA isolation and all plates contained several wells that held only 25μl nuclease free water to detect contamination. Standard curves for the ZIKV E glycoprotein primers and probe assay were generated on every plate by making 10-fold dilutions of a purified 444bp E glycoprotein PCR fragment starting at a known concentration. The 444bp PCR fragment was generated for this purpose by PCR amplification from ZIKV stock cDNA using primers z_F: 5’-CATACAGCATCAGGTGCATAGGAG-3’, z_R: 5’-AGCCATGAACTGACAGCATTATCC-3’ with Phusion HotStart II DNA Polymerase (Thermo Fisher Scientific, Waltham, MA). The fragment was purified with QIAquick PCR Purification Kit (Qiagen, Redwood City CA) and the concentration calculated using the average of 6 independent spectrophotometer readings (Nanodrop/\, Thermo Fisher Scientific, Waltham, MA). Five 96-well plates of individually serially diluted standard curves with concentrations ranging from 107 copies/well to 1 copy/well were run to generate the line equation used to analyze all qPCR assays. For each 96 well plate, 11 wells of each dilution were run including positive and no-template controls. When the dilution of this fragment is done correctly, we generate 7–10 positive wells out of 10 at the 10-copy range and 2–3 positive wells out of 10 in the single copy range. Thus, the assay can detect a single copy of ZIKV env cDNA per well. To determine the sensitivity of the assay in actual samples, 10-fold serial dilutions of vRNA from the ZIKV stock were added to plasma or RNA extracted from a mesenteric LN collected from a ZIKV negative RM. The assay was negative when 10–15 copies of ZIKV RNA were added to the cDNA synthesis reaction, which results in about 1 vRNA copy in each well. However, 6 of 6 wells were positive when 100–150 copies of ZIKV RNA were added to the cDNA synthesis reaction, which is equivalent to 10–13 copies of vRNA in each well. There was no amplification of ZIKV E glycoprotein sequences from the RNA isolated from any plasma or tissue samples from ZIKV negative animals. Thus, we estimate that the limit of quantitation in this ZIKV E glycoprotein PCR assay is 120 vRNA copies/ml of CVL, plasma or urine. While in tissue samples, the limit of quantitation is 33 vRNA copies/ug of total tissue RNA analyzed. Viral load data from plasma, urine, and CVL are expressed as vRNA copies/ml. Viral load data from tissues are expressed as vRNA copies/ug total RNA.
A commercial ELISA kit was used to test for the presence of ZIKV-specific-antibodies in plasma and CVL of inoculated animals. The NHP Zika virus serology test kit (XpressBio, Frederick MD) uses a Ugandan ZIKV NS1 protein as the capture antigen. There is about 97.5% amino acid identity between the Ugandan virus and contemporary circulating Asian ZIKV virus strains in the NS1 region. The kit was used as directed by the manufacture to test plasma samples. CVL samples, processed as described above, were diluted 1:1 and 1:2 and tested with the same kit.
Twenty-nine cytokines, chemokines and growth factors were measured in plasma and CVL samples using the Monkey Cytokine Magnetic 29-Plex Panel for the Luminex (Invitrogen, Carlsbad CA) according to the manufacturer's instructions. The analytes measured included IL-1β, IL-1RA, IL-2, IL-6, IFN-γ, IL-12, CCL3, CCL5, CCL11, CXCL8, CXCL9, CXCL10, CXCL11, and MIF. EDTA-plasma samples were diluted up to four fold with assay diluent and CVL samples were diluted up to 4 fold with a 1:1 mixture of PBS and assay diluent. Samples were incubated with antibody-coupled beads for 2 hours at room temperature, followed by incubation with a biotinylated detection antibody for 1 hour and streptavidin-phycoerythrin for 30 minutes. Each sample was assayed in duplicate, and cytokine standards supplied by the manufacturer were run on each plate. Multianalyte profiling was performed using a Luminex-100 system, and data were analyzed using Miliplex analyst software, version 5.1 (Millipore/Fisher Scientific, Waltham, MA). The median level of each analyte in a sample is reported. For these analytes, the sensitivity of the assay ranges from 0.5–20 pg/ml plasma according to the manufacturer.
GraphPad Prism version 5 for Apple OSX10.4 (GraphPad Software, San Diego California USA) and Macintosh computers (Apple Inc., Cupertino CA) were used for statistical analysis and graphing the data.
Zika virus strain PRVABC59; genbank accession number KU501215
Zika virus strain/H.sapiens-tc/BRA/2015/Brazil_SPH2015; genbank accession number KU321639.1
We produced a high titer (107 PFU/ml/6x109 vRNA copies/ml) ZIKV stock by culturing the plasma of a Brazilian blood donor [28] on Vero cells. The isolate was confirmed by next generation sequencing to be an Asian-lineage ZIKV. We mapped the sequences to the Zika-PRVABC59; genbank accession number KU501215 and Zika virus/H.sapiens-tc/BRA/2015/Brazil_SPH2015; genbank accession number KU321639.1. We found that nucleotide and AA sequence of the major variant in our Zika virus stock was identical to the Brazilian KU321639.1 reference sequence but had 35 positions with fixed nucleotide differences compared to the Puerto Rican KU501215 reference sequence. There were two other minor variants present in the stock at a frequency of between 5–10%; the defining nucleotide differences were not in a location of repeated nucleotides. Thus, our ZIKV stock is essentially clonal as it contains only a few infrequent variations from a single ZIKV sequence.
It has been reported that the dose of WNV or Dengue virus in an infected mosquito bite ranges from 104−106 PFU [31,32]. While the level of infectious ZIKV in semen is unknown, ZIKV RNA levels of 10 7–10 8 vRNA copies/ ml semen have been reported [4,33]. As one of the purposes of our study was to define the dose of ZIKV required for vaginal transmission, we chose to use a similar range of ZIKV doses for vaginal inoculation of RM. Thus, two animals were vaginally inoculated weekly with104 PFU (6x106 vRNA copies), two animals were inoculated with 105 PFU (6 x 107 vRNA copies) and two animals were inoculated with 106 PFU (6 x 108 vRNA copies). There was a 7-day interval between each vaginal inoculation (Fig 1). Two RM became infected (plasma ZIKV RNA+) after 1 vaginal inoculation with ZIKV (Fig 1). One (37812) of these 2 RM was exposed to a moderate virus dose (105 PFU/6 x 107 vRNA copies) in the luteal phase of the cycle (approx. cycle day 21) (Fig 2B) and the other (37072) to a low dose (104 PFU/6 x 106 vRNA copies) of ZIKV in the peri-ovulatory phase of the cycle (approx. cycle day 15) (Fig 2A). A third RM (37828) became infected after 2 high dose (106 PFU/6x108 vRNA copies) vaginal ZIKV inoculations with transmission occurring after the 2nd inoculation in peri-ovulatory phase of the cycle (approx. cycle day 14) (Fig 2C), and another RM (40125) after 5 vaginal inoculations with a high dose (106 PFU/6 x 108 vRNA copies) of ZIKV with transmission occurring after the 2nd inoculation in follicular phase of the cycle (approx. cycle day 7) (Fig 2D). Finally, after 8 weekly vaginal ZIKV inoculations, one low dose RM (36813) and one moderate dose RM (39933) remained uninfected. Both of these animals were treated with Depoprovera and 30 days later they were re-inoculated vaginally with the same dose of ZIKV they were previously inoculated with 8 times without transmission. After Depoprovera treatment, both of these RM became infected after 1 vaginal ZIKV inoculation (Figs 1 and 3). Thus, RM were readily infected with ZIKV after vaginal inoculation with a concentration of ZIKV within the range that is found in human semen [4,33].
In all 4 Depoprovera-naive RM, plasma ZIKV RNA was first detected at 4 or 6 days post-inoculation (PI), reached peak levels at 6–10 days PI and was undetectable by 9–14 days PI. The mean duration of virema was 8.2 days (Fig 2A–2D). ZIKV RNA levels in CVL and urine were also determined.
In 3 of 4 Depoprovera-naive RM, a blip of vRNA was detected in CVL 24–48 hours after vaginal inoculation, and then vRNA became undetectable (Fig 2A, 2C and 2D). In 2 of these RM, vRNA reappeared in CVL before plasma vRNA was detectable (Fig 2C and 2D). In the fourth RM, high and sustained levels of ZIKV RNA were found in CVL beginning at 3 days PI, prior to detection of plasma vRNA (Fig 2B). Among all 4 Depoprovera-naive RM, CVL ZIKV RNA was detected at 2–6 days PI, peaked at 2–9 days PI and was undetectable by 12–21 days PI. The mean duration of ZIKV RNA shedding in CVL was 8.1 days (Fig 2A–2D).
In 2 of 4 RM, a blip of vRNA was detected in urine within 24–48 hours post-inoculation (PI) (Fig 2B and 2C), with vRNA reappearing in urine before plasma vRNA was detectable in 1 of these 2 RM (Fig 2C). In the other 2 RM, high and sustained levels of ZIKV RNA were found in urine beginning at 9–12 days PI, long after detection of plasma vRNA. Among all 4 RM, urine ZIKV RNA was detected by 1–11 days PI, peak levels occurred at 7–14 days PI and vRNA was undetectable in urine by 9–32 days PI, (mean duration of urine ZIKV RNA shedding: 6 days) (Fig 2A–2D).
The detailed virology of the 2 RM that resisted systemic infection, despite 8 vaginal ZIKV inoculations spanning 2 menstrual cycles, until they were treated with Depoprovera is shown in Fig 3. On day 57 PI, 8 days after the last ZIKV vaginal inoculation on Day 49 PI and before DepoProvera treatment, vRNA was detected in one urine sample, but not plasma or CVL, of RM 36813 (Fig 3B). However following Depoprovera treatment, vRNA was present in CVL 2 days after the vaginal ZIKV rechallenge on day 100, while plasma vRNA was detected 2 days later on day 102 (Fig 3B). In the other Depo-treated RM, (39933) ZIKV RNA was first detected in plasma, CVL and urine on day 102, 4 days after vaginal ZIKV inoculation (Fig 3A).
We used a Luminex-based bead array assay to assess changes in the levels on cytokines and chemokines in the plasma and CVL in the 4 DepoProvera naive RM (Fig 4). All 4 RM (37072,40125,37182,38728) had clear increases in the level of macrophage inhibitory factor (MIF) in plasma. In addition 3 of 4 RM had increased plasma levels of L-1RA (37072,40125,37182), and CCL5 (RANTES) (37072,40125,38728), and 2 of 4 RM (37072,40125) had increased plasma levels of CCL11 (Eotaxin), CXCL10 (IP-10) and CXC11 (I-TAC) (Fig 4). The plasma levels of these mediators both increased and decreased after infection, but the highest levels of an analyte were generally found in plasma samples collected the day after peak vRNA levels and the lowest levels of most analytes were found in plasma samples with low vRNA levels (Fig 4). The pattern of changes in MIF levels were unique in that they increased prior to, or just after, initial detection of plasma vRNA; were lowest at the peak in plasma vRNA levels; and, in 3 of 4 RM (37072,40125,38728), increased to their highest level days after the peak in plasma vRNA (Fig 4).
The effect of vaginal ZIKV transmission on cytokine and chemokine levels in CVL was more dramatic and was detectable prior to changes in plasma levels of these analytes (Fig 4). All 4 RM had clear changes in the levels of IL-1b, IL-1RA, IL-6 and macrophage inhibitory factor (MIF) in CVL (Fig 4). In addition, 3 of 4 RM (40125,37182,38728) had increased levels of CXCL8 (IL-8), and 2 of 4 RM (37072,40125) had increased levels of CCL5 (RANTES) and CCL11. The levels of these mediators in CVL both increased and decreased after infection, but the highest levels of most analytes were generally found in CVL samples with high vRNA levels and the lowest levels of most analytes were found in CVL samples with low vRNA levels (Fig 4). In 3 animals (37072,37182,38728), IL-1Ra levels increased on the first day vRNA was detected in CVL and remained elevated until vRNA levels dropped (Fig 4). Of note, the levels of IL-1b and IL-1Ra were 10–100 fold higher in CVL than plasma (Fig 4), despite the dilution that occurs when CVL samples are collected. CVL sample collection began on Day 0, just prior to the first ZIKV inoculation, and thus for the animals (37072,38728) that became infected after 1 ZIKV inoculation there was a single pre-infection sample, for the animal (38728) infected after 2 inoculations there were 1 week of pre-infection samples and for the animal (40125) infected after 5 inoculations 4 weeks of pre-infection samples are available. The cytokine levels in the pre-infection CVL of the latter 2 animals were relatively stable, except in the CVL samples collected during menses (day -18 to -14) from 40125, in which many of the analytes were elevated (Fig 4).
We used a commercial ELISA assay to assess the levels of ZIKV-specific antibodies in plasma and CVL of the ZIKV inoculated animals. For all 6 RM infected with ZIKV after vaginal inoculation (Fig 1), paired plasma and CVL samples collected weekly from the day of first ZIKV inoculation to necropsy were tested. Of the 4 RM that became infected without DepoProvera treatment, ZIKV-specific antibodies were detected in plasma of one (37072) 7 days after vaginal ZIKV transmission and 14 days after vaginal ZIKV transmission in the other 3 RM (37812, 38728, 40125) (Fig 2). ZIKV-specific antibodies were never detected in the CVL samples of any of these 4 animals (Fig 2s). The 2 RM that remained ZIKV negative after 8 vaginal inoculations but then became infected after DepoProvera treatment and 1 additional vaginal inoculation, remained anti-ZIKV plasma antibody negative from the day of the first ZIKV inoculation until necropsy in the acute stage of infection (Fig 3).
To better understand the tissue tropism of ZIKV, we determined vRNA levels in tissues of all 6 RM infected with ZIKV by vaginal inoculation (Fig 5). The 2 RM treated with Depoprovera prior to infection (Fig 1) were necropsied at 4 and 8 days after vaginal ZIKV inoculation, when vRNA was detectable in plasma and CVL. At 4 days PI (39933), ZIKV RNA was present at low to moderate levels in the urinary tract, FRT and draining lymph nodes. vRNA was also detected in distal lymph nodes and spleen (Fig 5). At 8 days PI (36813), vRNA levels were 100–1000 fold higher in all tissues, with the highest levels in salivary glands and lymphoid tissues. ZIKV RNA was also detected in the central nervous system (CNS) at this early stage of infection (Fig 5). In addition, infectious ZIKV was isolated from 1 of 6 vRNA+ lymphoid tissues at 4 days PI (39933); while at 8 days PI (36813) ZIKV was isolated from 6 of 6 lymphoid tissues tested (Table 1).
The remaining 4 RM were necropsied between 30 and 35 days PI, about 2 weeks after vRNA was last detectable in plasma. At this stage, the RM were plasma ZIKV RNA negative and anti-ZIKV IgG positive. However, ZIKV RNA was detected at low to moderate levels in all lymphoid tissues tested from all 4 RM. In addition, low level ZIKV RNA was detected in the CNS (temporal lobe of brain) of one RM and the FRT (uterus) of a second RM (Fig 5). Zika RNA is also detected in tissues, including the brain and male and female reproductive tissues, during early and late stages of infection after SQ ZIKV inoculation of RM [34–36]. However, we were not able to recover infectious ZIKV from tissues of any of these 4 animals (Table 1). Thus, the significance of the ZIKV RNA that persists in tissues of RM long after it is cleared from plasma is unclear.
Given the severe disease ZIKV can cause in a developing fetus [37], the risk of transmission to women during pregnancy is of particular concern. Despite documented cases of ZIKV sexual transmission [1–13], the frequency and efficiency of sexual ZIKV transmission is unclear. Two modeling studies of ZIKV transmission dynamics in the recent outbreak in the Americas estimated that sexual transmission contributed between 3–45% to the overall basic reproduction number (R0) of ZIKV in a population [38] [39]. Obviously, this very wide range indicates that there is still considerable uncertainty about the significance of sexual transmission ZIKV in propagating and maintaining the virus in human populations [38] [39]. To better understand the potential for sexual transmission of ZIKV, a NHP model of vaginal transmission is needed. Macaques were experimentally infected with mouse-brain passaged ZIKV in the 1950s, however, until recently there were no published reports describing the biology of ZIKV infection in nonhuman primates. Since early 2016, animal models of human ZIKV have been developed using Type-1 IFN-antibody treated mice, Type-1 IFNR knockout mice [40–46] and RM [35,36,47–49]. To date, the reported non-human primate (NHP) studies have used intravenous (IV) or SQ routes of ZIKV inoculation to infect RM [35,36,47,48]. The data reported here demonstrate that ZIKV can be readily transmitted to mature cycling female RM by vaginal inoculation.
Perhaps, the most striking finding in this study is that the kinetics of virus replication and dissemination in RM after intravaginal ZIKV inoculation are markedly different than after SQ virus inoculation [34,36,49]. After SQ inoculation of RM with Asian lineage ZIKV, vRNA is detected in blood plasma as early as 1 d after infection and subsequently in both the urine and saliva [36,49]. The appearance of vRNA in urine and saliva is delayed and blunted when compared to plasma and ZIKV RNA was detected only infrequently in CVL of RM after SQ inoculation [36,49]. As in SQ inoculated RM, ZIKV shedding from the FRT is rare in ZIKV-infected women [50] the majority of whom were presumably infected by mosquito bite. In SQ inoculated RM, viral RNA is cleared from plasma and urine by day 10, but remains detectable in saliva and semen for more than 3 weeks [36]. In marked contrast, plasma vRNA is delayed by several days, and virus shedding from the FRT occurred, in all RM inoculated with ZIKV intravaginally (Fig 2). Of note, ZIKV is found in the FRT of a subset of infected women [51–53], and it is tempting to speculate that in these cases the virus was sexually acquired.
In addition to the delay in plasma vRNA in ZIKV vaginally inoculated RM compared to SQ infected RM, virus dissemination to tissues was slower and stepwise in the vaginally inoculated animals. Four days after vaginal inoculation, ZIKV RNA was present at low to moderate levels in the urinary tract, FRT, draining lymph nodes distal lymph nodes, spleen. However, at 8 days PI, vRNA levels were 100–1000 fold higher in all tissues, with the highest levels in salivary glands and lymphoid tissues indicating that the virus was still disseminating more than 1 week after infection. At 30 and 35 days PI, the vaginally infected RM were plasma ZIKV RNA negative but had low to moderate ZIKV RNA levels in all lymphoid tissues tested. In addition, low level ZIKV RNA was detected in the uterus of one of these 4 RM (Fig 5). Similarly, 7 days after SQ ZIKV inoculation high levels of ZIKV RNA were found in numerous tissues, including the brain and reproductive tract; and ZIKV RNA persisted through day 35 PI in neuronal, lymphoid and joint/muscle tissues [34,36]. However, while infectious ZIKV was isolated from multiple tissues at day 7 PI, infectious virus was not found in tissues collected at 28 days PI [34]. Thus, although ZIKV RNA seems to persist in target tissues for a considerable period after it is cleared from the blood, it remains to be seen if this persistent RNA contributes to pathogenesis or can serve as a reservoir for infectious virus.
In the RM infected by vaginal ZIKV inoculation, the levels ZIKV RNA in CVL was similar to plasma vRNA levels. Given the 10–100 fold dilution of cervicovaginal secretions that occurs during the CVL collection process, vRNA levels in CVL were at least equal to, and often higher than, plasma vRNA levels (Fig 2). Thus the FRT is able to support a high level of ZIKV replication. The timing of ZIKV shedding in CVL also demonstrated that virus replication in the FRT was independent of systemic replication. Often ZIKV RNA was detected in CVL before it appeared in plasma and ZIKV RNA could also be found in CVL after virus had been cleared from plasma. This suggests that the virus being shed in CVL is from local replication in the FRT that is independent of virus replication in other tissues. The presence of ZIKV in the FRT after its disappearance from blood and urine samples has also been documented in women [51,52], which suggests that the ZIKV preferentially replicates in the FRT of RM and women that acquire the infection through sex or vaginal inoculation.
There was substantial variability between the individual RM in susceptibility to infection after vaginal ZIKV inoculation in this study. It has been reported that the stage of the menstrual cycle at vaginal inoculation effects susceptibility to infection with SHIV and SIV in RM [54,55]Sodora. In these reports, susceptibility to viral infection was highest in menses and the luteal phase of the cycle [56]. In the current study, of the 4 ZIKV+ animals infected without Depo-Provera treatment, 37072 was infected in peri-ovulatory phase of the cycle (approx. cycle day 15); 38728 was infected in peri-ovulatory phase of the cycle (approx. cycle day 14); 37812 was infected in early luteal phase of the cycle (approx. cycle day 21); and 40125 was infected in follicular phase of the cycle (approx. cycle day 7) (Fig 2). Thus, there is no evidence that the stage of the menstrual cycle at exposure explains the variability vaginal ZIKV transmission in these 4 monkeys, however this initial observation needs to be confirmed in larger studies.
Depo-Provera, a brand of the injectable hormonal contraceptive depot-medroxyprogesterone acetate (DMPA), is the most widely used injectable contraceptive in the world. We chose to test the effects of Depoprovera on vaginal ZIKV transmission because DMPA treatment enhances infectivity of viruses in various rodent and nonhuman primate models of female genital tract infection [57–61]. In fact, progesterone treatment is needed to infect mice with ZIKV by vaginal inoculation [46]. In addition, some observational studies identified DMPA as a significant risk factor for acquisition of HIV and other sexually transmitted infections (STI) in women, while other studies failed to detect this association [62–65]. Our observation that, after Depoprovera treatment, both of the RM that initially resisted vaginal ZIKV transmission became infected with one vaginal ZIKV inoculation is consistent with the conclusion that Depoprovera enhanced susceptibility to vaginal ZIKV transmission. Caution is warranted in interpreting our study however as only 2 animals were treated with Depo-Provera in the study.
Several mechanisms have been proposed to explain enhanced STI acquisition with Depo-Provera including mucosal epithelium thinning, enhanced tissue inflammation, suppressed cell-mediated immune responses, and altered vaginal microbiota. However, none of these putative biological mechanisms are experimentally proven [66,67]. It was recently reported that Depoprovera use in women is associated with increased hemoglobin, immune activation markers (HBD, HBB, IL36G), and decreased epithelial repair proteins (TFF3, F11R) in reproductive tract secretions [68]. Further, in mice Depo-Provera reduced expression of the desmosomal cadherin desmoglein-1α in the genital epithelium, enhanced inflammatory cells numbers in genital tissue by increasing mucosal epithelial permeability, and increased susceptibility to HSV-2 infection [69]. The results of both of these recent studies suggest that Depo-provera mediated increases in mucosal permeability facilitate endogenous vaginal microbiota invasion and tissue inflammation by breaking down the epithelial barrier. Thus the most likely explanation for enhanced vaginal ZIKV virus transmission in the Depo-Provera treated animals is that increased permeability of the vaginal mucosa allowed the virus inoculum to access more target cells in the lamina propria.
Although our ZIKV inoculum was delivered to monkeys as cell-free virions suspended in tissue culture fluid, women are exposed to ZIKV virions in semen, which may affect virus transmission. Seminal plasma (SP) has a basic pH that neutralizes the acidic pH of the vagina, thus seminal plasma may limit the inactivation of ZIKV deposited into the vagina. In fact, it has been shown that SP boosts SIV and HIV-1 infection in vitro and semen amyloid proteins contribute to this activity [70–74]. However, the significance of the in-vitro observations is unclear, as the addition of semen, SP or semen amyloid proteins does not dramatically enhance vaginal SIV transmission [75]. However, it has been reported that SP marginally increases vaginal SIV transmission if low-dose viral inoculums are used [76,77]. In addition, human and macaque seminal plasma are complex biologic fluids that vary substantially in chemical composition between individuals, and between individual ejaculates making it impossible to replicate experiments without using an aliquot of the semen sample used in the original experiment. Due to the limited volume, it is not possible to use the same human or macaque seminal plasma material for more than a few experimental vaginal inoculations. These technical factors make it impractical to use seminal plasma in animal experiments modeling vaginal virus transmission if reproducible results are desired. To insure the reproducibility of the results in the studies reported here, we did not include seminal plasma in the inoculum.
RM infected by SQ inoculation with ZIKV during the first trimester of pregnancy have persistent plasma vRNA, leading to the hypothesis that the fetus or placenta may be the source of persistent virus replication in the immune suppressed pregnant female [49]. This conclusion is consistent with a report that the placental/fetal tissues from 24 of 44 women suspected of being infected with Zika virus during pregnancy were positive for ZIKV RNA by RT-PCR. [78]. However the results reported here, and previous results in RM [34] and women [51,52], demonstrate that ZIKV RNA persists in the FRT and lymphoid tissues in non-pregnant RM and these tissues are another possible source of persistent plasma vRNA in pregnant animals.
Although we detected anti-ZIKV IgG antibodies in plasma of all 4 animals infected for more than 10 days with ZIKV. We did not detect anti-ZIKV IgG antibodies in CVL of any animals. This is unexpected as antiviral antibodies are routinely found in CVL of RM and women [79–82]. It is possible that this result is due to a technical issue with the commercial ELISA kit we used. We are in the process of developing ELISA assays to measure anti-ZIKV IgG subclass and IgA antibody responses and these assays will clarify whether anti ZIKV antibody responses that were undetectable by the commercial assay are present in the CVL. Our inability to detect a plasma antibody response in the 2 animals inoculated 8 times with ZIKV is consistent the lack plasma vRNA in the animals and confirms that they remained uninfected despite the repeated ZIKV exposures. Apparently, in the absence of infection, the amount of ZIKV antigen in the inoculum is insufficient to elicit a systemic antibody response when placed on the mucosal surfaces of the FRT.
The systemic cytokine response is minimal after SQ ZIKV inoculation of RM [34], and it was suggested that the low levels of cytokine activation in vivo may be the result ZIKV inhibiting the innate immune pathways that direct synthesis and secretion of pro-inflammatory cytokines [34]. However, we found evidence that vaginal ZIKV transmission and subsequent systemic infection results in an acute inflammatory response characterized by increases in pro-inflammatory cytokines and chemokines in CVL and, to a lesser degree, plasma. Further, after vaginal ZIKV transmission, the inflammatory response in the FRT corresponded temporally to periods of local ZIKV replication. Thus, peak levels of ZIKV shedding/replication in the FRT were often associated with increased levels of pro-inflammatory cytokines (IL-1b, IL-6), anti-inflammatory mediators (IL-1RA) and a subset of chemokines in CVL. These changes are consistent with an acute antiviral inflammatory response to local ZIKV replication and viral mediated tissue damage in the FRT. However, the pattern and levels of the inflammatory mediators were very different in the blood and CVL. MIF and IL1Ra were elevated in both plasma and CVL of 3 of 4 RM. While IL-6 were elevated in CVL, but not in plasma, (Fig 4) all animals and CXCL8 was elevated in CVL but not plasma of 3 of 4 animals. In addition, given the 10-fold dilution of secretions that occurs during the collection of CVL, the concentration of all these inflammatory mediators was generally higher in CVL than plasma. Thus, after vaginal ZIKV transmission, there was an obvious local and systemic inflammatory response that was delayed and enhanced compared to that reported in SQ inoculated RM [34]. This finding suggests that the pathogenesis of ZIKV disease can vary with the route of transmission. Taken together, the distinct timing and nature of the inflammatory response in the FRT compared to blood and the unique pattern of virology in the FRT, is consistent with the conclusion that ZIKV replication in the FRT is independent of replication in the systemic compartment.
The pattern of inflammation in the FRT and systemic compartments also provides considerable insight into ZIKV pathogenesis. MIF, the only cytokine that was elevated in both plasma and CVL samples of all 4 animals. DENV infection induces MIF production and secretion and secreted MIF can enhance DENV replication and increase vascular leakage through autophagy [83]. Thus MIF may contribute to inflammation and hemostatic abnormality during DENV infection [84] and there is a correlation between MIF serum levels and disease severity in dengue patients [85]. The high concentrations of IL-1b and CXCL8 in the CVL after ZIKV infection suggest that enhanced neutrophil recruitment is a major response to ZIKV replication in the FRT [86–88]. Recruitment of neutrophils i requires the upregulation and release of IL-1β [89,90] and IL-1 also markedly prolongs the lifespan and stimulates the effector function of neutrophils and macrophages [91]. IL1Ra was elevated in the plasma of 3 of 4 RM and the CVL of all RM. Of note, the levels of IL-1Ra were 10–100 fold higher in CVL than plasma (Fig 4). IL-1Ra competes with IL-1 for binding to the IL-1 receptor, blocking IL-1–induced pro-inflammatory signaling, and thus, may affect viral pathogenicity. Elevated levels of IL-1Ra have been described in humans with a number of viral infectious diseases [92–94], but the role of IL-1Ra in viral pathogenesis is unclear. Changes in the levels of IL-6 were found in the CVL of all 4 animals tested. Although, IL-6 is considered a marker of inflammation L-6 levels do not necessarily correlate with the levels of other inflammatory cytokines and IL-6 directly affects the adaptive antiviral immune response. IL-6 affects differentiation of CD4 T cells [95] and can also modulate aspects of the innate immune response to viral infection [96–98].
The findings that ZIKV shedding in CVL is not related to plasma vRNA levels and that a local inflammatory response develops in the FRT that is distinct from the systemic response is consistent with the conclusion that ZIKV replicates, and persists, in the FRT independent of the systemic ZIKV infection. This conclusion is also supported by observation that after vaginal ZIKV inoculation of IFNR+/+ mice, ZIKV replicates in the FRT but not in systemic tissues [43]. Thus, data from both the NHP and mice models of vaginal ZIKV transmission support the conclusion that, after vaginal ZIKV transmission the virus preferentially replicates in the FRT independent of replication levels in other tissues. The unusual tropism of ZIKV for the FRT raises the possibility of additional unexpected effects of vaginal ZIKV transmission, including the potential for enhanced fetal infection and pathology. In addition, it remains to be shown that a vaccine that protects animal models from mosquito transmitted ZIKV can protect against vaginal ZIKV transmission.
|
10.1371/journal.ppat.1000714 | PPARγ Controls Dectin-1 Expression Required for Host Antifungal Defense against Candida albicans | We recently showed that IL-13 or peroxisome proliferator activated receptor γ (PPARγ) ligands attenuate Candida albicans colonization of the gastrointestinal tract. Here, using a macrophage-specific Dectin-1 deficient mice model, we demonstrate that Dectin-1 is essential to control fungal gastrointestinal infection by PPARγ ligands. We also show that the phagocytosis of yeast and the release of reactive oxygen intermediates in response to Candida albicans challenge are impaired in macrophages from Dectin-1 deficient mice treated with PPARγ ligands or IL-13. Although the Mannose Receptor is not sufficient to trigger antifungal functions during the alternative activation of macrophages, our data establish the involvement of the Mannose Receptor in the initial recognition of non-opsonized Candida albicans by macrophages. We also demonstrate for the first time that the modulation of Dectin-1 expression by IL-13 involves the PPARγ signaling pathway. These findings are consistent with a crucial role for PPARγ in the alternative activation of macrophages by Th2 cytokines. Altogether these data suggest that PPARγ ligands may be of therapeutic value in esophageal and gastrointestinal candidiasis in patients severely immunocompromised or with metabolic diseases in whom the prevalence of candidiasis is considerable.
| Since the early 1980s, Candida albicans has emerged as major cause of human disease, especially among immunocompromised individuals and those with metabolic dysfunction. The main host defense mechanisms against this yeast are engulfment and the production of reactive oxygen molecules by macrophages through Dectin-1 and the Mannose Receptor, two macrophage receptors for Candida albicans cell wall sugars. However, the contribution of these two receptors remains unclear. In our animal experiments, the lack of Dectin-1 in macrophages renders the animals more susceptible to gastrointestinal infection with Candida albicans, demonstrating the essential role of Dectin-1 in antifungal defense. In addition, our experiments established that the interaction between Dectin-1 and Mannose Receptor is important to orchestrate the host antifungal defense. Thus, Candida albicans clearance would be improved by Dectin-1 and Mannose Receptor up-regulation. Interestingly, we had established that the expression of these two receptors was increased by IL-13 through the activation of the nuclear receptor PPARγ, suggesting that PPARγ could be a therapeutic target to eliminate fungal infection. This paper, which highlights a new area of application of PPARγ ligands in infectious diseases, hence heralds the emergence of a new therapeutic strategy against fungal infection in severely immunocompromised patients or those with metabolic diseases.
| Innate immunity is a conserved mechanism of host defense and is responsible for immediately recognizing microbial invasion through the engagement of pattern-recognition receptors (PRRs). These PRRs can recognize highly conserved microbial structures, known as pathogen-associated molecular patterns (PAMPs). The PRR ligands comprise carbohydrate structures, peptidoglycans or lipopolysaccharides. The best characterized family of PRRs is the Toll-like receptors (TLRs) originally supposed to mediate cellular signaling, but the membrane-associated C-type lectin receptors have since emerged as major receptors in functions related to pathogen binding, uptake, and killing. They also contribute to the initiation and the modulation of the immune response. The C-type lectins form a group of proteins with at least one lectin-like carbohydrate-recognition domain (CRD) in their extracellular carboxy-terminal domains [1]. The C-type lectin Dectin-1 is a major β-glucan receptor on the surface of macrophages, DCs, neutrophils and it is also expressed on certain lymphocytes [2]. This type II transmembrane receptor consists of a single CRD involved in the calcium-independent recognition of β-1, 3-glucans exposed on particles such as zymosan, or many fungal species, including Saccharomyces, Pneumocystis, Aspergillus and Candida [3]–[5].
C.albicans is the most common cause of opportunistic mycotic infections in severely immunocompromised hosts and during metabolic disease [6]. The cell wall of this yeast is almost exclusively composed of glycans, such as mannans and β-glucans [7]. Mannans are the major component of outer cell wall while β-(1,3)- and β-(1,6)-glucans are more prominent in the inner layer. However, there is some surface exposure of β-glucans, particularly in areas where yeast cells bud during mother–daughter cell separation [8],[9]. The recognition of the multilayered carbohydrate structures of the fungal cell wall depends on various PRRs, such as the Mannose Receptor (MR), and the β-glucan receptor Dectin-1 [9],[10]. The respective roles of these PPRs in the non-opsonic recognition of C. albicans by macrophages remain unclear. Several studies support the view that the MR plays a crucial role in non-opsonized C.albicans recognition and phagocytosis [9],[11],[12]. This receptor has also been shown to be associated with the production of proinflammatory cytokines and reactive oxygen species [9],[13]. Recently, the β-glucan receptor Dectin-1 was found to be the main non-opsonic receptor involved in fungal uptake [14]. In addition, Dectin-1-induced-signaling leads to the production of cytokines and non-opsonic phagocytosis of yeast by murine macrophages [15],[16]. Dectin-1 also mediates respiratory burst [17] and its involvement has been suggested in the activation and regulation of phospholipase A2 (PLA2) and cyclooxygenase-2 (COX-2) [18]. Dectin-1 signaling pathway activation depends on its cytoplasmic immunoreceptor tyrosine-based activation motif (ITAM) the phosphorylation of which by Src kinase leads to the recruitment of spleen tyrosine kinase Syk in macrophages [19]. Although the contribution of the MR and Dectin-1 in non-opsonized C.albicans recognition, phagocytosis and killing is established, the point of intervention of these receptors in these processes remains unclear. In addition, depending on the context of the macrophage activation, the expression profile of PRRs is different. Thus, the change in the PRRs expression has to be taken into account in studying the involvement of these receptors in antifungal functions.
In mice, the expression of Dectin-1 can be influenced by various cytokines, steroids and microbial stimuli. Interleukin-4 (IL-4) and IL-13, for example, which are associated with the alternative activation of macrophages (macrophages M2), markedly increase the expression of Dectin-1 at the cell surface, whereas LPS and dexamethasone repress Dectin-1 expression [20]. Nevertheless, the Dectin-1 regulation pathway remains unclear. Empirical data suggest that the increase in Dectin-1 expression by IL-4 involved the STAT signaling pathway [21]. Moreover, another study showed in genetic models of macrophage specific Peroxisome proliferator-activated receptor γ (PPARγ) or STAT-6 knockout mice, that the IL-4/IL-13/STAT-6/PPARγ axis is required for the maturation of alternatively activated macrophages [22]. Therefore, in this context, the signaling pathway involved in the modulation of Dectin-1 expression remains to be elucidated.
In this study we determined the respective roles of the MR and Dectin-1 in the control of fungal infection. Interestingly, we showed that in vitro and in vivo, Dectin-1 is essential both to trigger the phagocytosis of non-opsonized Candida albicans and the respiratory burst after yeast challenge and to control fungal gastrointestinal infection. These data also established that the MR alone is not sufficient to trigger antifungal functions during macrophage alternative activation, indicating a cooperative role for Dectin-1 and the MR in the induction of the host antifungal response against Candida albicans. Moreover, we showed for the first time the involvement of the PPARγ signaling pathway in the regulation of Dectin-1 expression by IL-13. This report highlights that PPARγ ligands could be of therapeutic benefit in the resolution of fungal infections in patients severely immunocompromised or with metabolic diseases in whom the prevalence of candidiasis is considerable.
To explore the role of Dectin-1 and the MR in the control of fungal infection by alternatively activated macrophages (M2), we studied the phagocytosis of non-opsonized C.albicans and the production of yeast-induced reactive oxygen species (ROS) by macrophages, in the presence or absence of soluble receptor blocking agents (laminarin and mannan). The phagocytosis of non-opsonized C.albicans was significantly increased by IL-13 treatment (Figure 1A). This enhancement of phagocytosis in M2 activation was reduced by mannan. Moreover, laminarin or the association of mannan and laminarin blocked yeast internalization (Figure 1A).
Since C.albicans stimulates phagocytosis, and since this function contributes to the triggering of ROS production, we examined the respiratory burst induced by non-opsonized C.albicans in IL-13 polarized macrophages. As in the phagocytosis experiment, ROS production was enhanced by IL-13, and clearly reduced by the addition of soluble mannan and/or laminarin (Figure 1B).
Consistent with the critical role of PPARγ activation in the maturation of alternatively activated macrophages, we explored these two antifungal functions in PPARγ ligand-primed macrophages. Interestingly, these functions were enhanced by rosiglitazone, a PPARγ specific ligand, and decreased by mannan and/or laminarin pretreatment, as observed during alternative activation by IL-13 (Figure 1C, 1D). Altogether, these data showed that the antifungal functions of macrophages promoted by IL-13 or a PPARγ ligand involved Dectin-1 and the MR.
To unequivocally determine the role of Dectin-1 in C.albicans elimination, we generated Dectin-1 receptor conditional knockout mice, in which Dectin-1 was selectively disrupted in phagocytic cells. First, we generated mice that carried conditional Dectin-1 alleles (Dectin-1L2/L2 mice). To generate spatially controlled mouse mutants for the Dectin-1 gene in the macrophages, mice carrying the floxed Dectin-1 L2 alleles were crossed with transgenic mice that expressed the Cre recombinase under the control of the mouse lysozyme M promoter. Quantitative real-time RT-PCR and flow cytometry confirmed that the mRNA and protein levels of Dectin-1 were abrogated in phagocyte cells (Figure 2A, 2B).
To explore the antigen phenotype of the Dectin-1 knockout macrophages, we studied the protein level of several receptors and markers on the macrophage surface. No significant changes in protein levels were detected between the control (Cre 0) and Dectin-1 knockout (Cre Tg) macrophages (Figure 2C) showing that in Dectin-1 knockout macrophages there was no compensatory increase in the expression of other PRRs.
It had previously been shown that arachidonic acid release induced by C.albicans was inhibited by preincubation with soluble glucan phosphate [18]. In this context, to characterize the functional capacity of Dectin-1 knockout macrophages, we investigated the ability of non-opsonized C.albicans to stimulate arachidonic acid release by the control (Cre 0) or Dectin-1 knockout (Cre Tg) peritoneal macrophages. C.albicans challenge reduced release of arachidonic acid by Cre Tg macrophages but not by Cre 0 macrophages (Figure 2D). Interestingly, Cre Tg macrophages were able to release arachidonic acid in response to phorbol myristate acetate (PMA) known to mediate arachidonic acid release via surface receptor independent pathway (Figure 2E). We confirmed that Cre Tg macrophages had no abnormalities in their lipid metabolism and that Dectin-1 was required for arachidonic acid metabolism in response to C.albicans.
We then investigated the ability of these Dectin-1 knockout macrophages to produce inflammatory cytokines. TNFα production by control (Cre 0) and Dectin-1 knockout (Cre Tg) macrophages was evaluated in response to heat-killed C.albicans. The Dectin-1 knockout macrophages failed to produce TNFα in response to heat-killed C.albicans or non-opsonized zymosan (ZNO) (Figure 2F). However, LPS stimulation demonstrated that Dectin-1 knockout macrophages do not have a generalized defect in TNFα production (Figure 2F).
All these data showed that this mutation did not affect the phenotype and the Dectin-1 independent functional capacities of the macrophages and hence provide an appropriate model to explore the involvement of Dectin-1 in C.albicans clearance during alternative activation.
We investigated the effect of the macrophage-specific Dectin-1 deletion both on non-opsonized C.albicans recognition and on the antifungal functions in a M2 activation context.
We first looked at the binding of non-opsonized C.albicans on resident peritoneal macrophages. Untreated Dectin-1-control (Cre 0) and Dectin-1 knockout (Cre Tg) macrophages bound non-opsonized yeast at 4°C (Figure 3A). Interestingly, M2 polarization of macrophages by IL-13 strongly enhanced the binding of non-opsonized C.albicans equally in Cre 0 and Cre Tg macrophages, showing that Dectin-1 is not required for the initial binding of the yeast to untreated as well as to alternatively activated macrophages. Then to explore the role of the MR in binding non-opsonized C.albicans, we pretreated Cre 0 and Cre Tg macrophages with soluble mannan (Figure 3A). The addition of mannan did influence slightly the binding of non-opsonized C.albicans to resident macrophages, suggesting that the MR and other PRRs were implicated in this stage of recognition. Interestingly, the binding by alternatively activated macrophages was strongly inhibited by mannan. We concluded that the MR is the main PRR for the initial binding in M2 activation.
To further investigate the respective roles of the MR and Dectin-1 in antifungal functions we studied the ability of Dectin-1 control (Cre 0) and Dectin-1 knockout (Cre Tg) macrophages to engulf C.albicans and to produce reactive oxygen species in the presence of mannan. Non-opsonized C.albicans phagocytosis was decreased in Cre Tg resident macrophages (Figure 3B). IL-13 increased yeast internalization in Cre 0 macrophages, but importantly, it failed to improve the antifungal response in Cre Tg macrophages. The addition of mannan slightly changed the uptake of non-opsonized C.albicans by resident and M2 polarized macrophages. Together these data showed that Dectin-1 was essential for triggering the phagocytosis of non-opsonized C.albicans both in resident and in alternatively activated macrophages.
Consistent with the phagocytosis results, the reactive oxygen species production induced by non-opsonized yeast was increased by IL-13 only in Cre 0 macrophages. Moreover, the addition of mannan slightly decreased ROS production. These results showed that Dectin-1 is the main receptor involved in this antifungal function (Figure 3C) and that Dectin-1 is very important in triggering the phagocytosis of yeast and the C.albicans-induced respiratory burst in alternatively activated macrophages.
To assess the involvement of PPARγ in the MR- and Dectin-1-dependent antifungal functions, we studied the binding and phagocytosis of C.albicans, and ROS production in the presence of rosiglitazone, a specific PPARγ-ligand. Rosiglitazone strongly increased the binding of non-opsonized C.albicans by Cre 0 and Cre Tg macrophages but failed to trigger phagocytosis and ROS production by Cre Tg macrophages (Figure 3D–F). Moreover, the addition of mannan only affected the binding. These results suggest the involvement of PPARγ in the mechanisms of response to non-opsonized C.albicans dependent on the MR and/or Dectin-1 receptors during M2 activation by IL-13.
In this context of M2 polarization, we explored the involvement of the PPARγ pathway in the regulation of Dectin-1 expression. We stimulated resident peritoneal macrophages with synthetic (Rosiglitazone, MCC555 and GW1929), natural (15ΔPGJ2) PPARγ-ligands or IL-13. FACS profiles showed that the Dectin-1 protein level at the surface of the macrophages was markedly up-regulated by IL-13 and PPARγ-specific ligands (Figure 4A, 4B). Equally, Dectin-1 mRNA expression was significantly enhanced by IL-13, Rosiglitazone and 15ΔPGJ2 (Figure 4C). To assess the involvement of PPARγ in the modulation of Dectin-1 expression, a PPARγ-deficient cell line RAW264.7 [23] was transiently transfected with the pCMV-mPPARγ expression vector. After IL-13 or Rosiglitazone treatment, the level of Dectin-1 protein expression was higher in cells transfected with the pCMV-mPPARγ than in control cells transfected with the pCMV-luc vector (Figure 4D). This result suggests that the induction of Dectin-1 by IL-13 is PPARγ-dependent.
To unequivocally prove the involvement of the PPARγ-pathway, we blocked PPARγ activation with two irreversible PPARγ antagonists (T007 and GW9662) or with a specific PPARγ siRNA. Macrophages treated with the antagonists failed to up-regulate Dectin-1 expression after exposure to IL-13 and PPARγ-specific ligands, as shown by FACS analysis or quantitative real-time RT-PCR (Figure 5A, 5B). Moreover, we demonstrated that silencing PPARγ expression in macrophages with siRNA also abolished the specific increase of Dectin-1 by IL-13 (Figure 5C).
All these data together prove that the nuclear receptor PPARγ is required for the induction of Dectin-1 by IL-13 in mouse peritoneal macrophages.
Because cPLA2 regulates the synthesis of 15ΔPGJ2, an endogenous PPARγ-ligand, we studied the effects of a specific cPLA2 inhibitor (MAFP) on Dectin-1 expression. We showed that MAFP inhibited macrophage Dectin-1 mRNA expression (Figure 6A). In line, the level of Dectin-1 protein was decreased in a dose-dependent manner by the treatment of macrophages with MAFP (Figure 6B).The addition of 15ΔPGJ2 restored the induction of Dectin-1 by IL-13 (Figure 6C). Thus, IL-13 regulates Dectin-1 expression by controlling the production of the PPARγ endogenous ligand through cPLA2 activation.
To assess precisely the involvement of Dectin-1 in the development of gastrointestinal candidiasis and in the antifungal effect of PPARγ ligands, we studied the susceptibility to Candida infection of macrophage-specific Dectin-1 deficient (Cre Tg) mice treated or not with rosiglitazone. In macrophage-specific Dectin-1 deficient mice infected with 5.106 C.albicans cells, the yeast extensively colonized the stomach and cecum whereas in control mice the colonization was undetectable (>104) (Figure 7A). These results demonstrated that Dectin-1 plays an important role in host defense against gastrointestinal infection with C.albicans. Interestingly, treating mice with rosiglitazone did not improve the Candida clearance in the gastrointestinal tract, suggesting that rosiglitazone needs Dectin-1 to exert its antifungal effect. To ensure that the lack of the rosiglitazone effect was due to the absence of Dectin-1, we infected control and Dectin-1 deficient mice orally with a larger quantity of yeast (5.107 C.albicans) and then we studied the effect of rosiglitazone on the outcome of this gastrointestinal infection (Figure 7B). In this gastrointestinal model of Candida infection, the yeast colonized the stomach and the cecum in both control and Dectin-1 deficient mice. In addition, the gastrointestinal colonization was considerably higher in the Dectin-1 deficient mice, confirming the major involvement of Dectin-1 in the host defense against C. albicans. As expected, rosiglitazone improved Candida clearance only in control mice, demonstrating that the lack of the rosiglitazone effect in Dectin-1 deficient mice was dependent on the lack of Dectin-1.
We then investigated the ability of macrophages from infected macrophage-specific Dectin-1 deficient mice to phagocytose yeast and to release reactive oxygen intermediates. In these two models of Candida gastrointestinal infection, we showed that Cre Tg macrophages failed to engulf C. albicans and to produce ROS (Figure 7C, data not shown). In addition, the treatment in vivo with rosiglitazone increased phagocytosis and the production of ROS in macrophages from Cre 0 mice, whereas this treatment did not increase these functions in Cre Tg macrophages. Nevertheless, in macrophages from Cre 0 or Cre Tg mice, the rosiglitazone treatment increased the expression of the MR (Figure 7D). Altogether these data demonstrated that in the absence of Dectin-1 the MR is not sufficient to trigger the antifungal functions and that the antifungal effect of rosiglitazone is Dectin-1 dependent.
Candida albicans causes significant and recurrent infections among immunocompromised hosts and during metabolic dysregulation. The interactions between hosts and fungal pathogens are mediated by mannans and β-glucans, the major cell wall components of C.albicans. The MR and Dectin-1 are the main pattern recognition receptors of the phagocytic system involved in C.albicans elimination.
Classically, the MR was described as mediating non-opsonized C.albicans recognition through the mannan chains on the outer yeast cell wall [9],[11],[12]. The MR is mainly involved in the uptake and phagocytosis of yeasts but it has also been shown to be involved in the production of TNFα, IL-1β, IL-6 and reactive oxygen species [9],[13] and to modulate the pro-inflammatory effects in collaboration with TLRs. However, other studies have shown the contribution of the β-glucan receptor Dectin-1 in response to non-opsonized C.albicans. Dectin-1 triggered the phagocytosis of the yeast cell wall particle zymosan by macrophages and dendritic cells [10]. It is now established that Dectin-1 mediates macrophage phagocytosis of C.albicans yeast [24] but not hyphae [8]. In addition, Dectin-1 also signals the release of reactive oxygen species, TNFα, IL-2, IL-6, IL-10 and IL-23 [3],[15],[25],[26]. Despite the fact that these two receptors are involved in common functions, a better understanding of the role of these receptors in the interaction between fungi and macrophages and in the antifungal functions of innate immune cells is necessary.
This study provides new data illustrating the relevant link between Dectin-1, the MR and the antifungal response during alternative activation of macrophages. We have demonstrated that depending on the context of macrophage activation, the receptors involved in yeast initial binding are different. Interestingly, we showed that the MR is the main PRR for the initial binding only in a M2 activation context in which the macrophages strongly expressed the MR at their surface. In contrast, the results of binding of non-opsonized C.albicans by resident macrophages showed that the MR is not involved in the recognition of non-opsonized C.albicans. This finding is consistent with the study which showed that resident macrophages do not express the MR [2]. In addition, we showed that Dectin-1 is not involved in the initial binding of non-opsonized C.albicans in resident macrophages. Moreover, in M2 activated macrophages, Dectin-1 is also not implicated in recognition of non-opsonized C.albicans. This finding is in line with a study on dendritic cells which showed that the addition of anti-MR and anti-DC-SIGN blocking agents inhibited the binding of non-opsonized C.albicans whereas the addition of an anti-Dectin-1 blocking agent did not change this recognition [27]. However, after 60 min of interaction between non-opsonized C.albicans and macrophages in vitro, we showed that Dectin-1 is sufficient to trigger the phagocytosis of non-opsonized C.albicans and respiratory burst after challenge with the yeast. Indeed, our results obtained with laminarin are consistent with our data using Dectin-1 knockout macrophages that confirm that the impairment of this receptor strongly decreased the antifungal response. Heinsbroek and coworkers have reported that the phagocytosis of unopsonized C.albicans by thioglycollate-ellicited macrophages of Dectin-1 deficient mice was reduced by 80%, showing that Dectin-1 is the main PRR for the initial phagocytosis by thioglycollate-ellicited macrophages [14]. These authors also have showed the lack of effects on MR in phagocytosis of Candida by thioglycollate-ellicited macrophages in which the MR is mainly intracellular in location. These data are also consistent with studies which showed that silencing Dectin-1 expression in macrophages with micro-RNA abolishes the zymosan-induced ROS production [25] and that the Dectin-1 engagement was sufficient to trigger phagocytosis and ROS production stimulated by zymosan [26] or by C.albicans [8]. In our study we demonstrated that the initial binding of the yeast through the MR does not seem to be directly involved in Dectin-1-dependant C.albicans uptake and respiratory burst. Altogether these results strongly suggest that another receptor could be implicated in the initial step of the Dectin-1-dependant phagocytosis of non-opsonized C.albicans and ROS production. This hypothesis is in line with a recent study which showed that complement receptor 3 accumulates at the site of particle binding and hence suggests it has a role during fungal recognition [14].
The major contribution of Dectin-1 in C.albicans internalization and ROS production in vitro supports our in vivo study which showed that Dectin-1 knockout mice were more susceptible to gastrointestinal candidiasis. This increased Candida colonization of the stomach and cecum in macrophage-specific Dectin-1 deficient mice correlated with the decrease in the effective functions of their macrophages ex vivo, as observed in vitro. These data demonstrated that Dectin-1 is required for the host defense to GI infection with C.albicans and support the role of Dectin-1 in the in vivo control of C.albicans infection [28]. We recently showed that i.p. treatment of immunocompetent and immunodeficient (RAG-2−/−) mice with natural and synthetic PPARγ-specific ligands or with IL-13 decreased C. albicans colonization of GI tract 8 days following oral infection with the yeast [13]. Similarly, we demonstrated here that rosiglitazone, a specific PPARγ ligand, improved GI fungal clearance only in control mice and this amelioration was correlated with an increased in antifungal functions of their macrophages ex vivo (Candida phagocytosis and ROS production). Nevertheless, the treatment of macrophage-specific Dectin-1 deficient mice with rosiglitazone did not enhance the Candida elimination in GI tract while the rosiglitazone treatment increased the expression of MR. Altogether these data established that the MR alone is not sufficient to trigger the antifungal functions and the antifungal action of rosiglitazone is dependent on Dectin-1. We also show that the phagocytosis of yeast and the release of reactive oxygen intermediates in response to Candida albicans challenge are impaired in macrophages from Dectin-1 deficient mice treated with rosiglitazone. This in vivo study demonstrates that Dectin-1 is essential both to trigger the phagocytosis of non-opsonised C.albicans and the respiratory burst after yeast challenge, and to control fungal GI infection. In parallel, the involvement of the MR in the initial binding during M2 activation demonstrates that the MR and Dectin-1 are essential for an optimal antifungal host defence. Altogether these data suggest a cooperative role for these two receptors in the induction of the immune response against Candida by rosiglitazone. This cooperation between the MR, Dectin-1 and TLR2 was also demonstrated in the pathway involved in IL-1β production by C.albicans [29].
The involvement of Dectin-1 in improving the resolution of candidiasis by PPARγ ligands is in line with our results which showed for the first time that the increase in Dectin-1 cell surface expression by IL-13 was mediated by the PPARγ signaling pathway. The implication of PPARγ in the transcriptional regulation of Dectin-1 is also confirmed by an in silico analysis of the Dectin-1 promoter using Genomatix software. One putative PPARγ responsive element was found in the reverse strand of this promoter. These results highlight that PPARγ is required for the maturation of alternatively activated macrophages. Indeed, we have previously shown that the PPARγ pathway was required in vitro and in vivo for the induction of the expression of the M2 marker MR (CD206) and CD36 expression during alternative activation of monocytes/macrophages by IL-13 [13],[30],[31]. These results are consistent with the studies of Odegaard and colleagues who showed that the expression of genes preferentially expressed in alternatively activated macrophages such as Mrc1 (gene of MR CD206) and Clec7a (gene of Dectin-1) was decreased by 70–80% in the white adipose tissue of macrophage-specific PPARγ knockout mice [22]. Moreover, a recent study showed that PPARγ activation primed human monocytes into an enhanced M2 phenotype [32]. These authors also reported that thiazolidinediones treatment significantly increased the expression of the M2 marker MR (CD206) in PBMC isolated from patients. In our study, we also show that both 15-ΔPGJ2 and rosiglitazone up-regulated Dectin-1 expression through PPARγ. Indeed, in the absence of PPARγ in the murine macrophage cell line RAW 264.7, IL-13 or PPARγ agonists do not induce the increase of Dectin-1 expression, and the effect of the PPARγ agonists or of IL-13 on this expression is restored by the pCMV-PPARγ transfection in these cells. These findings join a paradigm initiated by Huang and coworkers for the regulation of nuclear receptor function by Th2-type cytokines in an alternative pathway of macrophage activation [33].
In this manuscript, we determined the signaling pathway triggered by IL-13 resulting in the increase of Dectin-1 expression. The use of MAFP, a cPLA2 inhibitor, blocked the Dectin-1 surface induction by IL-13 and this Dectin-1 over-expression is restored by the addition of 15-ΔPGJ2. Thus, we showed that IL-13 can positively regulate Dectin-1 expression partly by controlling the production of PPARγ endogenous ligands through cPLA2 activation. These data are supported by the work of Huang and coworkers who showed that IL-4 leads to the production of PPARγ endogenous ligands and by our confocal microscopy studies illustrating that IL-13 generates 15-ΔPGJ2 production and this nuclear localization in human monocytes and murine macrophages [30],[31],[33].
In summary, we have established that Dectin-1 is essential both to trigger the phagocytosis of non-opsonized C.albicans and respiratory burst after yeast challenge during alternative macrophage activation and to control fungal GI infection. We have also demonstrated the major contribution of the MR for the initial recognition of non-opsonized C.albicans. These findings suggest that the cooperation between Dectin-1 and the MR is necessary to orchestrate the antifungal response. Moreover, these results underline the importance of the IL-13/PPARγ/C-type lectin receptors axis for the antifungal response in macrophages and in the decrease of colonization of the gastrointestinal tract by C. albicans. Indeed, PPARγ ligand strongly enhances the expression of C-type lectin receptors at the surface of macrophages and hence promotes antifungal host defense. These data suggest new therapeutic strategies using PPARγ ligands against fungal infections in immunocompromised hosts and during metabolic diseases, because they increase the innate immune response by enhancing the expression of both the MR and Dectin-1 that are heavily involved in the recognition and elimination of non-opsonized C.albicans.
This study was carried out in accordance with Approval No. A3155503 and all procedures for animal care and maintenance conformed with the French and European Regulations (Law 87–848 dated 19/10/1987 modified by Decree 2001-464 and Decree 2001-131 relative to European Convention, EEC Directive 86/609 dated 24/11/1986).
The strain of C. albicans used throughout these experiments was isolated from a blood culture of a patient in the Toulouse-Rangueil University Hospital. The isolate was identified as Candida albicans based on common laboratory criteria and cultured on Sabouraud dextrose agar (SDA) plates containing gentamicin and chloramphenicol. Candida albicans was maintained by transfers on SDA plates. Growth from an 18- to 24-h SDA culture of C. albicans was suspended in sterile saline.
Fluorescent C.albicans was prepared by adding C.albicans to fluoroscein isothiocyanate (FITC; Sigma, France) dissolved in sodium carbonate buffer (pH 9.5) at room temperature for 3 h and washed by centrifugation three times in sodium carbonate buffer before storage in aliquots of water at 4°C. The viability of FITC-yeasts was not altered by the protocol of FITC-labeling.
The culture medium was Dulbecco's modified Eagle's medium (DMEM, Gibco Invitrogen Corporation, France) supplemented with glutamine (Gibco Invitrogen Corporation) penicillin, streptomycin (Gibco Invitrogen Corporation), and 10% heat-inactivated fetal calf serum (FCS).
Laminarin (soluble β-glucan from Laminaria digitata, Sigma) and mannan (from S. cerevisiae, Sigma) were prepared as 10 mg/ml stocks in Hepes-buffered saline solution (HBSS, Gibco Invitrogen Corporation, France), filter sterilized, and stored frozen until use. Solutions used during experiments were made at final concentration of 1.25 mg/mL in DMEM-based culture medium.
For the analysis of binding, phagocytosis of C.albicans and ROS production, cultured-macrophages were incubated at 4°C for 20 min with mannan and/or laminarin solution. The medium was removed by washing with cold DMEM until the experiment.
To generate Dectin-1 floxed (Dectin-1L2/L2) mice, genomic DNA covering the Dectin-1 locus was amplified from the 129Sv strain using high fidelity PCR. The resulting DNA fragments were assembled into the targeting vector that after linearization by NotI was electroporated into 129Sv ES cells. G418-resistant colonies were selected and analyzed for homologous recombination by PCR and Southern blot hybridization. Positive clones were verified by Southern blot hybridization. Therefore genomic DNA was prepared from ES cells, digested with XbaI or SacI, electrophoresed and transferred to a positively charged nylon transfer membrane (Amersham Biosciences, Saclay, France). A 0.5 kb DNA fragment (NotI–NheI) located between exons 6 and 7 (3′ probe) and a 0.5 kb DNA fragment (NotI–SacII) placed between exons 2 and 3 (5′ probe) were used as probes. The karyotype was verified and several correctly targeted ES cell clones were injected into blastocysts from C57BL/6J mice. These blastocysts were transferred into pseudopregnant females, resulting in chimeric offspring that were mated with female C57BL/6J mice that express the Flp recombinase under the control of the ubiquitous CMV promoter. The offspring that transmitted the mutated allele, in which the selection marker was excised and that had lost the Flp transgene (Dectin-1+/L2 mice), were then selected and used for systematic backcrossing with C57BL/6J mice to generate congenic Dectin-1 floxed mouse lines. A PCR genotyping strategy was subsequently used to identify Dectin-1+/+, +/L2, and L2/L2 mice. To generate phagocyte-specific mutant (LysM-Dectin-1−/−) mice, Dectin-1L2/L2 mice were mated with LysM-Cre C57BL/6J mice in which the Cre recombinase was expressed under the control of the phagocyte-selective lysozyme promoter [34]. LysM-Cre/Dectin-1L2/+ mice, heterozygous for the floxed Dectin-1 allele, were selected and subsequently inter-crossed to generate pre-mutant LysM-Cre/Dectin-1L2/L2 mice. At least two more rounds of breeding were required to generate age- and sex-matched mice for the experimental cohorts.
Murine resident peritoneal cells were harvested from female wild-type or macrophage-specific Dectin-1 deficient mice. Briefly, cells were obtained by injection into the peritoneal cavity of sterile HBSS. The collected cells were centrifuged, and the cell pellet was suspended in culture medium as described in “Reagents” section. Cells were allowed to adhere over 2 h at 37°C with 5% CO2 atmosphere in 24- or 96-well culture plates. Nonadherent cells were removed by washing with phosphate-buffered saline (PBS) (Gibco Invitrogen Corporation), and the remaining adherent cells were stimulated as described below.
Peritoneal macrophages were stimulated by rosiglitazone (5 µM), 15d-PGJ2 (1 µM), MCC555 (5 µM), GW1929 (1 µM) (Cayman Chemical, USA), or IL-13 (50 ng/mL) (Sanofi-Synthelabo, France). In some experiments, macrophages were incubated with the specific inhibitors of PPARγ, GW9662 (5 µM) and T0070907 (2 µM) (Cayman Chemical, USA) or of cPLA2, MAFP (10 µM, 20 µM) (Cayman Chemical, USA), 10 min before the addition of PPARγ ligands or IL-13. Macrophages were incubated for 20 h for binding, phagocytosis, and ROS assays and quantification of surface expressed markers; cells were cultured for 4 h for transcript quantifications.
For the analysis of the binding of C.albicans, 5.105 cultured-macrophages were incubated at 4°C for 20 min with mannan solution. The medium was removed by washing with cold DMEM, and peritoneal macrophages were subsequently challenged with six FITC-labeled yeasts per macrophage and binding was performed at 4°C. Binding was stopped after 20 min by washing the macrophages with ice-cold PBS. Macrophage monolayers were incubed with ice-cold PBS and gently scraped.The amount of C.albicans binding to the macrophages was determined using FACS based approach. The fluorescence was quantified on a Becton Dickinson FACScan using CellQuestPro software and used as indicator of the binding efficiency.
For analysis of phagocytosis of C.albicans, 5.105 cultured-macrophages were incubated at 4°C for 20 min with mannan and/or laminarin solution. The medium was removed by washing with cold DMEM, and the peritoneal macrophages subsequently challenged with six FITC-labeled yeasts per macrophage and phagocytosis was initiated at 37°C in an atmosphere of 5% CO2. Phagocytosis was stopped after 60 min by washing the macrophages with ice-cold PBS. Macrophage monolayers were incubed with ice-cold PBS-EDTA (5 mM) and gently scraped. The amount of C.albicans engulfed by macrophages was determined using FACS based approach. The distinction between internalized yeast cells and those attached to macrophage surface was done via quenching the FITC-fluorescence with trypan blue. The remaining fluorescence was quantified on a Becton Dickinson FACScan using CellQuestPro software and used as indicator of the phagocytosis efficiency.
The macrophages were plated in 96-well Falcon plates (2.105 macrophages/well). The oxygen-dependent respiratory burst of macrophages was measured by chemiluminescence (CL) in the presence of 5-amino-2,3-dihydro-1,4-phthalazinedione (luminol) using a thermostatically (37°C) controlled luminometer (Wallac 1420 Victor2, Finland). The generation of CL was monitored continuously for 1 hr after incubation of the cells with luminol (66 µM) and after Candida albicans challenge at a yeast-to-macrophage ratio of 3∶1 or non-opsonized zymozan (ZNO) at final concentration of 2 µg/mL. Statistical analysis was performed using the area under the curve expressed in counts x seconds.
After 20 h of culture, the culture medium was removed and macrophage monolayers were incubated with ice-cold PBS-EDTA (5 mM) and gently scraped. After washing by centrifugation, the surface Dectin-1 or CD36 expressions were detected respectively using FITC-Dectin-1 mAb (Serotec, Düsseldorf, Germany) or PE-CD36 mAb (Tebu-Santa Cruz) and compared with an irrelevant appropriate isotype control. To characterize Cre (0/Tg) macrophages, the analysis was performed on non adherent cells. The labeled mAbs anti-F4/80-PE-Cy5, anti-CD11b-Alexa 647, and anti-TLR2-alexa 488 were obtained from Serotec. mAb anti-SIGNR1 and anti-TLR4 were obtained from eBioscience, and anti-CD14 was obtained from (BD PharMingen). To evaluate the MR surface expression, we have used a MR-specific ligand conjugated to FITC; macrophages were incubated with FITC-labeled mannosylated bovine serum albumin. A population of 5000 cells was analyzed for each data point. All analyses were done in a Becton Dickinson FACScan using CellQuestPro software.
Total RNA was prepared with RNeasy® Mini Kit columns (QIAGEN) using the manufacturer's protocols. The synthesis of cDNA was completed with QuantiTect® Reverse Transcription (QIAGEN) according to the manufacturer's recommendations and primed with hexamers. Quantitative real-time PCR was performed on a LightCycler system (Roche Diagnostics) using QuantiFastTM SYBR® Green PCR (QIAGEN). Ten microliters of reaction mixture were incubated; the amplifications were performed for 40 cycles (10 s at 95°C and 60 s at 60°C) for Dectin-1 and β-actin. The primers (at a final concentration of 10 mM) were designed with the software Primer Express (Applied Biosystems, Foster City, CA). Sequences were as follows: (sens) 5′-TGG AAT CCT GTG GCA TCC ATG AAA-3′; (antisens) 5′-TAA AAC GCA GCT CAG TAA CAG TCC G 3′ for β-actine, and (sens) 5′-CAT CGT CTC ACC GTA TTA ATG CAT-3′ (antisens) 5′-CCC AGA ACC ATG GCC CTT-3′ for Dectine-1.
Real-time PCR data are represented as Ct values, where Ct was defined as the crossing threshold of PCR by the Light-Cycler® System. For calculating relative quantification of β-GR mRNA expression, we have used the following procedure.
ΔCtβ-GR = CtSample−CtVehicle.ΔCtβ-actin = CtSample−CtVehicle. Then, ΔΔCt represented the difference between ΔCtβ-actin and ΔCt β-GR calculated by the formula ΔΔCt = ΔCtβ-actin−ΔCtβ-GR. Finally, the N-fold differential expression of β-GR mRNA samples compared to the vehicle was expressed as 2ΔΔCt.
Each experiment was performed independently at least three times and the results of one representative experiment are shown.
For the plasmid transfection, the macrophage murine cell line RAW264.7 was maintained in an exponential growth phase by subsequent splitting in DMEM complemented with 10% of FCS. The day prior to transfection, cells were splited in 24-well dishes. Then the complete medium was replaced by DMEM without any serum and the cells were transiently transfected for 18 h at 60–80% of confluence with Fugen 6 (Roche, Switzerland). The ratio of DNA/Fugen was 1∶2 with 1 µg of DNA. On the day of stimulation, the medium was discarded and fresh complete medium was added with stimulations as indicated in the figures. The pCMV-luc (CEA, France) served as a control. The pCMV-mPPARγ, a gift from Ron Evans (The Salk Institute, San Diego, CA), was encoded for the mouse nuclear receptor PPARγ.
For the siRNA transfection, siRNA control and siRNA to knockdown PPARγ (sc-29456) were transfected into murine peritoneal macrophages with the siRNATransfection Reagent in the siRNA Transfection medium as described in manufacturer's protocol (Santa Cruz biotechnology, Inc.)
For in vitro cytokine expression, peritoneal macrophages were added to 96-well plates (2.105 macrophages/well) and then stimulated with heat-killed C.albicans at a yeast-to-macrophage ratio of 3∶1 for 24 h, or with non-opsonized zymosan at a final concentration of 2 µg/mL. or with LPS at a final concentration of 100 ng/mL. Supernatants were recovered and frozen at −70°C before analysis. The production of TNFα in the cell supernatants was determined with a commercially available OptiEIA kit (BD Biosciences) according to the manufacturer's instructions.
Murine peritoneal macrophages were prelabeled with [3H]arachidonic acid. Briefly, adherent murine peritoneal macrophages (5×105 per well in 24-well plates) were cultured for 18 hours at 37°C under an atmosphere of 5% CO2, in DMEM (0.5 mL) containing 1% FCS and 1 µCi/mL [3H]arachidonic acid as previously described [35]. After 18 h, the culture medium was removed and pre-labeled macrophages were washed three times with 0.5 mL DMEM containing 1% FCS; after, the cells were treated or not with 100 nM of phorbol 12-myristate 13-acetate (PMA) or C.albicans at a ratio of 3∶1 (yeast∶macrophage) for 2 h. The [3H]arachidonic acid metabolites released into the culture medium by stimulated or unstimulated macrophages were quantified by measurement of the radioactivity by beta liquid scintillation counting using a 1217 Wallac Rackbeta LKB 1217.
For C.albicans DNA extraction, 250 µL of each tissue homogenate was prepared with the High Pure PCR Template preparation kit (ROCHE) using the manufacturer's protocols.
The Light Cycler PCR and detection system (Roche Diagnostics, Mannheim, Germany) was used for amplification and online quantification as previously described [13].
All animal experimentation was conducted in accordance with accepted standards of humane animal care. The model of gastrointestinal candidiasis was established in 8-week-old female control or macrophage-specific Dectin-1 deficient mice. Mice were given 0.3 mL of the yeast suspension by the oral route (5.106 or 5.107 C.albicans CFU per mouse).
Therapeutic studies were performed on separate groups of 6 mice each infected with C. albicans. Mice received the treatment in 500 µl of NaCl 0.9% by the intraperitoneal route. The final DMSO concentration was lower than 0.1% (v/v). Mice were treated with rosiglitazone (Cayman) one day prior to infection, the day of infection and then every two days with a dose of 2.8 µg/g of mouse.
No colonized animals died during the course of the study. On day 5, all mice were euthanized using CO2 asphyxia and the peritoneal cells harvested. Macrophages of infected animals were used to investigate phagocytosis and ROS production and to evaluate the surface expression of the MR. Previous data shown that Candida infection had no effect on effector macrophage functions (phagocytosis and ROS production).
In parallel, standardized samples of stomach and cecum were aseptically removed and homogenized in 400 µL of sterile-normal saline using tissue-lyser beads (MP biomedical). Fungal burdens of the tissues are shown as log yeasts per gram of tissue after quantification of C.albicans by RT-PCR.
For each experiment, the data were subjected to one-way analysis of variance followed by the means multiple comparison method of Bonferroni-Dunnett. p<0.05 was considered as the level of statistical significance.
|
10.1371/journal.pntd.0000748 | A Virulent Wolbachia Infection Decreases the Viability of the Dengue Vector Aedes aegypti during Periods of Embryonic Quiescence | A new approach for dengue control has been proposed that relies on life-shortening strains of the obligate intracellular bacterium Wolbachia pipientis to modify mosquito population age structure and reduce pathogen transmission. Previously we reported the stable transinfection of the major dengue vector Aedes aegypti with a life-shortening Wolbachia strain (wMelPop-CLA) from the vinegar fly Drosophila melanogaster. Here, we report a further characterization of the phenotypic effects of this virulent Wolbachia infection on several life-history traits of Ae. aegypti. Minor costs of wMelPop-CLA infection for pre-imaginal survivorship, development and adult size were found. However, we discovered that the wMelPop-CLA infection dramatically decreased the viability of desiccated Ae. aegypti eggs over time. Similarly, the reproductive fitness of wMelPop-CLA infected Ae. aegypti females declined with age. These results reveal a general pattern associated with wMelPop-CLA induced pathogenesis in this mosquito species, where host fitness costs increase during aging of both immature and adult life-history stages. In addition to influencing the invasion dynamics of this particular Wolbachia strain, we suggest that the negative impact of wMelPop-CLA on embryonic quiescence may have applied utility as a tool to reduce mosquito population size in regions with pronounced dry seasons or in regions that experience cool winters.
| A virulent strain of the vertically-inherited bacterium Wolbachia pipientis (wMelPop-CLA) from the vinegar fly Drosophila melanogaster has been established in the dengue vector Aedes aegypti as part of a biological strategy for dengue control. In this medically important disease vector, wMelPop-CLA infection shortens mosquito lifespan and effectively blocks dengue productivity within the mosquito – two powerful effects that could decrease the vectorial capacity of mosquito populations for transmission of dengue viruses. Here, we further characterize the phenotypic effects of wMelPop-CLA on several life-history traits of Ae. aegypti, and report that this infection influences the survival of this mosquito species during sustained periods of embryonic quiescence. From an applied perspective, we suggest that this novel phenotype may be a useful tool to reduce mosquito population size in regions where embryonic quiescence contributes towards survival of this species through seasonal changes in rainfall or temperature, and thus further reduce the probability of dengue transmission at the beginning of each wet season. This study also highlights key fitness parameters needed to accurately model invasion dynamics of this virulent Wolbachia strain.
| Aedes aegypti, the primary vector of dengue viruses throughout the tropics, is a mosquito species that has strong associations with human habitation [1]. In the past, control of dengue has been complicated by an inability to eradicate Ae. aegypti from urban environments and implement sustained vector control programs [2]. These challenges have highlighted the critical need for new approaches to curb a worldwide resurgence in dengue activity [3].
A novel approach for dengue control that has been proposed involves the introduction of the obligate intracellular bacterium Wolbachia pipientis into field populations of Ae. aegypti. Wolbachia are maternally inherited bacteria that naturally infect a wide diversity of invertebrate species [4], [5], and can rapidly spread through arthropod populations by manipulations to host reproduction such as cytoplasmic incompatibility [6]. Wolbachia infections could limit dengue transmission through two distinct mechanisms. The first by introducing Wolbachia strains that reduce the survival rate and associated vectorial capacity of the mosquito population [7], [8]. The second mechanism relies on the ability of some Wolbachia strains to interfere with the ability of RNA viruses to form productive infections in insects [9], [10] and potentially modulate the vector competence of Ae. aegypti for dengue viruses.
Towards this aim, we previously reported the stable transinfection of Ae. aegypti with a life-shortening Wolbachia strain wMelPop-CLA (a mosquito cell-line adapted isolate of wMelPop) [11], originally derived from the vinegar fly Drosophila melanogaster [12]. In this mosquito host, wMelPop-CLA has been shown to both reduce adult life span [11] and directly interfere with dengue virus infection [13], suggesting that this Wolbachia strain may have applied utility as a biological tool to reduce dengue transmission. However, prior to application in a field setting, a thorough understanding of any fitness effects that occur in wMelPop-CLA infected mosquitoes is required to accurately model infection dynamics and the impact of wMelPop-CLA on Ae. aegypti populations.
To further characterize this novel symbiosis and identify any fitness parameters likely to influence its spread throughout mosquito populations, we examined the phenotypic effects of wMelPop-CLA infection on several life-history traits across embryonic, pre-imaginal and adult stages of Ae. aegypti. We compared the developmental time and survivorship of pre-imaginal stages from infected and uninfected Ae. aegypti strains, and the effect of this infection on adult body size. We also considered the effect of wMelPop-CLA infection on embryonic viability during egg quiescence and reproductive fitness as mosquitoes age.
The work reported in this manuscript used human volunteers for mosquito feeding as approved by the University of Queensland Human Ethics Committee - Approval 2007001379. Written consent was obtained from each participant used for blood feeding.
wMelPop-CLA infected PGYP1 and tetracycline-cured PGYP1.tet strains of Ae. aegypti [11] were maintained at 25°C, 75–85% relative humidity, with a 12∶12 h light∶dark photoperiod. Larvae were reared in plastic trays (30×40×8 cm) at a set density of 150 larvae in 3 L distilled water, and fed 150 mg fish food (TetraMin Tropical Tablets, Tetra, Germany) per pan every day until pupation. Adults were kept in screened 30×30×30 cm cages, and provided with constant access to 10% sucrose solution and water. Females (5 days old) were blood-fed using human blood. For routine colony maintenance, eggs from PGYP1 were hatched 5–7 days post-oviposition (i.e. without prolonged desiccation) to initiate the next generation. All fitness experiments with PGYP1 were conducted at G20 to G22 post transinfection. The tetracycline-cured PGYP1.tet strain, generated at G8–G9 post-transinfection, was re-colonized with resident gut microflora from wild-type larvae as previously described [11].
Eggs (120 h old) from PGYP1 and PGYP1.tet strains were hatched synchronously in nutrient-infused deoxygenated water for 1 h. After hatching, individual first instar larvae (n = 156 per strain) were placed into separate plastic 30 mL plastic cups with 20 mL of water, and fed 1 mg powdered TetraMin suspended in distilled water each day until pupation. The number of days spent in each pre-imaginal life stage (i.e., 1st, 2nd, 3rd and 4th instars, pupae), mortality at each stage, and sex of eclosing adults were recorded every 24 h. Stage-specific development and eclosion times for each strain were compared using Mann-Whitney U (MWU) tests conducted in Statistica Version 8 (StatSoft, Tulsa, OK).
As an indicator of adult body size, wing lengths of PGYP1 and PGYP1.tet mosquitoes (n = 50 of each sex) derived from the pre-imaginal development time assay were measured (distance from the axillary incision to the apical margin excluding the fringe of scales) [14]. Wing lengths of males and females from each strain were compared using MWU tests.
Individual PGYP1 and PGYP1.tet population cages (30×30×30 cm), each containing 200 males and 200 females per strain, were maintained over multiple gonotrophic cycles, with ad libitum access to 10% sucrose solution and water for the duration of their life span. During each cycle, females were provided with a human blood meal for 2×10 min periods on consecutive days, and 96 h post-blood meal a random sample of females (n = 48) was collected from each cage and isolated individually for oviposition. Following a set 24 h period for oviposition, females were returned to their respective cages and the proportion of females laying eggs determined. Eggs were conditioned and hatched 120 h post-oviposition as described above, and the total number of eggs (fecundity) and hatched larvae (fertility) from each female were recorded. To ensure that gravid females not sampled for oviposition could also lay eggs every cycle, oviposition cups were introduced into each stock cage (96 h post-blood meal) for a period of 48 h. Females were then blood fed to initiate the next gonotrophic cycle.
Cages were sampled until all females in the population were dead, which occurred after 7 and 16 gonotrophic cycles for PGYP1 and PGYP1.tet strains respectively. To ensure PGYP1.tet females did not become depleted of sperm, young males (3 days old) were supplemented to this cage after 8 gonotrophic cycles. Multiple linear regression analysis was used to detect trends in fecundity/fertility of mosquitoes from each strain over their lifespan. Student's t-test was used to compare the fecundity/fertility of mosquitoes from both strains of the same age.
PGYP1 and PGYP1.tet females were blood-fed on human blood, and 96 h post-blood meal isolated individually for oviposition in plastic Drosophila vials with wet filter paper funnels. After oviposition, egg papers were kept wet for 48 h, after which time they were removed from vials, wrapped individually in paper towel, and conditioned for a further 72 h at 25°C and 75–85% relative humidity. Egg batches were then moved to their respective storage temperature of 18°C, or 25°C in glass desiccator jars; maintained at a constant relative humidity of 85% with a saturated KCl solution [15]. For each temperature, 20 oviposition papers from each strain were hatched at seven time points at 7 day-intervals (5 to 47 days post-oviposition) by submersion in nutrient-infused deoxygenated water for 48 h. To hatch any remaining eggs, oviposition papers were dried briefly then submersed for a further 5 days and before the final number of hatched larvae was recorded. Multiple linear regression analysis was used to detect trends in the viability of eggs from each strain over time. MWU tests were used to compare viability of eggs between strains at the same storage age.
No significant differences in development times for larval stages of wMelPop-CLA infected PGYP1 or tetracycline-cured PGYP1.tet males were found (MWU, P>0.05 for all comparisons) (Table 1). In contrast, the mean development time for male PGYP1 pupae (64.88±1.38 h) was significantly greater relative to PGYP1.tet (57.00±1.25 h) (MWU, U = 1892.00, P<0.001), resulting in a longer cumulative time to eclosion for this strain (MWU, U = 1484.50, P<0.001). For females, development times for immature stages were not significantly different between strains; except for third instar larvae where PGYP1 development times were increased by ∼5 h relative to PGYP1.tet (MWU, U = 1929.00, P = 0.013) (Table 1). Despite this delay, eclosion times for PGYP1 females were not significantly different from PGYP1.tet (MWU, U = 2185.50, P = 0.15). Overall, the survivorship of immature stages from both strains to adulthood was identical (96.15%).
A comparison of the wing lengths of newly emerged adults from both strains revealed a minor, yet statistically significant adult size cost to wMelPop-CLA infection for both sexes. Wing lengths of PGYP1 males (2.36±0.01 mm, n = 50) were significantly shorter than those of PGYP1.tet males (2.46±0.02 mm, n = 50) (MWU, U = 661.50, P<0.0001). A smaller size difference (MWU, U = 955.00, P = 0.04) was found between PGYP1 females (3.03±0.03 mm, n = 50) and PGYP1.tet females (3.09±0.03 mm, n = 50).
PGYP1 and PGYP1.tet females had similar reproductive outputs in terms of the number of eggs oviposited and the number of viable larvae hatched per female during their first gonotrophic cycle (Fig. 1A and B). However, during subsequent cycles both fecundity and fertility of PGYP1 females decreased at an accelerated rate (fecundity: R2 = 0.5068, F1,299 = 307.30, P<0.001; fertility: R2 = 0.3517, F1,299 = 162.20, P<0.001) relative to females from the PGYP1.tet strain (fecundity: R2 = 0.3167, F1,602 = 278.95, P<0.001; fertility: R2 = 0.1506, F1,602 = 106.76, P<0.001). For example, as PGYP1 females aged the average number of larvae produced per female decreased such that by the second cycle a 15% cost to reproductive output was observed relative to uninfected PGYP1.tet females, which progressively declined to a 40% cost by the fifth cycle (t-tests, P<0.05 for all comparisons). A large proportion of PGYP1 females that were randomly sampled for oviposition at the six and seventh gonotrophic cycles did not produce eggs (Fig. 1C), leading to a further decline in fecundity and fertility of this strain (Fig. 1A and B). This appeared to be due to defects in feeding behaviour, as many of these older PGYP1 females were observed to be unsuccessful in obtaining a blood meal (data not shown). Such a dramatic decrease in oviposition rates was not evident for PGYP1.tet females as they aged (Fig. 1C).
The viability of quiescent embryos from the wMelPop-CLA infected PGYP1 strain decreased over time at 25°C and 18°C, whereas viability of embryos from the tetracycline-cured PGYP1.tet strain was relatively stable at both storage temperatures (Fig. 2). At 25°C (Fig. 2A), there was no significant difference in embryonic viability between PGYP1 (80.93±5.12%) and PGYP1.tet strains (74.96±4.37%) at 5 days post oviposition (MWU, U = 146.50, P = 0.1478). As quiescent embryos aged, however, PGYP1 embryonic viability decreased rapidly over time (R2 = 0.6539, F1,138 = 260.73, P<0.001), such that by 40 days post oviposition very few PGYP1 eggs hatched (0.44±0.24%). In contrast, PGYP1.tet embryonic viability remained relatively constant over time (R2 = 0.0005, F1,138 = 0.07, P = 0.7897) with ∼75% of quiescent eggs hatching at each time point. An analogous trend was observed at 18°C (Fig. 2B), where initially hatch rates were comparable between the two strains, but subsequently a greater loss in embryonic viability was observed for PGYP1 (R2 = 0.4035, F1,138 = 93.34, P<0.001) relative to PGYP1.tet (R2 = 0.0803, F1,138 = 12.05, P<0.001). This was particularly evident at 12 days post oviposition where embryonic viability declined more rapidly in PGYP1 (9.88±2.96%) compared to PGYP1.tet (68.06±4.12%) after being moved to a cooler storage temperature (MWU, U = 5.00, P<0.0001).
In its native D. melanogaster host, wMelPop induces minor phenotypic effects during pre-imaginal life-history stages [12], [16]. However, after adult emergence, somatic and nervous tissues of flies gradually become densely populated with Wolbachia leading to overt pathology and shortened life span [12]. Similarly, in this study we observed minor costs of wMelPop-CLA infection during Ae. aegypti pre-imaginal development, with the phenotypic effects of this virulent Wolbachia strain increasing as adult mosquitoes aged.
A small delay in the mean time to eclosion was observed for wMelPop-CLA infected Ae. aegypti males, but not females relative to their tetracycline-cured counterparts. Increased egg-to-adult development times have previously been characterized for certain D. melanogaster genotypes infected by wMelPop [16]. Differences in development time were also reflected by variations in adult body size, where size costs to wMelPop-CLA infection were more pronounced for infected males than infected females. Taken together, results from development time, immature survivorship and adult size assays suggest a minor physiological cost to wMelPop-CLA infection during Ae. aegypti pre-imaginal development. Additional studies that introduce larval competition [17], and which utilise a wide variety of potential nutrient sources likely to be encountered in field environments are required to fully evaluate the impact of wMelPop-CLA infection on this stage of Ae. aegypti life-history.
A common trait observed in many mosquito species, including Ae. aegypti, is a general decline in the numbers of eggs laid by females over successive gonotrophic cycles, which is thought to be caused by increasing ovarian follicle degeneration as mosquitoes age [18], [19]. Fecundity of both wMelPop-CLA infected and tetracycline-cured mosquito strains was initially comparable, consistent with previous assays using the PGYP1 Ae. aegypti strain [11]. Over subsequent gonotrophic cycles, however, fecundity declined at an accelerated rate in PGYP1 relative to the PGYP1.tet strain suggesting that wMelPop-CLA infection contributed to a reduction in reproductive fitness. This may be related to a progressive increase in pathology induced by this Wolbachia strain in reproductive tissue, as commonly observed in somatic and nervous tissue [12], as mosquitoes age. In Drosophila simulans, fecundity costs of wMelPop infection were initially high after transinfection of this strain from D. melanogaster, but attenuated over subsequent generations [20]. It remains possible that such costs to reproductive fitness will also diminish for PGYP1, as wMelPop-CLA and Ae. aegypti further adapt to each other.
Interestingly, as wMelPop-CLA infected females aged we observed a rapid decrease in the number of randomly sampled PGYP1 females that would oviposit in gonotrophic cycles 5 to 7. This time range correlates with the onset of wMelPop-CLA induced life-shortening in Ae. aegypti [11]. Such a decline in oviposition rate may be directly related to pathology induced in reproductive tissues, or most likely be due to unsuccessful blood feeding behaviour observed in wMelPop-CLA infected mosquitoes as they age [21]. Such an age-related decline in fecundity may limit or influence the rate at which the wMelPop-CLA infection to spreads through an Ae. aegypti population, and should therefore be considered in the development of models predicting invasion dynamics of this Wolbachia strain. A complete understanding of this magnitude of this effect will require further determination of the relative reproductive contribution of different age-classes of Wolbachia-infected and uninfected Ae. aegypti to mosquito population size in a more ecologically relevant field cage setting.
In addition to the previously characterized life-shortening [11] and viral interference phenotypes [13] of wMelPop-CLA infection in Ae. aegypti, a third major effect described in this study is the observation that this infection decreases the viability of quiescent embryos over time. The viability of eggs laid by tetracycline-cured Ae. aegypti remained high over the 1.5 month test period. In contrast, the viability of the wMelPop-CLA infected PGYP1 strain declined rapidly over time. This decrease in embryonic viability was particularly evident after PGYP1 eggs were moved to a cooler storage temperature, possibly reflecting decreased levels of cold tolerance in the presence of infection. Such decreases in embryonic viability are not observed in the closely related mosquito species Aedes albopictus, which is infected by two avirulent Wolbachia strains (wAlbA and wAlbB) [22]. Moreover, reductions in embryonic viability are also not seen in Ae. aegypti lines transinfected with wAlbB from Ae. albopictus [23].
The impact of wMelPop-CLA on survival of quiescent eggs may have important implications for the spread and maintenance of this infection in Ae. aegypti populations, as well as mosquito population dynamics. Larval habitats of container breeding mosquito species such as Ae. aegypti and other members of the subgenus Stegomyia, are often subject to high selection pressures due to drying during certain seasonal periods [24]. In this context, the effects of wMelPop-CLA on Ae. aegypti populations are likely to be highly dependent on geographical location where field releases occur.
In tropical regions, such as Thailand and Vietnam, where an abundance of both permanent and transient larval breeding sites exist and rainfall occurs on a regular basis or containers are maintained full of water by human intervention, it is likely that under certain release thresholds wMelPop-CLA will be able to spread and persist in local Ae. aegpyti populations. However, in regions with a pronounced dry season, such as northern Australia, where drying of eggs may occur, it would be expected that this effect would significantly reduce mosquito population size at the beginning of the following wet season due to wMelPop-CLA induced embryonic mortality. The magnitude of such an effect will be dependent on the ability of the wMelPop-CLA infection to invade an area under the action of CI before the onset of the dry season, as a concurrent decrease in Wolbachia prevalence in the mosquito population would also be expected if the infection had not spread to fixation prior to dry season onset.
From an applied perspective, we suggest that the ability of wMelPop-CLA to decrease mosquito viability during periods of embryonic quiescence may have potential utility in certain geographic locations as a tool to reduce mosquito population size at the beginning of each wet season. An analogous genetic strategy for population suppression has previously been proposed, involving the release of Ae. albopictus males adapted to tropical regions into temperate field populations of this mosquito species to reduce their over-wintering ability [25]. Given the importance of seasonal fluctuations in mosquito population density in influencing dengue epidemics [26], this phenotype may act synergistically with described effects of this infection on mosquito lifespan [11] and vector competence [13] to further reduce the probability of virus transmission in several disease-endemic countries worldwide. However, the observation that wMelPop-CLA influences fitness of both embryonic and adult life-history stages, also suggests that the invasion dynamics of this virulent Wolbachia strain are likely to be complex and highly sensitive to the ecological setting where field releases occur.
|
10.1371/journal.pntd.0004276 | Asymmetric Nerve Enlargement: A Characteristic of Leprosy Neuropathy Demonstrated by Ultrasonography | Neurological involvement occurs throughout the leprosy clinical spectrum and is responsible for the most feared consequences of the disease. Ultrasonography (US) provides objective measurements of nerve thickening and asymmetry. We examined leprosy patients before beginning multi-drug therapy aiming to describe differences in US measurements between classification groups and between patients with and without reactions.
Eleven paucibacillary (PB) and 85 multibacillary (MB) patients underwent nerve US. Twenty-seven patients had leprosy reactions (type 1, type 2 and/or acute neuritis) prior to US. The ulnar (at the cubital tunnel–Ut–and proximal to the tunnel–Upt), median (M) and common fibular (CF) nerves were scanned to measure cross-sectional areas (CSAs) in mm2 and to calculate the asymmetry indexes ΔCSA (absolute difference between right and left CSAs) and ΔUtpt (absolute difference between Upt and Ut CSAs). MB patients showed greater (p<0.05) CSAs than PB at Ut (13.88±11.4/9.53±6.14) and M (10.41±5.4/6.36±0.84). ΔCSAs and ΔUtpt were similar between PB and MB. The CSAs, ΔCSAs and ΔUtpt were similar between PB patients with reactions compared to PB patients without reactions. MB patients with reactions showed significantly greater CSAs (Upt, Ut and M), ΔCSAs (Upt and Ut) and ΔUtpt compared to MB patients without reactions. PB and MB showed similar frequencies of abnormal US measurements. Patients with reactions had higher frequency of nerve thickening and similar frequency of asymmetry to those without reactions.
This is the first study to investigate differences in nerve involvement among leprosy classification groups using US before treatment. The magnitude of thickening was greater in MB and in patients with reactions. Asymmetry indexes were greater in patients with reactions and did not significantly differ between PB and MB, demonstrating that asymmetry is a characteristic of leprosy neuropathy regardless of its classification.
| Leprosy is an infectious disease that affects the peripheral nerves, leading to nerve thickening, asymmetry and dysfunction; therefore, early detection of leprosy neuropathy is important for preventing deformities and disabilities. We examined peripheral nerve involvement using ultrasonography (US) in 96 leprosy patients prior to treatment, aiming to better understand differences in neuropathy patterns between leprosy classification groups and between patients with and without leprosy reactions. Patients underwent bilateral US of the ulnar (at the cubital tunnel and proximal to the tunnel), median and common fibular nerves to measure thickening and asymmetry. We found that nerve thickening was more severe in patients with a high bacillary load (multibacillary) and in those with previous reactions. Nerve asymmetry measurements were greater in the patients with previous reactions. Asymmetry did not differ significantly between the paucibacillary and multibacillary patients, demonstrating that asymmetry is a characteristic of leprosy neuropathy regardless of its classification.
| Neurological involvement is present throughout the leprosy clinical spectrum, and nerve impairment is responsible for the most feared consequences of the disease; therefore, some authors advocate that leprosy should be regarded as a chronic neurological condition rather than a skin disease [1–5]. Several authors have postulated that tuberculoid patients have asymmetric nerve thickening, lepromatous patients have symmetric and diffuse involvement, and borderline patients have variable and usually intense nerve enlargement [1,3,6,7]. Leprosy reactions (acute neuritis, types 1 and 2 reactions) can lead to additional nerve damage due to immune-mediated mechanisms. They may be superimposed on the chronic course of the disease and require immediate treatment [1,3,5,6].
Neurophysiologic and imaging studies can be used to investigate neurological impairment in leprosy. Although neurophysiology provides detailed information about dysfunction of affected nerves, it does not reveal anatomic changes, such as thickening and fascicular pattern changes [8,9]. High-resolution ultrasonography (US) permits examination of multiple nerve trunks over a long course in a few minutes, and compared with magnetic resonance imaging, US is considered more accessible and reasonably precise [8,10–12]. Furthermore, it is reported that US is more accurate than clinical palpation for assessment of peripheral nerve enlargement [11] and it provides objective measurements of peripheral nerve thickening and asymmetry [13]. To our knowledge, no published studies have investigated differences in US nerve measurements across the leprosy spectrum.
The purpose of this study was to investigate peripheral nerve thickening and asymmetry, as evaluated through US measurements, in leprosy patients, examining differences between leprosy types grouped according to the Ridley-Jopling and WHO classifications and to assess the influence of leprosy reactions on US findings.
The Ethics Committee of the Clinics Hospital of Ribeirão Preto Medical School approved the study (process n°02663112.0.0000.5440). Written informed consent was obtained from all participants. Some participants were minors and their parents provided written consent on behalf of them.
Ninety-six consecutive leprosy patients that were referred to Leprosy Reference Center at the Clinics Hospital of Ribeirão Preto Medical School were included in the study and underwent bilateral high-resolution US of the peripheral nerves before starting World Health Organization (WHO) multi-drug therapy. Leprosy diagnosis was established based on clinical signs and symptoms, skin smears, skin biopsy, and neurophysiologic examination when necessary. Patients were classified according to the Ridley-Jopling [14] classification in five groups: tuberculoid (TT), borderline-tuberculoid (BT), borderline-borderline (BB), borderline-lepromatous (BL), and lepromatous (LL). Patients were also grouped in paucibacillary (PB) and multibacillary (MB) according to the WHO operational classification [15].
Medical charts were reviewed for the identification of any leprosy reactions prior to US evaluation. Leprosy reactions were classified as cutaneous reactions (type 1 or 2) associated or not with acute neuritis. Type 1 cutaneous reactions were defined as presence of erythema and edema of skin lesions. There may be accompanying neuritis and edema of the hands, feet, and face. Type 2 reactions (erythema nodosum leprosum) were defined as presence of tender subcutaneous skin lesions. There may be accompanying neuritis, iritis, arthritis, orchitis, dactylitis, lymphadenopathy, edema, and fever. Neuritis was diagnosed if patients presented acute inflammation of one or more peripheral nerve trunk detected by swelling and/or functional impairment with spontaneous nerve pain and/or nerve tenderness on palpation. To statistical analysis patients were divided according to the presence of any kind of reaction prior to US examination or absence of reactions.
Patients with a history of diabetes mellitus, hypothyroidism, human immunodeficiency virus infection, trauma-related peripheral nerve disease or alcoholism were excluded from the study.
Musculoskeletal radiologists with previous fellowship training in nerve imaging performed all US sessions using a 12-MHz linear transducer model HDI-11 (Philips Medical Systems, Bothell, Washington, USA). Patients were examined in a seated position with 45° flexed elbows and 90° flexed knees. The ulnar (at the cubital tunnel area–Ut–and proximal to the tunnel–Upt), median (M) and common fibular (CF) nerves were systematically scanned along the transverse and longitudinal axes. Ulnar nerves were scanned from the middle third of the arm to the middle third of the forearm. M nerves were evaluated at the middle and distal thirds of the forearm. CF nerves were evaluated from the distal third of the thigh to the knee at the fibular head. In some cases, it was not possible to examine nerves bilaterally due to amputation, cutaneous ulcers or other cutaneous alterations at the site of examination.
Nerve cross-sectional areas (CSAs) were measured by freehand delimitation at the inner borders of the echogenic rims of the nerves. The measurements were performed using the electronic cursor at the time of examination, and the CSAs were assessed at the level of maximum nerve thickening. Ulnar nerve maximum CSAs were measured in two different regions: above the medial epicondyle proximal to the cubital tunnel (Upt) and at the cubital tunnel (Ut).
CSA measurements were used to calculate nerve asymmetry as follows: (1) the differential CSA index (ΔCSA), which was calculated as the absolute difference between CSAs for each nerve point from one side to the contralateral side, and (2) the differential Ut-Upt index (ΔUtpt) of the ulnar nerves, which was calculated as the absolute difference between the largest and smallest CSAs of Upt and Ut points of ulnar nerves on the same side. High ΔCSA values reflect nerve asymmetry with the contralateral nerve. High ΔUtpt values reflect non-uniform and focal thickening of the ulnar nerve between the tunnel and pre-tunnel areas.
For the classifications of the CSA, ΔCSA and ΔUtpt values as normal or abnormal, we used previously published values obtained from healthy volunteers [13], considering as abnormal values that were greater than the mean plus 2 standard deviations. Additionally, to compare the PB and MB groups, we performed two different analyses: one including all patients and another including only those patients with at least one abnormal measurement for each variable (CSA, ΔCSA and ΔUtpt). With second analysis, we expected to reduce the possible bias due to the existence of different frequencies of abnormalities in PB and MB.
Statistical analysis was performed using JMP software version 10.0 (SAS Institute Inc., Cary, NC). We performed the Wilcoxon test to compare the means of two groups. To analyze differences between the means of three or more groups, we performed the Kruskal-Wallis test. For the comparison of proportions, we used the two-tailed Fisher’s exact test. Probability (p) values less than 0.05 were considered significant.
The average age ± standard deviation of the patients was 45.9 ± 16 years (age range 16–85 years); 56 (58.3%) of the patients were men, and 40 (41.7%) were women. The clinical classifications and incidences of neuritis, type 1 and/or type 2 cutaneous reactions occurring prior to US are presented in Table 1. For clinical data and CSAs values of each patient see supporting information (S1 Table).
Table 2 shows the results for CSA, ΔCSA and ΔUtpt of the nerves studied for the five types of leprosy according to the Ridley-Jopling classification.
For all nerve points studied, we observed maximum mean values of CSA in LL and BL patients; the percentages of enlarged nerves (abnormal CSAs) were also greater in these two groups, reaching 81.3% incidence of enlarged ulnar nerves at the cubital tunnel (Ut) in LL patients. Although the mean values of ΔCSAs showed variable results, with maximum values in different types of leprosy, we observed higher incidences of asymmetric nerves (abnormal ΔCSAs) in LL and BL patients. LL patients showed the maximum ΔUtpt mean and the highest percentage of nerve asymmetry considering this measurement.
Table 3 shows the CSAs, ΔCSAs and ΔUtpt according to the WHO operational classification for all patients and for the patients with at least one abnormal measurement for each variable. The two analyses yielded similar results, with the MB patients displaying greater CSAs, ΔCSAs, ΔUtpt and frequency of abnormalities at the Upt, Ut and M nerves. On the other hand, at the CF nerve, the PB patients showed similar or slightly greater CSAs and ΔCSAs.
Because borderline patients show variable nerve impairment, we compared only the two polar forms, thereby aiming to better characterize and highlight nerve thickening and asymmetry. For this analysis, we selected only TT (n = 8) and LL (n = 8) patients who had at least one abnormal measurement for each variable. Significantly greater CSAs were observed in the LL patients at the Upt (7.64±1.91 /14.9±9.08 mm2), Ut (9.53±6.14 / 15.56±4.57 mm2) and M (6.36±0.84 / 12.04±4.75 mm2) nerves; no significant difference was observed at the CF nerve. Although the ΔCSAs were also greater in LL patients, no statistically significant differences were observed at the Upt (2.01±2.12 / 8.11±8.96 mm²), Ut (4.41±5.25 / 4.44±2.23 mm2) and M nerves (0.86±0.48 / 1.94±1.91 mm2). At the CF nerve, the TT patients showed a non-significantly greater ΔCSA (9.29±18.14 / 2.4±2.23 mm2). It is important to report that one TT patient had the maximum ΔCSA value observed in the study at this nerve (52.7mm2). No significant difference in the ΔUtpt was observed between TT and LL (6.02±6.34 / 8.18±5.93 mm2).
The CSAs, ΔCSAs and ΔUtpt were similar between PB patients with previous leprosy reactions (neuritis, type 1 and/or type 2 reactions) compared to PB patients with no reaction. MB patients with reactions showed significantly greater CSAs compared to MB patients without reactions at the Upt (14.20±11.09 / 8.20±5.12 mm2), Ut (14.73±15.73 / 10.62±6.13 mm2) and M nerves (10.88±5.99 / 8.40±4.32 mm2). ΔCSAs were also significantly greater in MB patients with reactions at the Upt (7.39±11.07 / 2.84±5.59 mm2) and Ut nerves (8.33±19.29 / 3.06±4.61 mm2). In addition, the ΔUtpt was significantly greater in MB patients with previous leprosy reactions (5.82±7.36 / 3.85±5.21 mm2).
Considering that abnormalities in at least one US measurement reflect neuropathy, we sought to evaluate differences in the frequencies of abnormalities between PB and MB patients and also between patients who did and did not have reactions prior to US. The results of this analysis are presented in Table 4.
We found that nerve asymmetry detected on US is characteristic of leprosy, with similar frequencies of abnormal measurements found in PB and MB patients. Furthermore, we observed a tendency toward higher ΔCSA and ΔUtpt values in the latter group. As expected, we also found that thickening (CSA) of the peripheral nerves was more pronounced in MB. Another important finding was that MB patients with previous leprosy reactions (neuritis, type 1 and/or type 2) had greater CSA, ΔCSA and ΔUtpt values than MB patients without reactions; however, among the PB patients there was no significant difference in nerve thickening and asymmetry comparing the groups with and without reactions.
This is the first study in which differences in peripheral nerve thickening and asymmetry among patients of different leprosy classification groups were systematically investigated using accurate US measurements prior to specific treatment (WHO multi-drug therapy); the inclusion of patients at different stages of treatment in previous studies [10,11,16] could weaken conclusions about variations in the pattern of nerve involvement. Two studies in which US evaluation was performed before treatment have been reported, but those studies did not address differences between leprosy classification types [8,13].
There is a growing interest in US as a diagnostic tool for peripheral neuropathies. Nerve palpation is subjective and requires expertise [5], and even among trained professionals, the reliability of palpation of superficial peripheral nerves is unsatisfactory, with poor agreement between evaluators [17]. In a study that compared clinical examination and US of peripheral nerves in 20 leprosy patients and 30 healthy volunteers, Jain et al. [11] concluded that clinical examination was subjective and inaccurate, whereas US provided an objective evaluation of nerve damage and could identify more extensive involvement. Another previous study showed that US abnormalities may be present in patients with normal neurophysiological findings [8]. The concept that US should always be performed in addition to neurophysiological studies during the investigation of peripheral neuropathies is currently gaining strength [8,9,12].
In our study, we found that peripheral nerve involvement, objectively evaluated by US, is common in all types of leprosy. Considering that the presence of one or more abnormal nerve measurements reflects neuropathy, we observed similar frequencies of neuropathy in the PB and MB patients. The frequency of abnormalities was high even among the PB patients (72.7% of the PB patients had at least one thickened or asymmetric nerve), suggesting that enlargement and asymmetry of the peripheral nerves may be more frequently detected on US, corroborating the findings of Jain et al. [11]. These results support the idea that leprosy is a neurological disease [1–3,5] and reinforce the importance of conducting a detailed neurological exam for all patients.
The ulnar nerve was the most commonly involved nerve in MB patients; up to 81.3% of LL patients showed abnormal CSA values in the cubital tunnel area. Furthermore, thickening of peripheral nerves was also frequent in PB patients, especially at the common fibular nerve. The thickening was clearly more pronounced in MB patients at superior limb nerves, even when we analyzed only the group of patients in which some neuropathy was detected at US: the CSAs were significantly greater in the MB group, and the 95% confidence interval showed no overlap between PB and MB mean values at the ulnar (Upt and Ut) and median nerves. At the common fibular nerve, no difference was found between PB and MB patients; furthermore, the common fibular nerve was the most frequently affected nerve in PB patients, in agreement with observations from clinical studies that suggest that this nerve can be impaired even early in the disease course [7,18].
Clinical expertise and previous studies indicate that PB patients (especially TT) have asymmetric nerve enlargement, whereas MB patients, notably the patients exhibiting the lepromatous polar form of the disease, have symmetric and diffuse involvement [1,6]. Frade et al. [13] have shown that ΔCSA and ΔUtpt possess high specificity for a diagnosis of leprosy when leprosy patients are compared to healthy individuals and that these measurements allow for the objective quantification of peripheral nerve asymmetry. We found that the asymmetry measurements (ΔCSA and ΔUtpt) did not significantly differ between the PB and MB patients; moreover, our data indicate a weak tendency toward greater asymmetry in the latter group. The ΔCSA means were greater in MB at the Ut, Upt, and median nerves, and the highest percentages of abnormal ΔCSA values for all studied nerves were found in the LL and BL patients. ΔUtpt showed the same trend, with greater values in the MB patients and the highest percentage of asymmetry in LL patients. We compared only the two polar forms of the disease (TT and LL), aiming to emphasize the findings. Our results confirmed previous analyses, showing significantly greater values of CSA (Upt, Ut, and median nerves) in LL patients. Although asymmetry measurements did not differ significantly between TT and LL patients, we observed a tendency toward greater asymmetry in LL patients. The results for the common fibular nerve showed an opposite tendency, with greater values of ΔCSA (although without statistical significance) in TT patients; nevertheless, we emphasize that one TT patient had the maximum ΔCSA value at common fibular nerve (52.7mm2), which could have increased the mean of this group.
Leprosy reactions, especially acute neuritis, can lead to severe nerve impairment and require immediate treatment with steroids and other immunosuppressive drugs. Consistent with the results of previous studies, our results indicate that more severe enlargement and asymmetry of nerves occurs in MB patients with previous or active reactions in all evaluated nerves except for the common fibular nerve. The nerve measurements among PB patients with and without reactions did not show significant differences. Despite the fact that all patients included in our study were examined before beginning WHO multi-drug treatment, the majority of patients who had a history of reactions were already receiving anti-reaction treatment (prednisone, thalidomide and/or azathioprine) at the time of the US exam; thus, nerve swelling and asymmetry might have been diminished in these patients. Taken together with previous results, these findings indicate that chronic M. leprae nerve infection and its ability to cause inflammation and fibrosis, as well as the presence of leprosy reactions, are important causes of nerve thickening and asymmetry. Previous studies have addressed the influence of reactions on peripheral nerve imaging findings. Martinoli et al. [10] investigated US and magnetic resonance imaging findings for 23 leprosy patients and concluded that patients with previous or active leprosy reactions had nerve enlargement and fascicular abnormalities. These authors also identified the presence of intraneural color Doppler signal in patients with active reactions. Jain et al. [11] have found that increased blood flow can be present in multiple nerves distant to the affected dermal lesion. In our study, we investigated all patients for the presence of Doppler signal; however, because most of the patients with clinical signs of reactions were already receiving anti-reaction treatment at the time of the US exam, Doppler signals were observed only in a small number of them, and the Doppler results are not reported here.
One limitation of our study is the lack of neurophysiological correlation, which could provide useful information concerning nerve function abnormalities. However, our main objective was the evaluation of anatomic alterations, represented by nerve thickening and asymmetry. Two other studies in which ulnar nerve neuropathy was investigated using US and electrophysiology [8,16] found US abnormalities in patients with normal neurophysiological findings. Those studies also demonstrated that leprosy patients can have normal ulnar nerve US findings with significant electrophysiological changes. The authors concluded that leprosy patients can exhibit abnormal nerve anatomy with preserved nerve function and vice versa [8]. Therefore, although we did not perform electrophysiological tests, we consider that our results improve the understanding of anatomic changes in leprosy neuropathy.
Despite the fact that the sample size of this study is the largest reported for any study of the use of US for leprosy neuropathy evaluation, the division of the patients into six clinical classification types resulted in the presence of a small number of subjects in each group. Perhaps the numbers of patients included in the groups with the polar forms TT and LL were not large enough to reveal significant differences in asymmetry measurements between the groups.
In conclusion, the present study contributes to the understanding of the pattern of peripheral nerve involvement in leprosy patients. The US results reported here have revealed that thickening and asymmetry are common in leprosy patients and that these abnormalities occur at similar frequencies in PB and MB. Moreover, the magnitude of thickening was greater in the MB patients and in those with previous leprosy reactions. Nerve asymmetry did not significantly differ between the PB and MB patients, demonstrating that asymmetry is a characteristic of leprosy neuropathy regardless of its classification.
|
10.1371/journal.ppat.1003587 | Highly Significant Antiviral Activity of HIV-1 LTR-Specific Tre-Recombinase in Humanized Mice | Stable integration of HIV proviral DNA into host cell chromosomes, a hallmark and essential feature of the retroviral life cycle, establishes the infection permanently. Current antiretroviral combination drug therapy cannot cure HIV infection. However, expressing an engineered HIV-1 long terminal repeat (LTR) site-specific recombinase (Tre), shown to excise integrated proviral DNA in vitro, may provide a novel and highly promising antiviral strategy. We report here the conditional expression of Tre-recombinase from an advanced lentiviral self-inactivation (SIN) vector in HIV-infected cells. We demonstrate faithful transgene expression, resulting in accurate provirus excision in the absence of cytopathic effects. Moreover, pronounced Tre-mediated antiviral effects are demonstrated in vivo, particularly in humanized Rag2−/−γc−/− mice engrafted with either Tre-transduced primary CD4+ T cells, or Tre-transduced CD34+ hematopoietic stem and progenitor cells (HSC). Taken together, our data support the use of Tre-recombinase in novel therapy strategies aiming to provide a cure for HIV.
| Current antiretroviral combination therapy can efficiently suppress virus replication, but cannot eliminate HIV. Therefore, no cure for HIV exists. A main hurdle for virus eradication is seen in the existence of resting cells that contain integrated replication-competent, but temporarily silenced, HIV genomes. Therefore, the most direct approach to eliminating virus reservoirs is to remove HIV genomes from infected cells. As previous studies suggested, this may be achievable by Tre-recombinase, an engineered enzyme that can excise integrated HIV from host cell chromosomes. The present work analyzes the expression of Tre-recombinase in human cells and demonstrates highly accurate Tre activity in complete absence of Tre-related cytopathic effects. Furthermore, in vivo analysis of Tre-recombinase demonstrates highly significant antiviral effects of Tre in HIV-infected humanized mice. The presented data suggest that Tre-recombinase might become a valuable component of a future therapy that aims at virus eradication.
| The introduction of highly active antiretroviral therapy (HAART) into clinical practice in the mid-1990s profoundly reduced morbidity and mortality among HIV-1-infected patients, changing an almost always fatal disease into a manageable chronic illness [1]. However, HAART is costly and occasionally not well tolerated [2], [3]. Particularly long-term HAART is frequently accompanied by emerging new toxicities, resulting in secondary complications that include metabolic disorders (e.g. diabetes, hyperlipidemia), osteoporosis, cardiovascular disease and chronic kidney disease (reviewed in [4]–[6]). Furthermore, large cohort studies demonstrated that the life expectancy of patients receiving HAART still remains considerably shorter than that of uninfected subjects (recently reviewed in [7]). Most importantly, the fact that HAART does not eradicate HIV and that treatment intensification, even when employing advanced drug regimens, fails to completely clear the virus (reviewed in [7], [8]) highlights the urgency of pursuing new strategies to find a cure for HIV infection.
It is generally believed that the main hurdle to virus eradication is the persisting HIV-1 infection in latent reservoirs, particularly in memory CD4+ T cells (reviewed in [9]–[15]). Latently HIV-1-infected resting CD4+ T cells are apparently established early in infection. One current strategy to eliminate this pool of long-lived cells aims to specifically activate the transcriptionally quiescent provirus (i.e. the integrated replication-competent HIV-1 genome), for example by modifying its chromatin structure through histone deacetylase (HDAC) inhibitors (reviewed in [12], [13], [15]–[17]). Upon HDAC inhibitor-induced HIV-1 antigen expression, it is expected that these cells either experience HIV-1-induced cell death or are eliminated by cytotoxic T cells (CTLs). It is fair to assume that such purging strategies would greatly benefit from a technology that can concurrently remove integrated HIV-1 from the pool of productively infected cells, thereby restoring, or at least improving the patient's immune function.
A novel strategy to remove integrated HIV-1 is based on a tailored site-specific recombinase (Tre), derived by molecular evolution of the bacteriophage recombinase Cre [18]–[20]. Tre targets a specific 34 bp sequence (loxLTR) derived from a primary HIV-1 strain [21] located in the proviral LTR regions, resulting in excision of the integrated proviral DNA from the genome of infected cultured cells [18]. This process not only suppresses viral replication, but in theory may also help eradicate HIV from an infected individual (reviewed in [22]).
Administering Tre-recombinase to patients will most likely require a gene therapy approach. In principle, genetic therapies against HIV either modify the patient's peripheral CD4+ T cells or patient-derived CD34+ hematopoietic stem cells (HSC) [23]–[25]. It is anticipated that the former strategy would lead to beneficial antiviral, although transient effects. The latter application will presumably be the preferred strategy in Tre-based virus eradication approaches, since, in theory, it allows perpetual repopulation of the patient's hematopoietic system with Tre-expressing HIV-1 target cells. These cells may be selected in vivo [26], since upon de novo infection they are able to remove the integrated HIV-1 proviral DNA, and thus remain functionally immune competent.
Independently of the selected gene therapy strategy, and prior to its potential use in HIV-infected patients, vector technology has to be developed that allows safe and efficient gene transfer followed by reliable transgene expression in target cells. Moreover, the absence of cytopathic and/or genotoxic effects upon vector-mediated Tre expression, and the accurate excision of HIV proviral DNA from chromosomal integration sites has to be demonstrated. Finally, the antiviral effects of Tre-recombinase have to be shown in vivo, i.e. in an appropriate animal model for HIV-1 infection. All these analyses will be of utmost importance for developing a potential Tre-based therapy to treat HIV infection.
For delivery of Tre-recombinase an HIV-1-derived replication-incompetent lentiviral vector (LV) was constructed that provides high safety levels due to a split packaging system; the self-inactivating (SIN) vector design; and a sequence element introduced to improve transcriptional termination [27], [28] (Figure 1A). To avoid transgene-related side effects, the gene sequence encoding Tre-recombinase was placed under the control of an engineered Tre-resistant tandem TAR repeat (2TAR), the cis-active target sequence of the HIV-1 Tat trans-activator [29]. This not only limits Tre expression to HIV-infected cells, but a duplicated TAR element also positively influences internal LTR promoter activity in the presence of Tat (Figure S1 in Text S1). Finally, constitutive expression of the GFP marker protein was facilitated by the PGK promoter (Figure 1A).
First, we investigated whether Tre-recombinase introduced by the LV construct is faithfully expressed in HIV-infected cells, and in turn, excises integrated chromosomal proviral DNA. We generated a reporter cell line, HeLa-smurf, which is stably infected with a replication-incompetent HIV-1 mutant with the env gene partially deleted and the nef open reading frame partially replaced by a marker gene (blb) (Figure 1B). Thus, Tre-mediated excision of the proviral genome results in loss of blue fluorescent protein (BFP) expression, which can be tracked by flow cytometry.
To monitor Tre activity, HeLa-smurf cells were transduced in triplicates with the GFP-expressing LV particles LV-Tre or a scrambled tre version, the Tre-negative control construct LV-Ctr, cultured for various time periods and analyzed for BFP and GFP expression. After transduction, both lentiviral constructs produced a GFP/BFP double positive cell population (see top right quadrants in Figure 1C, and plotted data in Figure S2 in Text S1). However, whereas the BFP/GFP double positive population of LV-Ctr-transduced cells subsequently remained stable over time, the BFP/GFP double positive population observed in LV-Tre treated cells started to decrease at 72 hours post transduction (Figure 1C and Figure S2 in Text S1). Simultaneously, the GFP-only positive population increased, suggesting that Tre-mediated excision of HIV-1 proviral DNA has occurred. Importantly, as the BFP/GFP double-positive population declined, expression of Tre-recombinase decreased, owing to the Tat-driven promoter providing temporal Tre expression (Figure 1D).
We performed a panel of assays to verify Tre-mediated excision of the provirus. First, we investigated the excised circular recombination product using a PCR assay (Figure 2A). An excision-specific fragment was only detected in DNA prepared from Tre-treated HeLa-smurf cells (LV-Tre), but not in cells treated with the negative control vector (LV-Ctr) (Figure 2B). The recombination product appeared as early as 24 hours after transduction, and can be detected until 6 days post transduction (p.t.) time point. As expected, the intensity of the PCR signal continuously increased, reaching a maximum at day 4 p.t., before starting to decline (Figure 2B). The direct sequencing of the LTR and its flanking regions confirmed that Tre-mediated recombination of loxLTR sequences had occurred in a highly accurate fashion, exactly maintaining the 34 bp loxLTR Tre-specific target sequence (Figure 2D). Importantly, the time-based occurrence and intensity of the excised proviral DNA closely corresponded with the declining BFP signal (Figure 1C and Figure S2 in Text S1), and coincided with temporal Tre expression (Figure 1D). Second, we only detected the remaining single LTR (i.e. genomic scar) in Tre-treated cells by PCR [30] (Figure 2C). Third, we identified a total of three proviral integration sites in genomic DNA from transduced HeLa-smurf cells by nrLAM PCR and high throughput sequencing [31] (Table S1 and Figure S3–Figure S5 in Text S1). As expected, proviral coding sequences (LTR/blb), as opposed to genomic sequences (LTR/int), were under-represented in Tre-treated cells compared to HeLa-smurf cells treated with the control vector as shown for one specific integration site by quantitative PCR (Figure 2E–G) as well as by integration site-independent next generation sequencing (Table 1). Thus, LV-Tre represents an efficient HIV-1 specific expression system for excising integrated HIV-1 provirus from cells.
The expression of antiviral genes may induce undesired effects, which could compromise host cell function. To analyze potential Tre-related cytotoxicities, Tre was overexpressed for a period of 15 weeks in Jurkat T cells; i.e. using the constructs LV-cTre and LV-cCtr in which the 2TAR promoter was replaced by the cellular EF1α promoter, permitting constitutive transgene (Tre) expression. Analyses of cellular metabolic activity (measured by an MTT assay), cell cycle progression (determined by DNA staining with propidium iodide) and apoptosis (assayed by Annexin V staining) did not reveal any deleterious effect of Tre expression on the host cells (Figure 3A–3D, respectively). This was also reflected by comparable cellular growth curves, independently of whether Tre-recombinase was expressed (Figure S6 in Text S1).
We also investigated possible effects of Tre expression on hematopoiesis and immune function. We transduced primary human CD4+ T lymphocytes with LV constitutively overexpressing the GFP marker protein and either Tre or the negative control. Subsequently, the GFP-positive cells (∼90% of the cultures) were analyzed with respect to immune activation by FACS, multiplex cytokine-ELISA and IL4 and IFNγ Elispot assays (Figure 3E–3G, respectively). These analyses suggest that prolonged overexpression of Tre does not negatively affect cellular activation of primary lymphocytes. We furthermore investigated the capacity of Tre vector-transduced CD34+ HSC to differentiate into various hematopoietic lineages using colony forming unit (CFU) assays. In all experiments, Tre vector-transduced HSC maintained their capacity to differentiate into the expected lineages (Figure 3H and 3I), with no significant differences from the controls.
For further analyses, Tre was constitutively overexpressed for 21 days in primary human CD4+ T cells and potential chromosomal alterations were subsequently analyzed by spectral karyotyping (SKY) [32] and array-comparative genomic hybridization (array-CGH) [33]. The combination of these cytogenetic assays exhibited neither chromosomal translocations, nor variations in DNA copy-number changes (Figure 4A and 4B and Figure S7 in Text S1). Thus, Tre appears to lack any obvious off-target activity in human cells, at least within the limitations of these experimental systems.
Next we investigated other potential undesired alterations due to Tre expression. A recent transcriptome analysis using whole human genome microarrays found no differences between Tre-treated and untreated control cells [34], indicating that the expression of Tre does not alter the expression profile of cells. To verify Tre's target specificity, the SeLOX algorithm [35], a locus of recombination site search tool, was employed to scan the human genome for potential Tre target sites. Seven independent sequences that occur in the human genome and display the highest sequence similarity to loxLTR (5–8 nucleotide mismatches in the specificity-determining loxLTR halfsites) were tested as sites for Tre-mediated recombination in E. coli and HeLa cells (Figure 5). No recombination (above background signal) was detectable after strong and prolonged expression of Tre in E. coli or HeLa, respectively, demonstrating that naturally occurring human loxLTR-like sites are not a substrate for the recombinase. However, by screening the Los Alamos HIV sequence database (http://www.hiv.lanl.gov/), two additional independent clinical HIV-1 isolates were identified with subtle single nucleotide loxLTR variations that served as Tre targets (Method S1 and Figure S8 in Text S1). These findings suggest that Tre does not alter or recombine human chromosomal sequences (Figure 4 and Figure 5) but may recognize certain s\ingle nucleotide alterations in the loxLTR site of a variety of clinical HIV-1 isolates.
In conclusion, the combined data indicate that LV-mediated expression of Tre-recombinase does not induce cytopathic effects in human hematolymphoid cells.
Two approaches, suggested as gene therapies against HIV [23], [24], were taken to test the ability of Tre to suppress HIV-1 infection in vivo. In the first approach, human CD4+ T cells were isolated from buffy-coats and transduced with LV-Tre or LV-Ctr particles, routinely resulting in ∼60% GFP+ cells (as measured by FACS; not shown). Then 6-week-old Rag2−/−γc−/− mice were conditioned with clodronate, irradiated and transplanted with 3×106 cells of the transduced total cell pools, which were characterized by CCR5 surface expression (Figure S9 in Text S1). Rag2−/−γc−/− animals lack B, T, and NK cells, can be engrafted with either CD4+ T cells or CD34+ HSC, and in both cases, support HIV-1 infection [36]–[39].
The engraftment of human lymphocytes was verified at 8 to 10 weeks post transplantation by FACS analysis of PBMCs, determining the percentage of mouse CD45+, human CD45+, human CD4+, and GFP+ cells. To assess Tre activity, mice with ≥1% human CD45+CD4+GFP+ peripheral cells were then infected by intra-peritoneal injection of 100 ng p24 antigen of the replication-competent CCR5-tropic HIV-1 pNLT2env(BaL)mcherry, bearing loxLTR sites derived from the primary HIV-1 isolate TZB0003 [21]. Since we wanted to monitor Tre-mediated antiviral effects as directly as possible, i.e. by analyzing plasma viremia, this challenge virus does intentionally not encode a functional Nef protein. It is known that intact Nef depletes CD4+ T cells in humanized mouse models, thereby also indirectly affecting viral loads [40], [41]. Therefore, Nef-mediated pathogenic effects were not addressed in the following in vivo analyses.
Upon infection, viral loads, GFP+ cells and PBMC surface markers (see Materials and Methods section) were subsequently monitored over time. Inspection of the data obtained from the individual animals revealed suppression of HIV-1 viremia in the plasma of the mice transplanted with Tre-transduced cells (LV-Tre; animal T1–11), but not in the negative control animals (LV-Ctr; animal T12–18; Figure 6). Moreover, in contrast to LV-Ctr treated animals, the percentage of human CD45+CD4+ cells increased during the 16 week observation period in the mice that received Tre-transduced CD4+ T cells (Figure 6). It should be noted that the engraftment of immune deficient mice with human cells is to a significant extent donor dependent, a fact that impacts on the animals' infection rates. Therefore, to obtain statistics, the HIV-1 RNA copies in the mice at week 2 post infection were set to 100% and the fold difference of the change in these baseline levels was followed over time. This analysis revealed a highly significant reduction of the viral load (p = <0.0001, n = 11; for statistical method see Materials and Methods section) in mice transplanted with Tre-transduced cells, as opposed to animals that received cells transduced with the negative control vector (p = 0.0811, n = 7) (Figure 7A). Moreover, the increase in the percentage of human CD45+CD4+ cells, particularly in Tre-treated mice as opposed to control animals was clearly significant (Figure 7B). This trend was also seen when assessing the percentage of CD4+GFP+ cells (Figure 7C). At 16 weeks after infection, animals were euthanized and various tissues analyzed. Immunohistochemistry on spleen sections clearly revealed reduced numbers of HIV-infected cells in a Tre-treated as compared to a control animal (Figure 7D). FACS analyses of single cell suspensions derived from bone marrow, liver, spleen and thymus showed an enrichment of CD4+GFP+ cells, particularly in animals that had received Tre-transduced cells (Figure S10 in Text S1).
In a second in vivo approach to study Tre-based antiviral effects, newborn Rag2−/−γc−/− mice were irradiated and transplanted by intrahepatic injection with 3×105 LV-Tre and LV-Ctr transduced human CD34+ HSC. The transduction rate of these cord blood-derived hematopoietic cells typically resulted in ∼30% GFP+ cells (not shown). Engraftment was verified at 10 to 12 weeks post transplantation by FACS analysis of PBMCs (determining the percentage of mouse CD45+, human CD45+, human CD19+, human CD3+, human CD4+ and GFP+ cells). Generally, animals with ≥0.5% of CD45+CD4+GFP+ lymphocytes in their peripheral blood were challenged with HIV-1 as described above. As shown, in all LV-Tre treated animals (HSC1–10), as opposed to mice that received LV-Ctr transduced HSC (HSC11–18), the individual viral load declined over time and the percentage of human CD45+CD4+ cells either increased or remained constant (Figure 8). Subsequent statistical analysis at week 12 post infection revealed that mean viremia in mice that received LV-Tre transduced HSC was significantly diminished (p = <0.0001, n = 10) compared to control (i.e. LV-Ctr) animals (p = 0.5377, n = 8) (Figure 9A), indicating a progressive loss of viral loads in Tre-treated animals over time. In contrast to the human CD4+ T cell transplanted animals, the percentage of CD45+CD4+ T cells did not change significantly in these mice (Figure 9B). This may be explained by the fact that only a fraction of the transplanted HSC (∼30%) were Tre-transduced, and thus protected from HIV replication. CD4+GFP+ cells were detected over the entire 12-week period, demonstrating the successful development of LV-transduced peripheral T cells in these HSC transplanted mice (Figure 9C). The immunohistochemical analysis of human CD3+ and HIV-1 p24 antigen expressing cells in lymph nodes of representative euthanized mice demonstrated that p24+ cells were distinctly depleted in mice that received LV-Tre transduced HSC (Figure 9D). Furthermore, FACS analysis of cell suspensions derived from bone marrow or spleen of these mice verified the presence of transgenic human cells representing various hematopoietic lineages, including CD4+ and CD8+ T lymphocytes, pre-B and activated B cells, cells committed to the monocyte/macrophage lineage, NK cells and NKT cells (see Figure S11 in Text S1).
Collectively, the in vivo experiments document the antiviral activity of Tre-recombinase at the organismal level.
The clinical development of HAART has been one of the great successes in modern medicine. However, the fact that HAART cannot eradicate HIV [7], [8] makes investigating novel antiviral strategies a prerequisite for developing a future cure for HIV infection [7], [9], [12]. In effect, gene therapy strategies represent a technology holding high promise for future antiviral disease treatments [22]–[25]. Indeed, various RNA-based technologies are currently being investigated in vivo, including, for example, the expression of RNA aptamers, siRNAs and shRNAs, TAR decoys, and ribozymes [42]–[44]. Moreover, the expression of membrane-bound fusion inhibitors is another appealing antiviral strategy [45], [46]. These approaches efficiently suppress virus replication, and thus reduce viral loads for extended periods of time. Another promising strategy appears to be disruption of the CCR5 gene [47], [48], for example by expressing engineered zinc finger nucleases (ZFN) [20]. In humanized mice transplanted with either CD4+ T cells or CD34+ HSC, ZFN-mediated CCR5 disruption has been shown to confer resistance to de novo infection by CCR5-tropic HIV-1, thereby controlling virus replication [49], [50].
In contrast, an antiviral strategy based on Tre-recombinase is independent of virus coreceptor usage (i.e. tropism) and can target cells that are already infected with HIV [18], [34]. Importantly, Tre-mediated provirus excision allows reversal of HIV infection at the cellular level, thereby avoiding viral cytopathic effects (e.g. effects associated with viral antigen expression) and possibly restoring host cell function. As shown here, Tre expression mediated highly significant antiviral effects, which were equally observed in animals engrafted with Tre-expressing CD4+ T cells or Tre-expressing CD34+ HSC. With respect to future clinical studies, this Tre effect is particularly impressive, since the animals had been transplanted with unselected cell pools, where only a fraction of the T cells (∼60%) or the HSC (∼30%) harbored the Tre expressing lentiviral vector. Apparently, Tre-mediated protection of only a subpopulation of HIV-1 target cells suffices to achieve significant antiviral effects in vivo. This may be explained by the in vivo selection of gene vector-transduced cells as well as potential bystander effects [22]–[25].
The results presented here suggest that Tre vectors are promising antiviral reagents for therapies based on the genetic modification of both peripheral T cells and hematopoietic stem cells. Clearly, the procedure for ex vivo treatment of peripheral T cells is comparably less complicated, and aims at a functional cure by achieving long-term control of HIV, preferably in the absence of HAART [7]. In contrast, the development of a sterilizing cure that eradicates HIV-1 from an infected organism, if achievable at all, most likely requires a highly complex strategy, involving the autologous transplantation of gene-modified HSC [22], [25]. It is then expected that the elimination of all HIV-infected cells may eventually depend on reconstituting the patient's immune functions, a process that presumably requires additional and potentially gene therapy-unrelated approaches such as, for example, immune activation and/or purging strategies [51]–[54]. It is likely that such multi-pronged eradication approaches will benefit from Tre-mediated provirus excision in the patient's immune effector cells (e.g. CD4+ T cells).
Important safety issues related to gene therapies are generally connected with potential cytopathic effects caused by the respective transgene and/or the vector technology used. The latter was addressed here by using an advanced lentiviral SIN vector design where transgene expression was placed under the control of a Tat-inducible promoter, limiting its expression to HIV-infected cells. This strategy circumvents a major shortcoming of various antiviral gene therapies, the continued expression of foreign transgenes [23], thereby minimizing undesired transgene-related side effects such as immunogenicity.
Obviously, expression of Tre-recombinase from a Tat-inducible promoter presumably precludes provirus excision in latently infected cells. It is therefore conceived that in clinical virus eradication approaches Tat-responsive Tre-expressing vectors will only be used in combination with purging drugs that, as previously shown, not only specifically activate the transcription of otherwise quiescent proviral genomes [55], [56], but, at the same time, will also enable Tat-mediated Tre expression from the vector used in the present study. In this context it is also important to note that a recent study demonstrated that drug-induced purging alone does not result in the elimination of patient-derived infected resting CD4+ T cells, even when autologous CTLs were present [57]. In fact, after virus reactivation these cells where only killed when the HIV-specific CTLs were pre-stimulated, suggesting that virus eradication depends at least on a combination of purging drugs with therapeutic vaccination strategies [57]. It is expected that such an approach would further benefit from the inclusion of an additional anti-HIV gene therapy [25], [58], such as Tre-mediated provirus excision [22]. It is also noted that recent computational modeling of HIV dynamics in the presence of a replication incompetent Tre-recombinase-expressing therapeutic vector suggested that such an approach may indeed clear all HIV from the system in the long term [59].
Clearly, our study does not investigate the efficacy of Tre for latent proviruses. This is a significant limitation that will be addressed in the future. Particularly, it will be of interest to see whether a residual Tat level exists in latently infected cells that enables Tre expression by the current vector design. Alternatively, Tat-independent vectors that employ drug-inducible promoters may permit conditional Tre expression in resting cells. For example, advanced doxycycline-responsive promoter systems hold the promise to further increase biosafety of gene therapies by actively controlling transgene expression [60]. In this context one may also conceive the direct delivery of Tre-recombinase into patients, for example by applying Tre-containing virus like particles [61]. It is noted, that excision of proviral DNA by recombinant cell permeable Tre-recombinase has been already demonstrated in cell cultures [34]. Thus, such advanced Tre delivery systems are conceived to play an important role in the future, particularly for targeting latently infected resting cells.
Another safety aspect that should not be underestimated is based on Tre's pronounced target site specificity. The fact that the site-specific recombination process mediated by such Cre-derived enzymes neither produces free DNA ends (e.g. double-strand breaks) nor requires additional host factors [62], minimizes the oncogenic risk. In agreement, the advanced molecular cytogenetic analyses presented here demonstrate the absence of Tre-related genome-wide off-target effects. This distinguishes Tre-recombinase from CCR5-specific ZFNs, which may suffer from off-target cleavage specificities [49], [63], [64]. Nevertheless, virus entry inhibition by CCR5 knockout represents a highly attractive antiviral strategy that may be exploited to its full extent when combined with Tre-recombinase technology, thereby not only blocking de novo infection but also targeting already infected cells for provirus excision.
The presented data suggest that antiviral gene therapies are feasible using conditionally expressed, engineered Tre-recombinases that precisely remove HIV-1 proviral DNA without cytopathic effects. Antiviral in vivo activity was observed by transduction of both CD4+ T cells and CD34+ HSC. Particularly the latter stem cell-based approach may be a valuable component of future eradication strategies to cure HIV [22], [25]. The fact that the current Tre-recombinase recognizes particularly HIV-1 subtype A isolates may limit its broad application. However, the recent identification of highly conserved HIV-1 LTR sequences [65] in combination with a novel loxLTR search tool [35] now permits the engineering of advanced Tre-recombinases with activity against the majority of HIV-1 variants.
Clearly, it is not expected that HIV-1 can be eradicated by Tre activity alone. As outlined above, future HIV eradication strategies are conceived to be a combination of various antiviral approaches (e.g. drug-based and gene therapies), host immunity enhancing treatments (e.g. therapeutic vaccination approaches), and purging attempts to overcome latency [51]–[53]. In summary, our data support the notion that Tre-recombinase technology can be a valuable component of such a multi-tiered strategy to treat HIV-infected patients.
The animal experiments were performed according to the guidelines of the German Animal Protection Law. The experimental protocols were reviewed and approved by the relevant German authority, the local ethics commission (Ärztekammer Hamburg; OB-050/07 and WF-010/2011) and the Freie und Hansestadt Hamburg, Behörde für Gesundheit und Verbraucherschutz (Nr.: 63/09 and 23/11).
The lentiviral (HIV-1-based) SIN vector backbone has been described previously [28], [66]. Briefly, the vector comprises self-inactivating long terminal repeats (SIN LTRs) (ΔU3, R, U5), splice donor (SD), splice acceptor (SA), packaging signal (Ψ), central polypurine tract (cPPT), and the Rev response element (RRE). A post-regulatory element derived from woodchuck hepatitis virus (PRE) ensures efficient posttranscriptional RNA processing [67], [68] and a duplicated simian virus 40 (SV40) upstream polyadenylation enhancer element (USE) optimizes termination of transgene transcription [28].
The transgene expression cassette includes either the open reading frame of Tre-recombinase (tre) [18] or a scrambled version, plus sequences encoding the enhanced green fluorescent protein (egfp) [28]. The scrambled version serves as a negative control (Ctr); since all ATG start codons were replaced by TGA stop codons, and all GTG triplets by CCT, the respective mRNA cannot be translated into a protein. The tre and scrambled sequence are under the control of a Tre-resistant variant of the HIV-1 NL4-3 LTR (GenBank accession number M19921), containing one or two TAR elements. The TAR duplication (2TAR) was generated by standard PCR technology using HIV-1 NL4-3 proviral DNA as the template for amplification. Expression of egfp is under the control of the constitutive human phosphoglycerate kinase (PGK) promoter [27], resulting in a dual-promoter vector design.
The proviral construct pNLT2ΔenvBLB was generated by replacing the BlpI×XhoI fragment of construct pNLT2ΔenvPuro [18] with the coding sequence for a fusion protein (BLB) composed of mTag-BFP (Evrogen) and blasticidin-S deaminase [69], both linked by the linker sequence 5′-GCGCTAGGTGCTGCCGCCGGTGGT-3′.
The proviral plasmid pNLT2env(BaL)mCherry, encoding CCR5-tropic replication-competent HIV-1 was constructed by inserting the env gene (2572 bp) derived from plasmid pWT/BaL [70] (NIH AIDS Research & Reference Reagent Program, Cat.No. 11414) into the previously described vector pNLT2ΔenvPuro [18]. Puromycin resistance encoding sequences were replaced by the gene (711 bp) encoding the autofluorescent protein mCherry, which was derived from the plasmid pRSET-mCherry [71] provided by Dr. Roger Y. Tsien, University of California San Diego.
HeLa and 293T cells were cultured at 37°C and 5% CO2 in Dulbecco's modified Eagle medium (DMEM; Biochrom) containing 100 units/ml of penicillin and streptomycin (PenStrep; Biochrom) and 10% fetal calf serum (FCS; Biochrom).
To generate HeLa-smurf cells, 2×105 HeLa cells were infected with pseudotyped pNLT2ΔenvBLB (MOI 1). Cells were sorted for BFP-positive cells 1 and 3 weeks after infection and cultured in DMEM supplemented with 100 units/ml of PenStrep, 10% FCS and 5 µg/ml blasticidin (Invivogen).
HIV-1 pseudotypes and lentiviral particles for infecting cultured cells were produced by transient cotransfection of 2×106 293T cells with the lentiviral or proviral plasmid and the respective packaging plasmids [27] using polyethylenimine (PEI) as a transfection reagent according to the manufacturer's protocol (Polysciences, Inc.). In detail, 6 µg of pNLT2ΔenvBLB and 1.5 µg of pCMV-VSV-G [72] were used for transfecting the proviral construct, or 6 µg of lentiviral vector, 1.5 µg of pRSV-Rev [27], 1.5 µg of pCMV-VSV-G [72] and 3 µg of pMDLg/pRRE [27]. At 72 hours post transfection viral supernatants were collected and passed through 0.2 µm pore size filters to ensure removal of any viral aggregates.
Titers of lentiviral particles and HIV-1 pseudotypes were determined as fluorescent forming units per ml (ffu/ml). This involved infecting 5×104 293T cells with different volumes of viral supernatant as described below. At 72 hours post transduction cells were harvested and analyzed by flow cytometry for GFP or BFP expression. Samples that contained 5 to 25% GFP or BFP positive cells were used to calculate viral titers.
Replication-competent HIV-1 for infecting humanized mice was produced essentially as before by transfecting 2×106 293T cells with 6 µg of the HIV-1 plasmid pNLT2env(BaL)mCherry. At day 3 post transfection, virus-containing supernatants were passed through 0.2 µm pore size filters, concentrated using a Centricon Plus-70 device (Millipore Corp), and adjusted with RPMI culture medium (without supplements) to 1 ng/µl of p24 antigen.
Cells were infected with various amounts of virus in the presence of 1 µg/ml polybrene (Sigma-Aldrich) and spinoculated at 300× g for 10 min at ambient temperature. After spinoculation cells were cultivated at 37°C and 5% CO2. Medium was changed 8 h post infection.
For transduction of primary CD4+ T cells, cultures were pre-stimulated with CD3/CD28 magnetic beads (Invitrogen) for 24 h according to the manufacturer's instructions. After prestimulation, various amounts of virus were added in the presence of 2 µg/ml polybrene (Sigma-Aldrich) and the cells were spinoculated as described above. After 24 h of incubation at 37°C and 5% CO2, the transduction procedure was repeated. Prior to further analyses, transduced cells were cultured in the presence of 500 IU IL-2 for a further 3 days at 37°C and 5% CO2.
For transduction of CD34+ HSC, cultures were prestimulated with the cytokine cocktail StemSpan CC110 (Stem Cell Technologies) for 24 h. Virus was added to the cells, which were maintained in StemSpan SFEM (Stem Cell Technologies) supplemented with cytokine cocktail CC110 and the cultures were subjected to spinoculation as described before. After 24 h of incubation at 37°C and 5% CO2, the transduction procedure was repeated.
Total protein was prepared and Western blot analysis was performed as described previously [73]. Rabbit polyclonal anti-Tre serum (Davids Biotechnologie), mouse anti-β-Tubulin (Sigma-Aldrich), polyclonal chicken anti-GFP (Novusbio), or polyclonal rabbit anti-GAPDH (FL-335; Santa Cruz) antibodies were used. Protein signals were quantified using an Odyssey Infrared Imaging System (LI-COR).
Total cellular RNA was prepared, reverse transcribed into cDNA and quantified by quantitative PCR as described previously [74].
To quantify glyceraldehyde-3-phosphate dehydrogenase (GAPDH) sequences the following primers were used: forward, 5′-GTCATCA ATGGAAATCCCATCA-3′; reverse, 5′-TGGTTCACACCCATGACGAA-3′; probe, 5′-(FAM)-TCTTCCAGGAGCGAGATCCCTC-(TAMRA)-3′.
24 hours before transduction, the HeLa-smurf cells' medium was changed to medium without blasticidin. Subsequently, 2×105 HeLa-smurf cells were infected with VSV-G pseudotyped LV-Tre or LV-Ctr (MOI 7.5) as described above.
From infected cultures genomic DNA, protein, and RNA were prepared at different time points (24–336 h) after transduction and analyzed for genomic gag-levels, Tre expression, GFP expression, occurrence of the circular recombination product, and the genomic scar, as described above and below.
In addition, expression of BFP and GFP was monitored by FACS analysis using a FACSCanto II (Becton Dickinson) system equipped with 405, 488 and 635 nm lasers.
Genomic DNA was isolated from eukaryotic cells using the QIAamp DNA Blood Mini Kit (QIAGEN GmbH). The circular recombination product generated by Tre-mediated recombination was detected as follows: 1 µg of genomic DNA was analyzed by PCR using 5′ Mastermix (5 Prime) with forward primer P2 (5′-GCTGCCCTCTGGTTATGTGTG-3′), binding in the blb sequence, and reverse primer P1 (5′-CTTAATACCGACGCTCTCGCAC-3′), binding in the gag sequence of pNLT2ΔenvBLB (PCR conditions: 1 cycle: 94°C for 2 min/56°C for 2 min/72°C for 2 min – 40 cycles: 94°C for 2 min/58°C for 1.5 min/72°C for 2 min - 1 cycle: 72°C for 10 min).
To detect the genomic scar, proviral integration sites were determined by HiLo-PCR [30]. In addition to the original protocol, a “nested” HiLo-PCR was performed with an aliquot of the first HiLo reaction to improve the yield of specific integration site fragments. The reaction conditions for the first and nested HiLo PCRs were as follows (50 µl total volume): 25 µl of Maxima Hot Start Green PCR Master Mix (2×) (Fermentas), and 1 pmol of HiLo primer or nested HiLo primer. HiLo PCR was carried out with 1 µg of genomic DNA from cell lines at the following conditions: 1 cycle: 95°C for 5 min – 25 cycles: 94°C for 1 min/65°C for 1 min/72°C for 3 min - 25 cycles: 94°C for 30 sec/37°C for 30 sec/72°C for 2 min – 1 cycle: 72°C for 5 min.
The nested HiLo PCR was carried out with 0.1 µl of the HiLo reaction under the following conditions: 1 cycle: 95°C for 5 min - 25 cycles: 94°C for 30 sec/65°C for 30 sec/72°C for 3 min - 15 cycles: 94°C for 20 sec/37°C for 30 sec/72°C for 2 min – 1 cycle: 72°C for 5 min. HiLo primers used in this study were: 5′-GAAATGCTAGGC GGCTGTCAAACCTCCACTCTA-3′ to amplify fragments upstream the HIV 5′-LTR, and 5′-TAGAGTGGAGGTTTGACAGCCGCCTAGCATTTC-3′ for fragments downstream of the HIV 3′-LTR. For the nested HiLo PCR the following primers were used: 5′-AGCACCATCCAAAGGTCAGT-3′ to amplify fragments upstream the HIV 5′-LTR, and 5′-AAGTAGTGTGTGCCCGTCTGTTG-3′ to amplify junction fragments downstream of the HIV 3′-LTR.
To enrich the number of genomic DNA/LTR junctions and proviral genome/LTR junctions, nrLAM-PCR was performed as described previously [31]. Briefly, 0.5 µg of genomic DNA was used as a template for linear amplification using 5′-biotinylated LTR-specific primer LTRI (5′-GATATCTGACCCCTGGCCCTG-3′). Biotinylated linear PCR products were immobilized on streptavidine-conjugated magnetic beads (Dynal-Invitrogen). Afterwards, a 5′-phosphorylated and 3′-modified (dideoxycytidine, ddC) linker-cassette ssDNAlinker (5′-CCTAACTGCTGTGCCACTGAATTCAGATCTCCCG GGTC-3′) was ligated to the 3′-end of the linear amplification product. Subsequently, the linear amplification product was amplified using two sets of nested primers. The first round of exponential amplification used 5′-biotinylated primer LTRII (5′-GTGTGTAGTTCTGCCAATC-3′) and primer LCI (5′-GACCCGGGAGATCTGAATTC-3′). Biotinylated double-stranded PCR products were immobilized on streptavidine-conjugated magnetic beads as before, and non-biotinylated complementary strands were eluted as substrate for further reaction. The second round of amplification was performed with primer LTRIII (5′-AGGGAAGTAGCCTTGTGTGTG-3′) and primer LCII (5′-GATCTGAATTCAGTGGCACAG-3′).
nrLAM-PCR products were then used as a template for quantitative PCR to determine the number of provirus/LTR and Chr11q13/LTR junctions. For quantification, SYBR green fluorescence was measured using the following sets of primers: 5′-CATGGAGCAATCACAAGTAGC-3′ and 5′-GTGGCTAAGATCTA CAGCTG-3′ (provirus/LTR junction); 5′-TTTAGTAGAGACAGGGTTTCACCATG-3′ and 5′-AGGGAAGTAGCCTTGTGTGTG-3′ (Chr11q13/LTR junction).
Semi-quantitative analysis was performed with the same set of primers under the following PCR conditions: 1 cycle: 98°C for 1 min – 19 cycles: 98°C for 10 sec/58°C for 30 sec/72°C for 45 sec - 1 cycle: 72°C for 5 min using Phusion Polymerase (Fermentas).
High throughput sequencing and data analysis were carried out as described previously [31]. Briefly, nrLAM-PCR products were amplified with bar-coded PCR primers fused to GS FLX-specific adaptors (for primer sequences see Table S1 in Text S1), pooled and subjected to pyrosequencing on a GS FLX sequencer (Roche), using adaptor primer A. Sequencing reads were sorted according to their multiple sequence identifier (MID) tags and quality filtered to eliminate all reads that did not match long terminal repeat (LTR) sequences at their 5′-end. We identified all reads that extended at least 20 nucleotides [31] beyond the LTR, and after trimming LTR sequences, matched the flanking sequences to both the human genome and pNLT2ΔenvBLB. We noted the number of reads that mapped to integration sites or the blb-encoding sequences of pNLT2ΔenvBLB to generate the data shown in Table 1. The BLAT alignment tool, described in [31], as well as the CLC Genomics Workbench package (CLCbio) were used to map sequence reads, and the UCSC genome browser (http://genome.ucsc.edu/) was used to visualize integration sites (Figure S3–Figure S5 in Text S1).
Isolation of CD4+ T cells from buffy coats was carried out using the RoboSep negative selection human CD4+ T cell enrichment kit in conjunction with a RoboSep automated cell separator (StemcellTechnologies) according to the manufacturer's instructions. Likewise, the preparation of CD34+ HSC from umbilical cord blood was performed with the EasySep human cord blood CD34+ selection kit (StemcellTechnologies) and the RoboSep system.
MTT assay was performed with 100 µl of cell suspension and the MTT kit (Roche) according to the manufacturer's instructions. A VersaMax micro plate reader (Molecular Devices) was used for colorimetric assay evaluation.
Analysis of apoptotic events was performed using the Annexin V FITC kit (Invitrogen) together with the antibody Annexin V-APC conjugate (Becton Dickinson). For analysis, 5×105 cells were harvested, stained according to the manufacturer's protocol, and analyzed using a BD FACSCanto II (Becton Dickinson) system.
To determine cell cycle distribution 1×106 cells were harvested, washed with PBS, suspended in 500 µl PBS/1% EDTA and fixed drop-wise with 5 ml of 80% ice cold ethanol. After incubating for 20 min on ice, the cells were incubated for 24 h at −20°C. Afterwards, the cells were pelleted and rehydrated in 450 µl of PBS supplemented with 16.6 µl RNAse A (10 mg/ml; Sigma-Aldrich) and 33 µl propidium iodide solution (0.5 mg/ml; Sigma-Aldrich). Incubation at 37°C for 30 min, was followed by further incubation at ambient temperature in the dark for 2 h prior to flow cytometry using a BD FACSCanto II system.
Primary human CD4+ T cells were stimulated for 12–24 h with phorbol myristate acetate (PMA) (50 ng/ml final conc.) and ionomycin (0.67 µM final conc.). Specific cytokine levels were monitored by ELISA, Elispot and intracellular cytokine staining (ICS).
Human Th1, Th2, Th17 Cytokine Multi-Analyte ELISArray (Qiagen) was performed with supernatants from 1×106 stimulated cells or unstimulated controls, according to the manufacturer's instructions. ICS was essentially performed as previously described [75] with the modification that monensin (Biolegend) was used to inhibit secretion. Mouse anti-human CD3-APC H7 (BD Biosciences), mouse anti-human CD4-APC (Becton Dickinson) and mouse anti-human CD154-PE (BD Pharmingen) antibodies were used for surface staining according to the manufacturer's instructions, except that a 4-fold excess of the CD154 antibody was directly added to the cells during stimulation. For intracellular staining, mouse anti-human IFNγ-PE-Cy7 antibody (BD Pharmingen) was used. Live/dead staining was performed in parallel using the LIVE/DEAD Fixable Aqua Dead Cell Stain Kit for 405 nm excitation (Life Technologies).
Elispot analysis was essentially performed as previously described [76]. Briefly, polyvinylidene plates (96-well; Millipore) were coated with 50 ng of recombinant anti-human IFNγ antibody (Mabtech), or 50 ng of recombinant anti-human IL4 antibody (Mabtech) in phosphate-buffered saline at 4°C for 12 h. Afterwards, 3×103 to 1×105 cells were seeded on the coated plates and stimulated with PMA/ionomycin as indicated above. Secreted IL4 or IFNγ was detected using the biotinylated detection antibodies anti-human IL4 (Mabtech) or anti-human IFNγ (Mabtech).
The differentiation potential of transduced HSC cells was performed with methocult H4435-enriched methylcellulose (Stem Cell Technologies) according to manufacturer's protocol. For this, 100 transduced or mock treated cells were suspended in 1 ml of methylcellulose and seeded into a 3.5 cm diameter cell culture dish (Stemcell Technologies). After incubation at 37°C and 5% CO2 for 14 days, various cell colonies were identified and counted.
Tre overexpressing primary human CD4+ T cells were arrested in mitosis 21 days post transduction by treating the cells with 0.1 µg/ml colcemid for 4 hours. Cells were then treated with 75 mM KCl, incubated at 37°C for 15 min and fixed in 75% methanol/25% acetic acid. Cell suspension was dropped onto glass slides. Metaphase chromosomes were hybridized with the SKY probe mixture and analysed as previously described [77] using the SpectraCube system (Applied Spectral Imaging) coupled to an epifluorescence microscope (Leica).
Lentiviral transduced Tre expressing primary human CD4+ T cells were harvested 21 days post transduction and genomic DNA was extracted using the QIAamp DNA Blood Mini kit (Qiagen) for array-CGH analysis. DNA was hybridized against DNA from mock-transfected cells on an Agilent SurePrint G3 Human CGH Microarray Kit 2×400K. The minimum number of probes affected to designate an aberration was set to 3. The median over all probe spacing was 5.3 kb (4.6 kb in RefSeq genes) on the array used.
Potential off-target Tre recombination sites were identified by screening the human genome using the bioinformatics tool, SeLOX [35]. The respective genomic sites were cloned into the recombination reporter plasmid pEVO-Tre-target [18]. In E. coli, recombinase expression was induced with L-arabinose (Sigma-Aldrich) at 1 mg/ml. Plasmid DNA was isolated from overnight cultures and digested with BsrGI and XbaI (NEB), resulting in different fragment sizes for recombined versus non-recombined substrate on agarose gels. Recombination on the Tre target loxLTR served as positive control. In eukaryotic cell culture, HeLa cells were cotransfected with the reporter plasmids and the expression plasmid pIRESneo-Tre [18]. DNA was isolated from the cells 48 h post transfection and analyzed for recombination by polymerase chain reaction using the primers F: 5′- GACAATAACCCTGATAAATGC-3′, and R: 5′-CCTTAAACGCCTGGTGCTAC-3′.
Humanized Balb/c Rag2−/−γc−/− (provided by M. Ito, Central Institute for Experimental Animals, Kawasaki, Japan) were bred and maintained under specific pathogen-free conditions using individually ventilated cages (IVC).
To generate human T cell transplanted Rag2−/−γc−/− mice, 6 week old animals were preconditioned by intra-peritoneal (i.p.) injection of 100 µl of clodronate liposomes (obtained from Dr. N. van Rooijen, Department of Molecular Cell Biology, Amsterdam, Netherlands). Twenty four hours later, animals were irradiated using a dose of 2×2 Gy (6 h and 2 h before transplantation) from a Cesium 137 source at 3.75 Gy/min (CSL-12; Conservatome). Subsequently, mice were transplanted with 3×106 lentiviral vector (LV-Tre or LV-Ctr) transduced human CD4+ T cells in 150 µl PBS containing 0.1% human AB serum (PAN Biotech GmbH) by i.p. injection. Analysis of human cell engraftment was verified by FACS analysis of peripheral blood samples at 8 to 10 weeks post transplantation, using retro-orbital sampling. Likewise, following HIV-1 infection blood samples were analyzed every second to fourth week for a period of 4 months.
Animals transplanted with human hematopoietic stem/progenitor cells (CD34+ HSC) were generated by injecting newborn Rag2−/−γc−/− mice 24 h after birth intra-hepatically (i.h.) with 3×105 lentiviral vector (LV-Tre or LV-Ctr) transduced CD34+ cells in 30 µl PBS containing 0.1% human AB serum. Prior to i.h. injection, the newborns were irradiated with 2×2 Gy as before. Engraftment was verified by FACS analysis of peripheral blood samples at 10 to 12 weeks post transplantation and, following HIV-1 infection, every second to third week for a period of 3 months.
CD4+ T cell or CD34+ HSC transplanted mice were infected by i.p. injection of 100 ng p24 antigen (108 HIV-1 RNA copies) of CCR5-tropic HIV-1 pNLT2env(BaL)mcherry, containing loxLTR Tre-recombinase target sites. Animals were bled from the retro-orbital venous sinus two weeks post infection, followed by collection of blood every second, third or fourth week. Viremia was assayed by diluting cell-free mouse plasma with human serum (PAN Biotech GmbH) using the ultrasensitive (<20 HIV-1 RNA copies/ml) Cobas AmpliPrep/Cobas TaqMan HIV-1 Test version 2.0 (Hoffmann-La Roche Ltd.).
For analysis of peripheral cells, 50 µl to 100 µl of blood was collected from the retro-orbital venous sinus (r.o.) into 100 µl bleeding-buffer (PBS plus 10 mM EDTA) and red blood cells were lysed by treatment with Red Blood Cell Lysing Buffer (Sigma-Aldrich). The white blood cell pellet was resuspended in FACS-buffer (PBS containing 2% FCS and 2 mM EDTA) and stained with monoclonal antibodies.
Single cell suspensions of various organs (thymus, spleen, liver and bone marrow) for antibody staining and FACS analysis were prepared at necropsy by manual tissue dissection and filtering through a sterile 70 µm nylon mesh (BD Biosciences).
Stained cells were analyzed in FACS-buffer plus 1% paraformaldehyd using a FACSCanto II (Becton Dickinson) system with BD FACSDiva Software v5.0.3 and FlowJo software v7/9 for PC (Treestar). To monitor human cell engraftment, r.o. collected cells were stained with monoclonal antibodies raised against mouse CD45.2 (104) and human CD45 (H130), human CD3 (UCHT1), and human CD19 (HIB19) (all from eBioscience Inc.). The transduction rate was monitored by vector-derived GFP expression. HIV-1 infected mice were analyzed by staining with monoclonal antibodies directed against the human antigens CD45 (H130), CD4 (RPA-T4) (both from eBioscience Inc.), CD3 (UCHT1), CD8 (B9.11) (both from Beckman Coulter Inc.), CCR5 (3A9) and CXCR4 (12G5) (both from BD Pharmingen). Isotype antibodies and cells obtained from non-transplanted mice served as negative staining controls.
Formalin fixed and wax embedded sections were analyzed. Deparaffinized sections were incubated in citrate buffer in an 85°C waterbath overnight for human CD3 antigen detection. Monoclonal anti-human CD3 (Dako M7254, clone F7.2.38) was used in a 1∶1000 dilution. Biotinylated anti-mouse monoclonal antibody in combination with horseradish peroxidase streptavidin was used for visualisation. The TNB-Amplification Kit (Dako) and diaminobenzidine were used as substrates. Sections were counterstained with haemalumn.
To visualize p24 antigen, the monoclonal antibody clone Kal-1 (Dako) was used. Deparaffinized sections were boiled for 20 min in retrieval buffer S 1699 (Dako) using a pressure cooker set at 100°C. Streptavidin alkaline phosphatase and the TNB-Amplification Kit with Permanent Red were used for visualisation.
For staining mesenteric lymph nodes, detected in HSC transplanted Rag2−/−γc−/− mice, sections were incubated in titrated concentrations of mouse monoclonal anti-HIV p24 (Kal-1; Dako) and anti-human CD3 (SP7; Thermo Scientific) antibodies using an automated Ventana Discovery Module (Ventana Medical Systems). Stainings were developed according to the manufacturer's protocol as described previously [38].
Significant values between the initial analysis (at week 2 post HIV-1 infection) and the final analysis (at week 12 or 16 post HIV-1 infection) within the LV-Tre animal group and the LV-Ctr animal group were calculated using the Student's paired two-tailed t-test of the GraphPad Prism Program version 5.03 (GraphPad Software). The two-tailed p values less than 0.05 were considered significant.
|
10.1371/journal.ppat.1007992 | NS2B/NS3 mutations enhance the infectivity of genotype I Japanese encephalitis virus in amplifying hosts | Genotype I (GI) virus has replaced genotype III (GIII) virus as the dominant Japanese encephalitis virus (JEV) in the epidemic area of Asia. The mechanism underlying the genotype replacement remains unclear. Therefore, we focused our current study on investigating the roles of mosquito vector and amplifying host(s) in JEV genotype replacement by comparing the replication ability of GI and GIII viruses. GI and GIII viruses had similar infection rates and replicated to similar viral titers after blood meal feedings in Culex tritaeniorhynchus. However, GI virus yielded a higher viral titer in amplifying host-derived cells, especially at an elevated temperature, and produced an earlier and higher viremia in experimentally inoculated pigs, ducklings, and young chickens. Subsequently we identified the amplification advantage of viral genetic determinants from GI viruses by utilizing chimeric and recombinant JEVs (rJEVs). Compared to the recombinant GIII virus (rGIII virus), we observed that both the recombinant GI virus and the chimeric rJEVs encoding GI virus-derived NS1-3 genes supported higher replication ability in amplifying hosts. The replication advantage of the chimeric rJEVs was lost after introduction of a single substitution from a GIII viral mutation (NS2B-L99V, NS3-S78A, or NS3-D177E). In addition, the gain-of-function assay further elucidated that rGIII virus encoding GI virus NS2B-V99L/NS3-A78S/E177E substitutions re-gained the enhanced replication ability. Thus, we conclude that the replication advantage of GI virus in pigs and poultry is the result of three critical NS2B/NS3 substitutions. This may lead to more efficient transmission of GI virus than GIII virus in the amplifying host-mosquito cycle.
| Flaviviral vertebrate amplifying host(s), invertebrate vector(s), genetics, and environmental factors shape the viral geographical distribution and epidemic disease pattern. Newly emerging dengue virus genotypes, West Nile virus clades, or Zika virus strains exhibited an enhancement in mosquito vector competence. However, hosts and viral determinants responsible for the occurrence of JEV genotype replacement remains unclear. Here, we demonstrated that emerging GI viruses with enhanced transmission potential in amplifying hosts such as pigs and avian species was encoded by three critical GI-specific mutations in NS2B/NS3 proteins. This discovery provides insight into the viral genetic mechanism underlying the GI virus advantage and adaptation in the pig/avian species-mosquito cycle. Our results also emphasize the importance of monitoring viral evolution in amplifying vertebrate hosts to clarify the role of avian species in local transmission of GI virus in JE endemic and epidemic countries.
| Japanese encephalitis virus (JEV), a mosquito-borne flavivirus, has a single-stranded, positive-sense RNA genome encoding three structural proteins (capsid, precursor membrane protein, and envelope protein) and seven non-structural proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5) [1]. JEV was first isolated in 1935. Since then JEV has been classified into five genotypes (GI-GV) with variant geographic distribution in Asia and Australia. Traditionally, GIII virus is the most widely distributed and dominant JEV genotype in JEV endemic/epidemic regions [2]. The ecological factors of GIII endemic or epidemic areas are well characterized, and the virus is maintained primarily in the Culex tritaeniorhynchus-amplifying host (pigs and avian species) transmission cycle, and humans and horses are accidental, dead-end hosts [3–8]. An estimated 67,900 JE human cases occur annually with a 20–30% case-fatality rate, and 30–50% of surviving patients suffer neurological sequelae [9].
GI virus emerged in the 1990s and has gradually replaced GIII virus as the most frequently isolated JEV genotype from Culex tritaeniorhynchus, stillborn piglets, and JE patients in Japan, Korea, Vietnam, Thailand, Taiwan, China, and India [10–15]. We suggest that GI virus might compete with GIII virus in the same pig-mosquito cycle and exhibit a transmission advantage over areas previously dominated by GIII virus [11, 16]. Emerging GI virus replicates more efficiently in Aedes albopictus mosquito-derived cells, birds, and ducklings than GIII virus [17–19]; however, it remains unknown if the replication efficiency of GI vs. GIII virus occurs in pigs and/or Culex tritaeniorhynchus mosquitoes, which play critical roles in local transmission of JEV. In addition, the experimental evidence associated with the difference in viral genomic factor(s) does not fully support the occurrence of GIII replacement during the past 20 years [11, 18, 20].
In this study, we investigated the in vitro and in vivo replication characteristics of emerging GI virus and previously dominate GIII virus in Culex tritaeniorhynchus mosquito vector and amplifying hosts (pigs, young chickens, and ducklings). We did not include dead-end hosts in this study because of the remote possibility of the involvement of dead-end hosts in the transmission of JEV. More importantly, we constructed, generated and applied various genotypic chimeras of recombinant JEVs from infectious cDNA clones to determine the viral genetic determinants that enhanced and increased the replication efficiency of GI virus over GIII virus. These genetic determinants would be critically important for selection of viral hosts and genes to monitor GI virus activity and evolution in a natural transmission cycle.
To analyze the growth curve of GIII and GI JEVs in mosquito-derived cells, we infected Aedes albopictus-derived C6/36 cells and Culex tritaeniorhynchus-derived CTR209 cells with GIII viruses (CH1392 and T1P1 strains) and GI viruses (YL2009-4 and TC2009-1 strains) at 28°C (Fig 1A and 1B). The growth curves and virus titers from GI and GIII-infected C6/36 cells were similar with no statistical significance between them during a 60-hour infection period (Fig 1A). In CTR209 cells, the growth curves were statistically non-significant differences (p>0.05) between genotypes; although, GI YL2009-4 virus was less efficient and generated an average of 1.61–2.21 log lower viral titer than GIII-CH1392 virus (Fig 1B). In addition, comparable ratios of viral NS3 proteins to α-tubulin were detected in GIII and GI virus-infected C6/36 and CTR209 cells at 48 hours post-infection (HPI) (S1A, S1B, S1D and S1E Fig). The newly evolved WNV genotype (WN02) showed an ability to adapt to mosquitoes reared at a higher temperature [21]. Therefore, we further compared GIII to GI viruses in CTR209 cells at an elevated temperature. When the temperature was increased to 34°C, similar titers for both GIII and GI viruses were observed even though higher relative intensity of NS3 proteins from GI viruses was detected in CTR209 cells (Fig 1C, S1C and S1F Fig). This suggested GIII and GI viruses might have different efficiency to generate infectious particle [22] or the stability difference of NS3 protein [23] in infected CTR209 cells. In addition, the viability of GIII and GI virus-infected CTR209 cells was non-significantly different at 28°C and 34°C (S2 Fig).
To analyze viral replication in mosquitoes in vivo, we fed female Culex tritaeniorhynchus mosquitoes with pig blood mixed with 8x106 focus forming units (ffu) of GIII or GI JEVs. We determined the infection rate by detection of viral titer in each mosquito 14 days post-infection (DPI). GIII CH1392 virus and GI TC2009-1 virus had higher infectious rates (58.33% and 66.67%) than GIII T1P1 virus and GI YL2009-4 virus (25.00% and 16.67%) whereas no genotypic differences were observed in infectivity or in viral titer among positive mosquitoes (Fig 2). These in vitro and in vivo results suggest that the Culex tritaeniorhynchus mosquito vector may play a minor role in genotype replacement.
To evaluate the role of amplifying hosts in genotype replacement, we compared the in vitro replicative ability of GIII CH1392 and T1P1 strains to GI YL2009-4 and TC2009-1 strains in pig- and poultry-derived cells (PK-15, DF-1, CER, and DE cells) and primate-derived control cells (VERO cells) at 37°C (Fig 3A–3C, S3A and S3B Fig). A similar growth curve of GIII and GI viruses was observed in VERO cells (Fig 3A); in contrast, GI YL2009-4 and TC2009-1 viruses produced 1–2 -log higher viral titers than GIII CH1392 and T1P1 viruses in PK-15 cells at 24 and 36 HPI (p< 0.05) (Fig 3B). In poultry-derived cells (DF-1, CER, DE cells), two GI viruses and GIII CH1392 virus replicated to similar viral titers but their viral titers were higher than GIII T1P1 virus (Fig 3C, S3A and S3B Fig). Natural amplifying hosts for JEV have a higher body temperature of 40–44°C in avian species and 38–40°C in pigs. In pigs, fever can be as high as 41°C following JEV infection [24]. Therefore, we inoculated JEVs into VERO, PK-15, and poultry-derived cells (DF-1, CER, DE cells) and analyzed viral replication at 41°C. Interestingly, as the temperature increased, the overall viral titers of GI viruses were significantly higher than GIII viruses in amplifying host-derived cells and at the single time point of 36 HPI in VERO cells (p< 0.05) (Fig 3D–3F, S3C and S3D Fig). An apparently higher relative intensity of NS3 proteins was also observed at 48 HPI in GI virus-infected PK-15 and DF-1 cells but not in VERO cells at 41°C (S1G–S1L Fig). However, the ratio of NS3 to β-actin was inconsistent with extracellular infectious particles in the GIII virus-infected DF-1 cells. This suggested two GIII viruses might have different efficiency to generate infectious particle [22] or the stability difference of NS3 protein [23] in infected DF-1 cells.
The higher thermal stability at elevated temperature for GI viruses could be a contributing factor related to enhancement of viral replication of GI viruses. To determine the influence of viral thermal stability, we incubated equal amounts of GIII and GI viruses at 37°C and 41°C, and examined residual viral titers at 3, 6, 9, and 12 hours post-treatment (HPT) (S4 Fig). The viral infectious rates of both genotypes dropped to less than 50% after 3-HPT at 37°C and even more at 41°C (S4A and S4B Fig). There was no statistical difference between the half-life of infectivity of GIII and GI viruses at both 37°C and 41°C (S4C Fig). Besides, there was no genotypic difference on cell viability (VERO, PK-15, and DF-1) after GIII and GI virus infection except lower viability of GI virus-infected PK-15 cells, producing higher viral tier, at 41°C at 60 HPI (p< 0.05) (S2 Fig). These results suggest that the replication advantage of GI virus over GIII virus occurred in an amplifying host but was independent of external viral thermal stability and virus-infected cell viability.
Enhancement of GI virus replication ability in pig- and poultry-derived cells was further investigated in vivo. JEV–infected pigs exhibited either an asymptomatic manifestation or developed fever with sufficiently high viremia to ensure the transmission of virus by engorged mosquitoes [8]. We subcutaneously inoculated 105 ffu of GIII CH1392 virus, GI YL2009-4 virus, and or PBS into ten-week old, specific-pathogen free (SPF) pigs (three SPF pigs per group), and then monitored daily body temperature and viremia. At 3 DPI, three GI virus-infected pigs developed fever averaging 40.6°C while GIII virus-infected and PBS-inoculated pigs maintained normal body temperature until euthanasia at 8 DPI (Fig 4A). No detectable viremia was observed in the infected pigs at 1-DPI. A significantly higher viremia was detected the following day in GI YL2009-4 virus-infected pigs with an average viral titer of 104.6 ffu/ ml compared to GIII CH1392 virus-infected pigs with a viral titer of 102.7 ffu/ml (p< 0.05). However, viremia was undetectable at 4-DPI in all virus- and PBS-inoculated pigs (Fig 4B). The higher viremia in GI virus-inoculated pigs was consistent with the higher RNAemia detected in the GI virus-infected pigs at 2 DPI (p< 0.05) (S5A Fig). Although only a limited number of SPF pigs were used in this study, we found that GI YL2009-4 virus infection induced a higher viremia and higher fever than GIII CH1392 virus at 3 DPI.
Viremic migratory birds are suspected of spreading GI viruses between countries but the role of avian species in local transmission of GI virus has never been determined [14]. Experimentally JEV-infected young chickens and ducklings showed an age-dependent ability to induce sufficient viremia for mosquito infection [25]. To compare the replication ability of GIII to GI virus in domestic avian species, we subcutaneously inoculated 104 ffu of GIII CH1392 and T1P1 strains, GI YL2009-4 and TC2009-1 strains or PBS into one-day old chickens (Fig 4C) and two-day old ducklings (Fig 4D). As early as 1 DPI, we found that 100% (8/8) and 75% (6/8) of GI virus-infected chickens and ducklings developed viremia, while 37.5% (3/8) and 12.5% (1/8) of GIII virus-infected chickens and ducklings, respectively, developed viremia (Fig 4C and 4D). GIII and GI virus titers were highest at 2 DPI and subsequently dropped the following 2 days (Fig 4C and 4D). These results showed that GI viruses replicated to a significantly (p<0.05) higher viremia (0.60–1.73-log higher) as well as earlier and longer lasting than GIII viruses in chickens and ducklings. The higher viremia was supported by the higher RNAemia detected in GI virus-infected poultry compared to GIII virus-infected poultry at 2 DPI (p< 0.05) (S5B and S5C Fig). However, no clinical signs were observed in the JEV-infected and PBS-inoculated poultry during the 6-day observation period.
To investigate the viral genetic determinants for the enhancement of GI virus infectivity, we initially analyzed the amino acid variances between seventy-one GIII viruses and seventy-seven GI viruses. Nine and twenty-four GI virus-specific and highly consensus substitutions were identified in structural and non-structural proteins, respectively, and three were non-conservative, charge-altering substitutions (Table 1). These substitutions as well as the other forty-three variations were observed between GIII and GI viruses used in this study. We also identified the cyclization structural variants formed by GIII and GI viruses in the untranslated region (UTR, S6 Fig), which might influence viral RNA synthesis [26]. To pin-point the target gene, we constructed five infectious clones of derived chimeric viruses (pCMV GIII/GI UTR, pCMV GIII/GI C-E, pCMV GIII/GI NS1-5, pCMV GIII/GI NS1-3, pCMV GIII/GI NS4-5) and one GI virus infectious clone (pCMV GI) by introducing the variant region of GI virus into the GIII virus infectious clone (pCMV GIII) (Fig 5A). These infectious cDNA clones were transfected into BHK-21 cells and subsequently detected by the expression of viral NS1 protein using an immunofluorescence assay (Fig 5B). Recombinant viruses, recovered from the transfected cells, formed similar sized plaques in BHK-21 cells (Fig 5B). Next we analyzed the replication ability of five chimeric recombinant viruses (rGIII/GI UTR, rGIII/GI C-E, rGIII/GI NS1-5, rGIII/GI NS1-3, and rGIII/GI NS4-5), recombinant GIII (rGIII) virus, and recombinant GI (rGI) virus in C6/36 cells at 28°C and in vertebrate cells (VERO, PK-15, and DF-1) at 41°C (Fig 5C–5F). All recombinant viruses reached similar titers in the range of 107.09 to 107.62 ffu/ml in C6/36 cells at 48 HPI. However, rGI virus exhibited a robust replication and significantly higher titer than rGIII virus in VERO (100.6-fold difference at 24 HPI, p< 0.05), PK-15, (101.1 and 101.4-fold differences at 24 and 48 HPI, p< 0.05) and DF-1 cells (100.6-fold difference at 48 HPI, p< 0.05) (Fig 5D–5F). Interestingly the rGIII/GI NS1-5 and rGIII/GI NS1-3 viruses replicated as efficiently as rGI virus, and produced a significantly higher titer than rGIII virus in VERO, PK-15, and DF-1 cells (p< 0.05). In contrast, rGIII/GI UTR, rGIII/GI C-E, rGIII/GI NS4-5, and rGIII viruses all maintained a low replication efficiency. These results demonstrated that the major genetic determinants for the enhancement of GI virus infectivity are located within NS1-3 proteins.
Conversely, we investigated the influence of the replication ability of GIII virus- derived NS1-5 and NS1-3 proteins on the rGI backbone for amplification in host-derived cells. The chimeric clones pCMV GI/GIII NS1-5 and pCMV GI/GIII NS1-3 were constructed to produce rGI/GIII NS1-5 and rGI/GIII NS1-3 viruses as described above (Fig 5A and 5B). The recombinant viruses yielded similar titers compared to rGIII virus but produced significantly lower titers than rGI virus in VERO, PK-15, and DF-1 cells at 41°C (p< 0.05) (Fig 5D–5F). These results further support the conclusion that GI-derived NS1-3 genes made a major contribution to enhanced replication of GI virus in host-derived cells at elevated temperature.
To verify the role of NS1-3 genes in vivo, we compared the replication ability of rGIII/GI NS1-3 virus to rGIII virus by subcutaneously inoculating 107 ffu and 104 ffu of the viruses into ten-week old SPF pigs and 1-day old chickens, respectively (Fig 6). The rGIII and rGIII/GI NS1-3 virus-infected pigs developed viremia with an average titer of 102.45 and 103.05 ffu/m at 2 DPI. The following day, rGIII/GI NS1-3 virus-infected pigs reached peak viremia but no viremia was detected in rGIII virus-inoculated pigs (Fig 6A). This difference was also observed in the plasma viral RNA collected from the infected pigs at 2 DPI (S7A Fig). These results highlighted the higher and extended viremia in rGIII/GI NS1-3 virus-infected pigs compared to rGIII virus-infected pigs. In addition, we also observed that rGIII/GI NS1-3 virus induced significantly higher viremia and RNAemia in chickens with 5-fold and 6-fold increases in viral titer and viral RNA compared to rGIII virus (p< 0.05) at 60 HPI, respectively (Fig 6B and S7B Fig). These results further suggest that the major viral determinants for the enhancement of GI virus infectivity in pigs and chickens were located on NS1-3 proteins.
To verify the specific substitution(s) of GI NS1-3 proteins involved in the enhancement of GI replication, we conducted a loss-of-function experiment by introduction of a single GIII virus-specific and highly consensus substitution for NS1-3 proteins of the rGIII/GI NS1-3 chimeric virus instead of the rGI virus (Fig 7A). The influence of these substitutions on the replication ability of rGIII/GI NS1-3 virus was evaluated in C6/36 cells at 28°C or in VERO, PK-15, and DF-1 cells at 41°C. As expected, the replication advantage of the rGIII/GI NS1-3 virus was consistently observed in PK-15 and DF-1 cells but not in C6/36 and VERO cells as compared to the rGIII virus at 48 HPI (Figs 5 and 7). However, the enhanced replication of the rGIII/GI NS1-3 virus was significantly reduced after the introduction of three single-substitutions (NS3-S78A, NS3-P105A, or NS3-D177E) in PK-15 cells (p< 0.05) and five single-substitutions (NS2A-I6V, NS2A-T149S, NS2B-L99V, NS3-S78A, or NS3-D177E) in DF-1 cells (p<0.05), respectively. The NS2A-R187K substitution had a minor effect on viral titer in both cell lines (p> 0.05) (Fig 7D and 7E). These results suggest that NS1 substitutions might not be critical factors for influencing the phenotype of the rGIII/GI NS1-3 virus.
The loss-of-function experiments were also used to evaluate substitutions for the degree of influence on the in vivo replication ability of rGIII/GI NS1-3 virus in 1-day old chickens. The rGIII and parental rGIII/GI NS1-3 viruses were inoculated as infection controls (Fig 7F and S8A Fig). As expected, rGIII/GI NS1-3 virus replicated to a significantly higher viremia than rGIII virus, and yielded 104.25 ffu/ml (Fig 7F) and 106.93 viral RNA copies/ml (S8A Fig) in the chickens at 60 HPI. In contrast, the recombinant viruses encoding a single NS2B-L99V, NS3-S78A, or NS3-D177E substitution induced a significantly lower viremia or RNAemia than the parental rGIII/GI NS1-3 virus (p< 0.05). The other recombinant viruses encoding a single NS2A-I6V, NS2A-T149S, NS2A-R187K or NS3-P105A substitution showed a viremia comparable to rGIII/GI NS1-3 virus in chickens. These results further supported a conclusion that the residues NS2B-99, NS3-78, and NS3-177 were involved in the replication enhancement of GI virus in vitro and in vivo.
To investigate the inter-dependency among the substitutions, we introduced single or multiple GI virus NS2B-V99L, NS3-A78S, and NS3-E177D substitutions into rGIII viruses (Fig 8A). Seven mutant rGIII viruses were generated and their infectivity evaluated using C6/36 cells at 28°C and VERO, PK-15, and DF-1 cells at 41°C. rGIII, rGI, and rGIII/GI NS1-3 viruses were included as controls (Fig 8B–8E). As expected, all recombinants yielded similar viral titers in C6/36 and VERO cells but rGIII virus exhibited a significantly lower titer than rGIII/GI NS1-3 viruses in PK-15 and DF-1 cells at 48 HPI. Mutant rGIII viruses encoding single NS2B-V99L, NS3-A78S, double NS2B-V99L-NS3-A78S, or triple substitutions yielded significantly higher viral titers (0.59 to 1.00-log increase in viral titer) in both PK-15 and DF-1 cells as compared to rGIII virus (p< 0.05). The effect of substitutions on viral replication was further evaluated by inoculating mutant rGIII viruses into 1-day old chickens. rGIII/GI NS1-3 virus replicated to higher titer (1.02 log) (Fig 8F) and produced higher levels of viral RNA (0.99 log) (S8B Fig) than rGIII virus at 48 HPI. With the exception of the NS2B-V99L substitution, the remaining mutant rGIII viruses encoding single and multiple substitutions had 0.68 to 0.96-log higher viral titers and 0.46 to 1.01-log higher viral RNA production than rGIII virus in 1-day old chicken. However, we were unable to detect an apparent synergistic effect among three substitutions.
Emerging GI virus has gradually replaced GIII virus as the dominant JEV isolated from human cases, stillborn piglets, and Culex tritaeniorhynchus since the 1990s. The mechanism of genotype replacement remains unclear, especially the role of the genetic determinants affecting the local pig-Culex tritaeniorhynchus transmission cycle. In this study, we identified the contribution of NS2B/NS3 mutations correlated with enhanced replication of GI virus in amplifying hosts: domestic pigs as well as in day-old chickens and two-day old ducklings. The role of Culex tritaeniorhynchus mosquito might be less significant in the genotype replacement.
There are two geographic variants of GI JEVs: viruses from GI-a clade are mainly distributed in tropical areas of Asia and Australia [13, 16, 27] and viruses of GI-b clade used in this study are widely distributed in southern and eastern Asia [16]. Previous studies have suggested that GI-a or GI-b viruses can compete with GIII for the same mosquito vector and amplifying hosts but are less likely to co-circulate in the same geographic locations [13, 20, 27, 28]. Mosquito vectors play a critical role in the occurrence of newly emerging flavi- and alphaviruses [29–32]. The previous reports indicated that GI-b virus replicated to higher titer in Aedes albopictus mosquito-derived cells [18] but inconsistent results on the infectivity of GI-a, GI-b, and GIII viruses were observed in Culex quinquefasciatus [33, 34]. Culex tritaeniorhynchus is the primary mosquito vector for GI and GIII viruses and account for 93.58% and 74.22% of mosquito-derived isolates, respectively [20]. We used in vitro Culex tritaeniorhynchus-derived cells as well as in vivo Culex tritaeniorhynchus mosquitoes to study the infectivity of GI-b and GIII viruses in the current study. Our study results indicated that both genotypes of JEV replicated to similar viral titers regardless of the assay system used (Fig 2) and suggested that the primary vector, Culex tritaeniorhynchus mosquito, for GI-b and GIII viruses is less likely to play a significant role in the genotype replacement of GIII to GI-b.
Migratory birds were suspected of spreading GI-b virus from southern to southeastern Asia and thus associated with JEV genotype replacement [19, 35]. The ardeid birds (herons and egrets) and pigs play an important role in local transmission of JEV [8]. All host-derived GI viruses are isolated from pigs [20]. However, JEV can infect young domestic poultry (chickens and ducklings) which are suspected of being involved in local transmission of JEV due to viral titers being sufficiently higher than the minimum infectious dose for mosquito hosts [25]. Earlier or higher viremia in GI-b virus-infected birds [19] and ducklings [36] suggested that a domestic avian-mosquito cycle might enhance the transmission of GI viruses. GI-b virus induced higher viremia than GIII virus in pigs, young chickens and ducklings (Fig 4). However, previous studies showed similar replication ability of GI-b, JE-91 strain, isolated in 1990 and GIII virus in DF-1 cells and ducklings [18, 33]. We speculate that this difference could be a result of the genetic variation of earlier GI-b isolates and 3 days older ducklings used in the previous studies [18, 33]. The NS2B/NS3 protein sequences of GI and GIII viruses used in the previous studies were unavailable. Higher viremia induced by GI-b virus could enhance viral transmission by mosquitoes since the infection rate of JEV was dose-dependent in mosquitoes [37]. In addition, GI-b virus produced higher levels of viral RNA than GIII virus in pig tonsil and nasal mucosa explants, potentially enhancing oronasal transmission between pigs [38, 39]. Collectively, our study suggests that the replacement of GIII virus with GI-b virus as the dominant and circulating virus in the host-mosquito cycle is the result of enhancement of transmission efficiency in amplifying hosts, including pigs and domestic avian species, not in mosquito vectors.
Mutation in the NS1 protein of Zika virus and the NS3 protein of HCV has been shown to enhance viral fitness during flaviviral evolution [29, 40]. Genomic sequencing and analysis has suggested that substitutions in E, NS4B, and NS5 proteins were involved in the evolutionary advantage of GI virus [18, 20]. Experimental evidence provided in our study, however, suggested that the enhancement of GI-b virus infectivity in amplifying hosts was associated with NS2B/NS3 substitutions (Figs 7 and 8), especially the residues NS2B-99 and NS3-78 in the protease domain and NS3-177 in the loop connecting protease and helicase domains of NS2B/NS3 proteins (Fig 9). Other substitutions such as NS2B-D65E, NS2B-A105P, NS3-N182S, or NS1/NS2A only played a minor role in improving viral fitness. The GI virus NS2B-99, NS3-78, and NS3-177 residues were also observed in GII viruses or GV viruses (S2 Table). This implied the substitutions associated with genotype replacement of GIII virus by GI virus and geographical distribution of the other genotypes may be different. The NS2B/NS3 proteins were involved in viral RNA replication, polypeptide processing, and infectious particle assembly through enzymatic-dependent or -independent processes [41, 42]. Thus, the GI-b virus NS2B/NS3 substitutions may enhance viral replication at post viral entry, as supported by the observation that infectivity rates were similar between rGI and rGIII viruses in the infectious center assay (S9 Fig). Moreover, flaviviral NS2B/NS3 proteins harbor multiple strategies to evade host innate immunity [42, 43]. JEV NS2B/NS3 protease was able to cleave interferon stimulator [44]. This interferon antagonistic ability of JEV was critical for efficient replication and increased virulence in mice [45]. A novel interferon antagonist of NS1 protein has recently been identified in newly emerging Zika viruses associated with the current epidemics [46]. In contrast to PK-15 and DF-1 cells, the virus titer (Figs 7 and 8) and focus size (S10 Fig) had no significant influence by NS2B/NS3 substitution in interferon-deficient VERO cells, suggested that the virus titers and focus size may associate with interferon antagonism or other host factors. Therefore, we hypothesize those two possible mechanisms of enhancement in viral post-entry and innate immunity antagonistic ability result in the replication advantage of GI-b virus in amplifying hosts.
Flaviviruses adaptation to elevated temperatures have been shown to enhance fitness in avian species and mosquito vectors [21, 48]. The enhancement of GI-b virus infectivity in amplifying hosts was observed at elevated temperatures (Fig 3A–3F). A West Nile virus study found that thermostability of replication was associated with higher viremia in avian species [48]. The GI-b NS2B/NS3 substitutions might enable NS2B/NS3 protein complex or NS3/NS5 replicase complex to interact more effectively with heat shock proteins for proper folding and hence stability at elevated temperature [23, 49]. Thus, it is possible that higher body temperature or development of fever in amplifying hosts could positively modulate interferon activity of vertebrate hosts against viral infection [50, 51]. Therefore, the influence of temperature on enzymatic activity, heat shock proteins-interacting ability, and interferon antagonistic ability of GI-b and GIII virus NS2B/NS3 proteins should be investigated in future studies.
The study of GI-a and GIII virus infectivity in Culex quinquefasciatus were inconsistent in previous studies [33, 34]. The NS2B-99, NS3-78, and NS3-177 residues were conserved in GI-a and GI-b viruses but additional substitutions were identified on E-141, NS2A-105, and NS5-438. Thus, the virological factors underlying the replacement of GIII virus by GI-a virus may require re-evaluation of viral replication ability in both Culex tritaeniorhynchus and amplifying hosts using dominant, circulating GI-a isolates in the future.
There are two limitations of the current study: the limited number of SPF pigs used to reveal the replicative advantage of GI vs. GIII viruses and the lack of analytical data to determine the viral dissemination and transmission ability in Culex tritaeniorhynchus. However, even with these limitations we should not underestimate the disease burden of JEV caused by GI virus infection. We have no doubt that GI viruses are more efficiently transmitted in the amplifying host-mosquito cycle and have similar virulence compared to the GIII virus in human [52]. Thus, we suggest that it is important to continually monitor GI virus evolution and clarify the role of avian species in local transmission of GI virus.
Animal Use protocols were approved by the Institutional Animal Care and Use Committees (IACUCs) in National Chung Hsing University (NCHU) (protocol number: 102–107) and National Pingtung University of Science and Technology (NPUST) (protocol number: 104–013). All experimental protocols followed the Guide for the Care and Use of Laboratory Animals published by the National Institutes of Health.
A total of 196 1 day-old specific pathogen free (SPF) chickens (Gallus gallus domesticus) and 80 2 day-old minimum disease-free ducklings (Cairina moschata) were purchased from JD-SPF Biotech Co., Ltd and Livestock Research Institute (Council of Agriculture in Taiwan), respectively. These animals were kept in isolators at the avian holding facility in NCHU. Thirteen second-generation SPF pigs (Lee-Sung Strain) were purchased from the Agricultural Technology Research Institute in Taiwan, and housed in the negative air-pressure animal facility certified by the Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC) in the Animal Disease Diagnostic Center of NPUST, Taiwan. Avian and pig euthanasia performed by the use of CO2 inhalation and electrical stunning were approved by IACUCs in NCHU and NPUST, respectively.
C6/36 cells (provided by Dr. Yi-Ling Lin from Academic Sinica, Taiwan) derived from the midgut of Aedes albopictus and CTR209 cells (provided by Dr. Kyoko Sawabe from National Institute of Infectious Disease, Japan) derived from embryos of Culex tritaeniorhynchus [53] were grown in Roswell Park Memorial Institute (RPMI) 1640 medium (Gibco) supplemented with 5% fetal bovine serum (FBS, Gibco) and in VP12 medium [54] supplemented with 10% FBS, respectively. Porcine kidney cells (PK-15, provided by Dr. Chienjin Huang from NCHU, Taiwan), chicken embryo fibroblast cells (DF-1, provided by Dr. Shan-Chia Ou from NCHU, Taiwan), chicken embryo related cells (CER) and duck embryo cells (DE and CER, provided by Dr. Poa-Chun Chang from NCHU, Taiwan), and monkey kidney cells (VERO, provided by Dr. Gwong-Jen J. Chang from Centers for Disease Control and Prevention (CDC), United States of America) were all grown in Dulbecco’s Modified Eagle Medium (DMEM, Gibco) supplemented with 10% FBS except for VERO cells with 5% FBS. Baby hamster kidney cells (BHK-21, provided by Dr. Wei-June Chen from Chang Gung University (CCU), Taiwan) were cultured in Minimum Essential Medium (MEM, Gibco) supplemented with 10% FBS. Mosquito cells or mammalian cells were maintained in incubators supplied with 5% CO2 at 28°C or 37°C. GIII subcluster II JE CH1392 and T1P1 viruses (provided by Dr. Wei-June Chen from CCU, Taiwan), and GI-b subcluster II JE YL2009-4 and subcluster I TC2009-1 viruses (from an already-existing collection in our lab) were used in this study. JEV strains CH1392, YL2009-4, and TC2009-1 were isolated from pools of field-captured Culex tritaeniorhynchus in 1990 and 2009, and T1P1 was isolated from the pool of Armigeres subalbatus in 1997 [55, 56]. All viral stocks were amplified in C6/36 cells and stored at -80°C until used.
JEVs or rJEVs were inoculated onto C6/36, CTR209, PK-15, DF-1, CER, DE, and VERO cells at an MOI of 0.5 at 28°C for mosquito cells and at 37°C for mammalian cells. Infected cells were washed three times with 1X PBS after a one-hour incubation, and then subsequently incubated at 28°C or 34°C for mosquito cells and at 37°C or 41°C for mammalian cells. The supernatants from infected cells were collected and stored at −80°C until used. All experiments were conducted in triplicate, and the viral titers in the supernatants were determined by the micro-antigen focus assay.
The micro-antigen focus assay was used to determine viral titers in the supernatants of infected cells, in the plasma recovered from infected animals, or in the supernatants of homogenized mosquitoes. Briefly, VERO cells were seeded in to 96-well plates at a cell count of 2.25×104 cells/100 μL/well and incubated at 37°C, 5% CO2 overnight to allow a cell monolayer to form. The serially diluted samples were added into wells for one hour at 37°C. Infected VERO cells were overlayed with 1% methylcellulose mixed with DMEM medium supplemented with 2% FBS, and incubated at 37°C for 32 to 36 hours. The methylcellulose overlays were discarded by washing with 1X PBS, and the infected VERO monolayers were fixed with 75% acetone in PBS for 20 minutes. The fixed cells were air-dried and stained with mouse anti-JEV polyclonal antibody (provided by Dr. Gwong-Jen J. Chang from CDC, USA), followed by staining with horseradish peroxidase (HRP)-conjugated goat anti-mouse IgG antibody (Jackson ImmunoResearch, West Grove, PA). The foci were developed after the addition of Vector-VIP (Vector Laboratories, Burlingame, CA) into each well. The viral titer was calculated by the average number of foci-forming unit (ffu) per ml or per mosquito.
Laboratory-hatched female Culex tritaeniorhynchus mosquitoes were fed with 10% sucrose and maintained at 28°C. Mosquitoes were starved for 1 day prior to the blood-feeding experiment. The mosquitoes were fed per os with a JEV viremic blood meal, a mixture of pig blood cells and 8×106 ffu/ml of JEV according to the previous study [57] and the stock titer of JEVs used in this study. The virus-infected mosquitoes were maintained in the different cage at 28°C. Infected mosquitoes were collected from cages by aspiration and homogenized individually at 14 days post- infection (DPI).
Thirteen ten-week old, JEV-seronegative SPF pigs were used to determine the replication ability of field-isolated JEVs (nine pigs) and rJEV (four pigs). SPF pigs were anesthetized with stresnil (China Chemical and Pharmaceutical Co., Ltd) and subcutaneously inoculated with PBS or 105 ffu of GIII CH1392 virus, GI YL2009-4 virus, or 107 ffu of rJEVs. Experimental pigs were monitored, the daily body temperatures were recorded, and clinical signs of infection were noted. Pig plasmas were recovered at different days post infection. All pigs were euthanized with electrical stunning at 8 DPI.
Eighty 1-day old chickens (16 per viral group) and eighty 2-day old ducklings (16 per viral group) were subcutaneously inoculated with one dose of PBS or 104 ffu of GIII CH1392 virus, GIII T1P1 virus, GI YL2009-4 virus, or GI TC2009-1 virus. The daily activity and clinical signs of infection for chickens and ducks were monitored after JEV infection. We collected blood from four infected chickens or ducklings 1 day prior to infection and at 1, 2, 4, and 6 DPI. Plasma was mixed with anticoagulant at a final concentration of 0.33% sodium citrate (Sigma-Aldrich) in 0.85% sodium chloride (Sigma-Aldrich), and centrifuged at 3,000 rpm for 15 minutes. The ten-fold diluted plasma was recovered from the supernatant and stored at −80°C until used.
One hundred sixteen 1 day-old chickens (4 per group in Figs 6 and 7; 6 per group in Fig 8) were subcutaneously inoculated with PBS or 104 ffu of rJEVs. The plasma was recovered from infected chickens at 48 or 60 HPI as described above, and stored at −80°C until used.
The infectious clone encoding the full genome of GIII JE RP9 virus was constructed using pBR322 plasmid and referred to as pCMV GIII (kindly provided by Dr. Yi-Ling Lin from Academic Sinica, Taiwan) in this study. JEV viral RNA was transcribed by CMV promoter and terminated by SV40 poly-A terminator. The precise JEV 3’ terminal sequence was generated by a ribozyme sequence of hepatitis delta virus (HDVr) incorporated right after the 3’UTR of JEV (5). To construct GIII and GI JEV chimeric infectious clones, we replaced five genetic fragments or the complete viral genome of pCMV GIII with the corresponding genes of GI JE YL2009-4 virus by blunt-end ligation with T4 DNA ligase (New England Biolabs) to generate six recombinant viruses (rGI, rGIII/GI UTR, rGIII/GI C-E, rGIII/GI NS1-5, rGIII/GI NS1-3 and rGIII/GI NS4-5). We replaced the pCMV GI infectious cDNA clone with the corresponding gene fragment of GIII virus using the protocol of Gibson assembly reaction (New England Biolabs) and generated two additional recombinant viruses, rGI/GIII NS1-5 and rGI/GIII NS1-3 (Fig 5A). All fragments were amplified by PCR reactions (KOD, Novagen). PCR templates and primers are listed in S1 Table. cDNA constructs, extracted from the transformed competent cells using Mini-prep kit (Qiagen), were sequenced to authenticate the complete viral genome insert.
The site-specific mutation was individually introduced into pCMV GIII/GI NS1-3 or pCMV GIII by site-directed mutagenesis using the following reaction mixes: 1.5 mM MgSO4, 0.2 mM dNTPs, 0.4 mM mutagenesis primers (S1 Table), 0.5U KOD Hot Start DNA polymerase (KOD, Novagen), and the respective cDNA clones. Mutated clones were identified in cDNA constructs, extracted from the transformed competent cells using Mini-prep kit (Qiagen), and sequenced to authenticate the complete viral genome insert.
BHK-21 cells were seeded into 12-well plates and grown at 37°C overnight. The next day, the mixture of 1 μg of the infectious clone and Opti-MEM (Life Technologies) was added into a mixture of Lipofectamine 2000 (Life Technologies) and Opti-MEM, and then the final mixture was incubated at room temperature for 30 minutes. Next the mixture was added onto the 80% confluent cells for five hours at 37°C, and then replaced with the culture medium. After a 3- to 4-day incubation, the production of rJEVs was detected in the transfected cells by an immunofluorescence assay utilizing mouse anti-JEV NS1 antibody (provided by Dr. Yi-Ling Lin from Academic Sinica, Taiwan). The recombinant viruses secreted from transfected cells were harvested and subsequently amplified in C6/36 cells. Virus plaques were identified by plaque assay. The viral RNA was extracted from virus-infected C6/36 cells with RNeasy mini kit (Qiagen), and transcribed into cDNA with the JEV 3’UTR primer 5’-AGATCCTGTGTTCTTCCTCA-3’ using the Superscript III transcription reaction (Thermo Fisher Scientific). The complete genome of recombinant virus was confirmed by sequencing the PCR products amplified from the viral cDNA template.
The statistical analysis was performed by GraphPad Prism v5.01. Student’s two-tailed t-test was used to compare two data groups. The multiple-group comparison was calculated by One-way ANOVA, and the post test analysis performed by using Turkey’s Multiple or Dunnett’s Multiple Comparison Test. P<0.05, the significant difference in the two-group and multiple-group comparison.
|
10.1371/journal.pcbi.1001057 | Is My Network Module Preserved and Reproducible? | In many applications, one is interested in determining which of the properties of a network module change across conditions. For example, to validate the existence of a module, it is desirable to show that it is reproducible (or preserved) in an independent test network. Here we study several types of network preservation statistics that do not require a module assignment in the test network. We distinguish network preservation statistics by the type of the underlying network. Some preservation statistics are defined for a general network (defined by an adjacency matrix) while others are only defined for a correlation network (constructed on the basis of pairwise correlations between numeric variables). Our applications show that the correlation structure facilitates the definition of particularly powerful module preservation statistics. We illustrate that evaluating module preservation is in general different from evaluating cluster preservation. We find that it is advantageous to aggregate multiple preservation statistics into summary preservation statistics. We illustrate the use of these methods in six gene co-expression network applications including 1) preservation of cholesterol biosynthesis pathway in mouse tissues, 2) comparison of human and chimpanzee brain networks, 3) preservation of selected KEGG pathways between human and chimpanzee brain networks, 4) sex differences in human cortical networks, 5) sex differences in mouse liver networks. While we find no evidence for sex specific modules in human cortical networks, we find that several human cortical modules are less preserved in chimpanzees. In particular, apoptosis genes are differentially co-expressed between humans and chimpanzees. Our simulation studies and applications show that module preservation statistics are useful for studying differences between the modular structure of networks. Data, R software and accompanying tutorials can be downloaded from the following webpage: http://www.genetics.ucla.edu/labs/horvath/CoexpressionNetwork/ModulePreservation.
| In network applications, one is often interested in studying whether modules are preserved across multiple networks. For example, to determine whether a pathway of genes is perturbed in a certain condition, one can study whether its connectivity pattern is no longer preserved. Non-preserved modules can either be biologically uninteresting (e.g., reflecting data outliers) or interesting (e.g., reflecting sex specific modules). An intuitive approach for studying module preservation is to cross-tabulate module membership. But this approach often cannot address questions about the preservation of connectivity patterns between nodes. Thus, cross-tabulation based approaches often fail to recognize that important aspects of a network module are preserved. Cross-tabulation methods make it difficult to argue that a module is not preserved. The weak statement (“the reference module does not overlap with any of the identified test set modules”) is less relevant in practice than the strong statement (“the module cannot be found in the test network irrespective of the parameter settings of the module detection procedure”). Module preservation statistics have important applications, e.g. we show that the wiring of apoptosis genes in a human cortical network differs from that in chimpanzees.
| Network methods are frequently used in genomic and systems biologic studies, but also in general data mining applications, to describe the pairwise relationships of a large number of variables [1], [2]. For example, gene co-expression networks can be constructed on the basis of gene expression data [3]–[10]. In many network applications, one is interested in studying the properties of network modules and their change across conditions [11]–[16]. For example, [17]–[19] studied modules across multiple mouse tissues, [20] studied module preservation between human brain and blood tissue, and [21] studied module preservation between human and mouse brains.
This article describes several module preservation statistics for determining which properties of a network module are preserved in a second (test) network. The module preservation statistics allow one to quantify which aspects of within-module topology are preserved between a reference network and a test networks. For brevity, we will refer to these aspects as connectivity patterns, but we note that our statistics are not based on network motifs. We use the term “module” in a broad sense: a network module is a subset of nodes that forms a sub-network inside a larger network. Any subset of nodes inside a larger network can be considered a module. This subset may or may not correspond to a cluster of nodes.
Many cluster validation statistics proposed in the literature can be turned into module preservation statistics. In the following, we briefly review cluster validation statistics. Traditional cluster validation (or quality) statistics can be split into four broad categories: cross-tabulation, density, separability, and stability statistics [22]–[24]. Since cross-tabulation statistics compare cluster assignments in the reference and test clusterings, they require that a clustering procedure is also applied to the test data. On the other hand, density and density/separability statistics do not require a clustering in the test data set. These statistics typically evaluate clusters by how similar objects are within each cluster and/or how dis-similar objects are between different clusters [25]. Stability statistics typically study cluster stability when a controlled amount of artificial noise is added to the data. Although stability statistics also evaluate clusters, they are more relevant to comparing clustering procedures rather than quantifying cluster preservation and hence we do not consider them here.
While many cluster validation statistics are based on within- and/or between cluster variance, several recent articles used prediction error to evaluate the reproducibility (or validity) of clusters in gene expression data [24], [26], [27]. These papers argued that the use of a measure of test set clusters defined by a classifier made from the reference data is an appropriate approach to cluster validation when the aim is to identify reproducible clusters of genes or microarrays with similar expression profiles. For example, the in-group proportion (IGP), which is similar to the cluster cohesion statistic [28], is defined as the proportion of observations classified to a cluster whose nearest neighbor is also classified to the same cluster [24]. One can also calculate a significance level (p-value) for the IGP statistic. A comparison of the IGP statistic to alternative cluster quality statistics found that the IGP performs well [24]. Thus, we use the IGP statistic as benchmark statistic for assessing the use of module preservation statistics in case that modules are defined as clusters. Our simulation studies and applications show that one of our module preservation statistics is sometimes closely correlated with the IGP statistic if the modules are defined as clusters. But cluster validation statistics (such as the IGP) may not be appropriate when modules are not defined as clusters. In general, assessing module preservation is a different task from assessing cluster preservation. In our simulations, we demonstrate that module preservation statistics can detect aspects of module preservation that are missed by existing cluster validation statistics.
Table 1 presents an overview of the module preservation statistics studied in this article. We distinguish between cross-tabulation based and network based preservation statistics. Cross-tabulation based preservation statistics require independent module detection in the test network and take the module assignments in both reference and test networks as input. Several cross-tabulation based statistics are described in the first section of Supplementary Text S1. While cross-tabulation approaches are intuitive, they have several disadvantages. To begin with, they are only applicable if the module assignment in the test data results from applying a module detection procedure to the test data. For example, a cross-tabulation based module preservation statistic would be meaningless when modules are defined as gene ontology categories since both reference and test networks contain the same sets of genes. But a non-trivial question is whether the network connections of a module (gene ontology category) in the reference network resemble those of the same module in the test network. To measure the resemblance of network connectivity, we propose several measures based on network statistics. Network terminology is reviewed in Table 2 and in Methods.
Even when modules are defined using a module detection procedure, cross-tabulation based approaches face potential pitfalls. A module found in the reference data set will be deemed non-reproducible in the test data set if no matching module can be identified by the module detection approach in the test data set. Such non-preservation may be called the weak non-preservation: “the module cannot be found using the current parameter settings of the module detection procedure”. On the other hand, one is often interested in strong non-preservation: “the module cannot be found irrespective of the parameter settings of the module detection procedure”. Strong non-preservation is difficult to establish using cross-tabulation approaches that rely on module assignment in the test data set. A second disadvantage of a cross-tabulation based approach is that it requires that for each reference module one finds a matching test module. This may be difficult when a reference module overlaps with several test modules or when the overlaps are small. A third disadvantage is that cross-tabulating module membership between two networks may miss that the fact that the patterns of connectivity between module nodes are highly preserved between the two networks.
Network based statistics do not require the module assignment in the test network but require the user to input network adjacency matrices (described in Methods). We distinguish the following 3 types of network based module preservation statistics: 1) density based, 2) separability based, and 3) connectivity based preservation statistics. Density based preservation statistics can be used to determine whether module nodes remain highly connected in the test network. Separability based statistics can be used to determine whether network modules remain distinct (separated) from one another in the test network. While numerous measures proposed in the literature combine aspects of density and separability, we keep them separate and provide evidence that density based approaches can be more useful than separability based approaches in determining whether a module is preserved. Connectivity based preservation statistics can be used to determine whether the connectivity pattern between nodes in the reference network is similar to that in the test network. As detailed in Methods, several module preservation statistics are similar to previously proposed cluster quality and preservation statistics, while others (e.g. connectivity based statistics) are novel.
Table 1 reports the required input for each preservation statistic. Since each preservation statistic is used to evaluate the preservation of modules defined in a reference network, it is clear that each statistic requires the module assignment from the reference data. But the statistics differ with regard to the module assignment in the test data. Only cross-tabulation based statistics require a module assignment in the test data. Network based preservation statistics do not require a test set module assignment. Instead, they require the test set network adjacency matrix (for a general network) or the test data set of numeric variables (for a correlation network).
We distinguish network statistics by the underlying network. Some preservation statistics are defined for a general network (defined by an adjacency matrix) while others are only defined for a correlation network (constructed on the basis of pairwise correlations between numeric variables). Our applications show that the correlation structure facilitates the definition of particularly powerful module preservation statistics. Preservation statistics 4–11 (Table 1) can be used for general networks while statistics 12–19 assume correlation networks. Network density and module separability statistics only need the test set adjacency matrix while the connectivity preservation statistics also require the adjacency matrix in the reference data.
It is often not clear whether an observed value of a preservation statistic is higher than expected by chance. As detailed in Methods, we attach a significance level (permutation test p-value) to observed preservation statistics, by using a permutation test procedure which randomly permutes the module assignment in the test data. Based on the permutation test we are also able to estimate the mean and variance of the preservation statistic under the null hypothesis of no relationship between the module assignments in reference and test data. By standardizing each observed preservation with regard to the mean and variance, we define a statistic for each preservation statistic. Under certain assumptions, each statistic (approximately) follows the standard normal distribution if the module is not preserved. The higher the value of a Z statistic, the stronger the evidence that the observed value of the preservation statistic is significantly higher than expected by chance.
Several studies have explored how co-expression modules change between mouse tissues [19] and/or sexes [18]. Here we re-analyze gene expression data from the liver, adipose, muscle, and brain tissues of an F2 mouse intercross described in [13], [17]. The expression data contain measurements of 17104 genes across the following numbers of microarray samples: 137 (female (F) adipose), 146 (male (M) adipose), 146 (F liver), 145 (M liver), 125 (F muscle), 115 (M muscle), 148 (F brain), and 141 (M brain).
We consider a single module defined by the genes of the gene ontology (GO) term “Cholesterol biosynthetic process” (CBP, GO id GO:0006695 and its GO offspring). Of the 28 genes in the CBP, 24 could be found among our 17104 genes. Cholesterol is synthesized in liver and we used the female liver network as the reference network module. As test networks we considered the CBP co-expression networks in other tissue/sex combinations.
Each circle plot in Figure 1 visualizes the connection strengths (adjacencies) between CBP genes in different mouse tissue/sex combination. The color and width of the lines between pairs of genes reflect the correlations of their gene expression profiles across a set of microarray samples. Before delving into a quantitative analysis, we invite the reader to visually compare the patterns of connections. Clearly, the male and female liver networks look very similar. Because of the ordering of the nodes, the hubs are concentrated on the upper right section of the circle and the right side of the network is more dense. The adipose tissues also show this pattern, albeit much more weakly. On the other hand, the figures for the brain and muscle tissues do not show these patterns. Thus, the figure suggests that the CBP module is more strongly preserved between liver and adipose tissues than between liver and brain or muscle.
We now turn to a quantitative assessment of this example. We start out by noting that a cross-tabulation based approach of module preservation is meaningless in this example since the module is a GO category whose genes can trivially be found in each network. However, it is a very meaningful exercise to measure the similarity of the connectivity patterns of the module genes across networks. To provide a quantitative assessment of the connectivity preservation, it is useful to adapt network concepts (also known as network statistics or indices) that are reviewed in Methods. Figure 2 provides a quantitative assessment of the preservation of the connectivity patterns of the cholesterol biosynthesis module between the female liver network and networks from other sex/tissue combinations. Figure 2A presents the composite summary statistic (, Equation 1) in each test network. Overall, we find strong evidence of preservation (, Equation 1) in the male liver network but no evidence () of preservation in the female brain and muscle networks. We find that the connectivity of the female liver CBP is most strongly preserved in the male liver network. It is also weakly preserved in adipose tissue but we find no evidence for its preservation in muscle and brain tissues. The summary preservation statistic measures both aspects of density and of connectivity preservation. We now evaluate which of these aspects are preserved. Figure 2B shows that the module shows strong evidence of density preservation () (Equation 30) in the male liver network but negligible density preservation in the other networks. Interestingly, Figure 2C shows that the module has moderate connectivity preservation (Equation 31) in the adipose networks.
The measure summarizes the statistical significance of 3 connectivity based preservation statistics. Two of our connectivity measures evaluate whether highly connected intramodular hub nodes in the reference network remain hub nodes in the test network. Preservation of intramodular connectivity reflects the preservation of hub gene status between the reference and test network. One measure of intramodular connectivity is the module eigengene-based connectivity measures (Equation 17), which is also known as the module membership measure of gene [13], [29], [30]. Genes with high values of are highly correlated with the summary profile of the module (module eigengene defined as the first principal component, see the fifth section in Supplementary Text S1). A high correlation of between reference and test network can be visualized using a scatter plot and quantified using the correlation coefficient . For example, Figure 2I shows that in the female liver module is highly correlated with that of the male liver network (, ). Further, the scatter plots in Figure 2 show that the measures between liver and adipose networks show strong correlation (preservation): (), (), (), while the correlation between in female liver and the brain and muscle data sets are not significant. This example demonstrates that connectivity preservation measures can uncover a link between CBP in liver and adipose tissues that is missed by density preservation statistics.
We briefly compare the performance of our network based statistics with those from the IGP method [24]. The R implementation of the IGP statistic requires that at least 2 modules are being evaluated. To get it to work for this application that involves only a single module, we defined a second module by randomly sampling half of the genes from the rest of the entire network. Figure 2D shows high, nearly constant values of the IGP statistic across networks, which indicates that the CBP module is present in all data sets. Note that the IGP statistic does not allow us to argue that the CBP module in the female liver network is more similar to the CBP module in the male liver than in other networks. This reflects the fact that the IGP statistic, which is a cluster validation statistic, does not measure connectivity preservation.
Here we study the preservation of co-expression between human and chimpanzee brain gene expression data. The data set consists of 18 human brain and 18 chimpanzee brain microarray samples [31]. The samples were taken from 6 regions in the brain; each region is represented by 3 microarray samples. Since we used the same weighted gene co-expression network construction and module identification settings as in the original publication, our human modules are identical to those in [32]. Because of the relatively small sample size only few relatively large modules could be detected in the human data. The resulting modules were labeled by colors: turquoise, blue, brown, yellow, green, black, red (see Figure 3A). Oldham et al (2006) determined the biological meaning of the modules by examining over-expression of module genes in individual brain regions. For example, heat maps of module expression profiles revealed that the turquoise module contains genes highly expressed in cerebellum, the yellow module contains genes highly expressed in caudate nucleus, the red module contains genes highly expressed in anterior cingulate cortex (ACC) and caudate nucleus, and the black module contains mainly genes expressed in white matter. The blue, brown and green modules contained genes highly expressed in cortex, which is why we refer to these modules as cortical modules. Visual inspection of the module color band below the dendrograms in Figures 3A and 3B suggests that most modules show fairly strong preservation. Oldham et al argued that modules corresponding to evolutionarily older brain regions (turquoise, yellow, red, black) show stronger preservation than the blue and green cortical modules [32]. Here we re-analyze these data using module preservation statistics.
The most common cross-tabulation approach starts with a contingency table that reports the number of genes that fall into modules of the human network (corresponding to rows) versus modules of the chimpanzee network (corresponding to columns). The contingency table in Figure 3C shows that there is high agreement between the human and chimpanzee module assignments. The human modules black, brown, red, turquoise, and yellow have well-defined chimpanzee counterparts (labeled by the corresponding colors). On the other hand, the human green cortical module appears not to be preserved in chimpanzee since most of its genes are classified as unassigned (grey color) in the chimpanzee network. Further, the human blue cortical module (360 genes) appears to split into several parts in the chimpanzee network: 27 genes are part of the chimpanzee blue module, 85 genes are part of the chimpanzee brown module, 52 fall in the chimpanzee turquoise module, 155 genes are grey in the chimpanzee network, etc. To arrive at a more quantitative measure of preservation, one may quantify the module overlap or use Fisher's exact test to attach a significance level (p-value) to each module overlap (as detailed in the first section of Supplementary Text S1). The contingency table in Figure 3C shows that every human module has significant overlap with a chimpanzee module. However, even if the resulting p-value of preservation were not significant, it would be difficult to argue that a module is truly a human-specific module since an alternative module detection strategy in chimpanzee may arrive at a module with more significant overlap. In order to quantify the preservation of human modules in chimpanzee samples more objectively, one needs to consider statistics that do not rely on a particular module assignment in the chimpanzee data.
We now turn to approaches for measuring module preservation that do not require that module detection has been carried out in the test data set. Figures 4A,B show composite module preservation statistics of human modules in chimpanzee samples. The overall significance of the observed preservation statistics can be assessed using (Equation 1) that combines multiple preservation statistics into a single overall measure of preservation, Figure 4A. Note that shows a strong dependence on module size, which reflects the fact that observing module preservation of a large module is statistically more significant than observing the same for a small module. However, here we want to consider all modules on an equal footing irrespective of module size. Therefore, we focus on the composite statistic which shows no dependence on module size (Figure 4B). The median rank is useful for comparing relative preservation among modules: a module with lower median rank tends to exhibit stronger observed preservation statistics than a module with a higher median rank. Figure 4B shows that the median ranks of the human brain modules. The median rank of the yellow module is 1, while the median ranks of the blue module is 6, indicating that the yellow module is more strongly preserved than the blue module. Our quantitative results show that modules expressed mainly in evolutionarily more conserved brain areas such as cerebellum (turquoise) and caudate nucleus (yellow and partly red) are more strongly preserved than modules expressed primarily in the cortex that is very different between humans and chimpanzees (green and blue modules). Thus the module preservation results of , corroborate Oldham's original finding regarding the relative lack of preservation of cortical modules.
Since the modules of this application are defined as clusters, it makes sense to evaluate their preservation using cluster validation statistics. Figure 4C shows that the IGP statistic implemented in the R package clusterRepro [24] also shows a strong dependence on module size in this application. The IGP values of all modules are relatively high. However, the permutation p-values (panels C and D) identify the green module as less preserved than the other modules (, Bonferroni corrected p-value 0.43). Figures 4E,F show scatter plots between the observed IGP statistic and and , respectively. In this example, where modules are defined as clusters, the IGP statistic has a high positive correlation () with and a moderately large negative correlation () with . The negative correlation is expected since low median ranks indicate high preservation.
While composite statistics summarize the results, it is advisable to understand which properties of a module are preserved (or not preserved). For example, module density based statistics allow us to determine whether the genes of a module (defined in the reference network) remain densely connected in the test network. As an illustration, we will compare the module preservation statistics for the human yellow module whose genes are primarily expressed in caudate nucleus (an evolutionarily old brain area), and the human blue module whose genes are expressed mostly in the cortex which underwent large evolutionary changes between humans and chimpanzees. In chimpanzees, the mean adjacency of the genes comprising the human yellow module is significantly higher than expected by chance, with a high permutation statistic , . But the corresponding permutation statistic for the human blue module is only weakly significant, , (see Supplementary Text S2 and Supplementary Table S1). Thus, the mean adjacency permutation statistic suggests that the blue module is less preserved than the yellow module.
For co-expression modules, one can define an alternative density measure based on the module eigengene (Figures 5A and E). The higher the proportion of variance explained by the module eigengene (defined in the fifth section in Supplementary Text S1) in the test set data, the tighter is the module in the test set. The human yellow module exhibits a high proportion of variance explained, , and the corresponding permutation statistic is , . In contrast, for the human blue module we find and the corresponding permutation statistic is , . The permutation statistics again suggest that the yellow module is more preserved than the blue module.
Although density based approaches are intuitive, they may fail to detect another form of module preservation, namely the preservation of connectivity patterns among module genes. For example, network module connectivity preservation can mean that, within a given module , a pair of genes with a high connection strength (adjacency) in the reference network also exhibits a high connection strength in the test network. This property can be quantified by correlating the pairwise adjacencies or correlations between reference and test networks. For the genes in the human yellow module, the scatter plot in Figure 5B shows pairwise correlations in the human network (-axis) versus the corresponding correlations in the chimpanzee network (-axis). The correlation between pairwise correlations (denoted by ) equals and is highly significant, . The analogous correlation for the blue module, Figure 5F is lower, 0.56, but still highly significant, , in part because of the higher number of genes in the blue module.
A related but distinct connectivity preservation statistic quantifies whether intramodular hub genes in the reference network remain intramodular hub genes in the test network. Intramodular hub genes are genes that exhibit strong connections to other genes within their module. This property can be quantified by the intramodular connectivity (Equation 7): hub genes are genes with high . Intramodular hub genes often play a central role in the module [5], [33]–[35]. Preservation of intramodular connectivity reflects the preservation of hub gene status between the reference and test network. For example, the intramodular connectivity of the human yellow module is preserved between the human and chimpanzee samples, (Figure 5C). In contrast, the human blue (cortical) module exhibits a lower correlation (preservation) (Figure 5G). The value is more significant because of the higher number of genes in the blue module.
Another intramodular connectivity measure is , which turns out to be highly related with [29]. Figure 5D shows that for the human yellow module is highly preserved in the chimpanzee network (). The corresponding correlation in the human blue module is lower, (Figure 5H). In summary, the observed preservation statistics show that the human yellow module (related to the caudate nucleus) is more strongly preserved in the chimpanzee samples than the human blue module (related to the cortex).
To further illustrate that modules do not have to be clusters, we now describe an application where modules correspond to KEGG pathways. KEGG (Kyoto Encyclopedia of Genes and Genomes) is a knowledge base for systematic analysis of gene functions, linking genomic information with higher order functional information [36]. KEGG also provides graphical representations of cellular processes, such as signal transduction, metabolism, and membrane transport. To illustrate the use of the module preservation approach, we studied the preservation of selected KEGG pathway networks across human and chimpanzee brain correlation networks. While pathways in the KEGG database typically describe networks of proteins, our analysis describes the correlation patterns between mRNA expression levels of the corresponding genes. As before, we define a weighted correlation network adjacency matrix between the genes (described in the third section of Supplementary Text S1 and [5]). For the sake of brevity, we focused the analysis on the following 8 signaling pathways: Hedgehog signaling pathway (12 genes in our data sets), apoptosis (24 genes in our data sets), TGF-beta signaling pathway (26 genes), Phosphatidylinositol signaling system (39 genes), Wnt signaling pathway (55 genes), Endocytosis (59 genes), Calcium signaling pathway (78 genes), and MAPK signaling pathway (93 genes). All of these pathways have been shown to play critical roles in normal brain development and function [37]–[41]. We provide a brief description of the functions of these pathways in Methods; more detailed description can be found in the KEGG database and in numerous textbooks.
Figures 6A,B show the composite preservation statistics and . Both statistics indicate that the apoptosis module is the least preserved module. To visualize the lack of preservation, consider the circle plots of apoptosis genes in Figures 7 L, M that show pronounced differences in the connectivity patterns among apoptosis genes. While we caution the reader that additional data are needed to replicate these differences, prior literature points to an evolutionary difference for apoptosis genes. For example, a scan for positively selected genes in the genomes of humans and chimpanzees found that a large number of genes involved in apoptosis show strong evidence for positive selection [42]. Further, it has been hypothesized that natural selection for increased cognitive ability in humans led to a reduced level of neuron apoptosis in the human brain [43].
Figure 6A shows that exhibits some dependence on module size. Since we want to compare module preservation irrespective of module size, we focus on the results for the statistic (Figure 6B). A reviewer of this article hypothesized that gene sets (modules) known to be controlled by coexpression (such as Wnt, TGF-beta, SRF, interferon, lineage specific differentiation markers, and NF kappa B) would show stronger evidence of preservation than gene sets without a priori reason for suspecting such control (calcium signaling, MAPK, apoptosis, chemotaxis, endocytosis). Interestingly, the results for the statistic largely validate this hypothesis. Specifically, the 4 most highly preserved pathways according to are Wnt (controlled by coexpression), calcium (not controlled), Hedgehog (controlled), and Phosphatidylinositol (not commented upon). The 4 least preserved pathways are apoptosis (not controlled), TGF-beta (controlled), MAPK (not controlled), endocytosis (not controlled).
Since KEGG pathways are not defined via a clustering procedure it is not clear whether cluster preservation statistics are appropriate for analyzing this example. But to afford a comparison, we also report the findings for the IGP statistic [24]. Figures 6C and D show that IGP identifies Phosphatidilinositol and TGF-beta as the least preserved modules while apoptosis genes are highly preserved. We find no significant relationship between the IGP statistic and our module preservation statistics and (Figures 6E and F). This example highlights that module preservation statistics can lead to very different results from cluster preservation statistics.
To understand which aspects of the pathways are preserved, one can study the preservation of density statistics (Figure 7B) and of connectivity statistics (Figure 7C). According to , the coexpresssion network formed by apoptosis genes is not preserved. It neither shows evidence of connectivity preservation () nor evidence of density preservation (, ). The Hedgehog pathway also shows no evidence of density preservation (, ) but it shows weak evidence of connectivity preservation (, ). The relatively low preservation Z statistics of the Hedgehog pathway may reflect a higher variability due to a small module size (it contains only genes while the other pathways contain at least 22 genes). To explore this further, we studied the observed preservation statistics, which are less susceptible to network size effects than the corresponding statistics. The scatter plots in Figure 7D–H show the correlations between eigengene based connectivity measures between the two species. For the Hedgehog pathway, we find that () which turns out to be higher than that of the TGF- pathway.
The lack of preservation of the apoptosis pathway cannot be explained in terms of low module size. Figure 7E shows that it has the lowest observed statistic, .
This application outlines how module preservation statistics can be used to study the preservation of KEGG pathway networks. The analysis presented here is but a first step towards characterizing molecular pathway preservation between human and chimpanzee brains, and should be extended through more detailed analyses with additional data sets in the future. A limitation of our microarray data is that they measured expression levels in heterogeneous mixtures of cells. KEGG and GO (gene ontology) pathways all essentially describe interactions that take place within cells. So when data have been generated from a heterogeneous mixture of different cell types, it is possible that these relationships are somewhat obscured. It is not obvious that all of the elements of a KEGG pathway should be co-expressed, particularly since the pathways describe protein-protein interactions.
We briefly describe an application that quantifies module preservation between male and female cortical samples. The details are described in Supplementary Text S3 and in Supplementary Table S2. We used microarray data from a recent publication [30] to construct consensus modules [44] in male samples from 2 different data sets. We then studied the preservation of these modules in the corresponding female samples. Cross-tabulation measures indicate that for 3 of the male modules there are no corresponding modules in the female data. However, our network preservation statistics show that in fact the three modules show moderate to strong evidence of preservation. Thus, in this application the network preservation statistics protect one from making erroneous claims of significant sex differences.
In Supplementary Text S4, we re-analyze the mouse liver samples of the F2 mouse intercross [13], [17] to study whether “female” co-expression modules (i.e., modules found in a network based on female mice) are preserved in the corresponding male network. This application demonstrates that module preservation statistics allow us to identify invalid, non-reproducible modules due to array outliers. A comprehensive table of module preservation statistics for this application is presented in Supplementary Table S3.
Our preservation statistics allow one to evaluate whether a given module is preserved in another network. A related but distinct data analysis task is to construct modules that are present in several networks. By construction, a consensus module can be detected in each of the underlying networks. A challenge of many real data applications is that it is difficult to obtain independent information (a “gold standard”) that allows one to argue that a module is truly preserved. To address this challenge, we use the consensus network application where by construction, modules are known to be preserved. This allows us to determine the range of values of preservation statistics when modules are known to be preserved. In Supplementary Text S5 and Supplementary Table S4, we report three empirical studies of consensus modules [44] which are constructed in such a way that genes within consensus modules are highly co-expressed in all given input microarray data sets. The consensus module application provides further empirical evidence that module preservation statistics and the recommended threshold values provide sufficient statistical power to implicate preserved modules.
In Table 1, we categorize the statistics according to which aspects of module preservation they measure. For example, we present several seemingly different versions of density and connectivity based preservation statistics. But for correlation network modules, close relationships exist between them as illustrated in Figure 8. The hierarchical clustering trees in Figure 8 show the correlations between the observed preservation statistics in our real data applications. As input of hierarchical clustering, we used a dissimilarity between the observed preservation statistics, which was defined as one minus the correlation across all studied reference and test data sets. Overall we observe that statistics within one category tend to cluster together. We also observe that separability appears to be weakly related to the density and connectivity preservation statistics. Cross-tabulation statistics correlate strongly with density and connectivity statistics in the study of human and chimpanzee brain data, but the correlation is weak in the study of sex differences in human brain data.
We derive relationships between module preservations statistics in the sixth section of Supplementary Text S1. In particular, the geometric interpretation of correlation networks [29], [45] can be used to describe situations when close relationship exist among the density based preservation statistics (, , , ), among the connectivity based preservation statistics (, , , ), and between the separability statistics (, ). These relationships justify aggregating the module preservation statistics into composite preservation statistics such as (Equation 1) and (Equation 34).
To illustrate the utility and performance of the proposed methods, we consider 7 different simulation scenarios that were designed to reflect various correlation network applications. An overview of these simulations can be found in Figure 9. A more detailed description of the simulation scenarios is provided below.
Table 3 shows the performance grades of module preservation statistics in the different simulation scenarios. The highest grade of indicates excellent performance. We find that the proposed composite statistics (mean grade ) and (mean grade ) perform very well in distinguishing preserved from non-preserved modules. In contrast, cross-tabulation based statistics only obtain a mean grade of . Since several simulation scenarios test the ability to detect connectivity preservation (as opposed to density preservation), it is no surprise that on average cluster validation statistics do not perform well in these simulations. For example, the IGP cluster validation statistic (Table 4) obtains a mean grade of across the scenarios. But the IGP performs very well (grade 4) when studying the preservation of strongly preserved clusters (scenario 2).
Table 3 also shows the performance of individual preservation statistics. Note that density based preservation statistics perform well in scenarios 1 through 5 but fail in scenarios 6 and 7. On the other hand, all connectivity based statistics perform well in scenarios 6 and 7. The relatively poor performance of is one of the reasons why we did not include it into our composite statistics.
In the following, we describe the different simulation scenarios in more detail.
Additional descriptions of the simulations can be found Supplementary Text S6 and in Supplementary Table S5. As caveat, we mention that we only considered 7 scenarios that aim to emulate selected situations encountered in co-expression networks. The performance of these preservation statistics may change in other scenarios. A comprehensive evaluation in other scenarios is needed but lies beyond our scope. R software tutorials describing the results of our simulation studies can be found on our web page and will allow the reader to compare different methods using our simulated data.
Preservation statistics described in this article have been implemented in the freely available statistical language and environment R. A complete evaluation of observed preservation statistics and their permutation statistics is implemented in function modulePreservation, which is included in the updated WGCNA package originally described in [46]. For each user-defined reference network both preservation and quality statistics are calculated considering each of the remaining networks as test network. Our tutorials illustrate the use of the modulePreservation function on real and simulated data. All data, code and tutorials can be can be downloaded from http://www.genetics.ucla.edu/labs/horvath/CoexpressionNetwork/ModulePreservation.
This article describes powerful module preservation statistics that capture different aspects of module preservation. The network based preservation statistics only assume that each module forms a sub-network of the original network. Thus, we define a module as a subset of nodes with their corresponding adjacencies. In particular, our connectivity preservation statistics (, , , and ) do not assume that modules are defined as clusters. While we have used connectivity based statistics in biologic applications (e.g., modular preservation in human and mouse networks [20], [21]), this article provides the first methodological description and evaluation of these and other module preservation statistics. We also demonstrate that it is advantageous to aggregate multiple preservation statistics into composite statistics and . While we propose module preservation statistics for general networks (e.g., ), all of our applications involve gene co-expression networks.
For a special class of networks, called approximately factorizable networks, one can derive simple relationships between network concepts [29], [45]. Analogously, we characterize correlation modules where simple relationships exist between i) density-based preservation statistics, ii) connectivity based preservation statistics, and iii) separability based preservation statistics (see the sixth section of Supplementary Text S1). We also briefly describe relationships between preservation statistics in general networks.
Table 3 shows the performance grades of module preservation statistics in different simulation scenarios. We find that composite statistics and perform very well in distinguishing preserved from non-preserved modules. While the dependence of on the module size is often attractive, our applications show situations when it is unattractive. In this case, we recommend to use the composite statistic , which has an added bonus: its computation is much faster than that of since it does not involve a permutation test procedure. Our applications provide evidence that the statistic can lead to biologically meaningful preservation rankings among modules.
Our applications provide a glimpse of the types of research questions that can be addressed with the module preservation statistics. In general, methods for quantifying module preservation have several uses. First and foremost they can be used to determine which properties of a network module are preserved in another network. Thus, module preservation statistics are a valuable tool for validation as well as differential network analysis. Second, they can be used to define a global measure of module structure preservation by averaging the preservation statistic across multiple modules or by determining the proportion of modules that are preserved. A third use of module preservation statistics is to define measures of module quality (or robustness), which may inform the module definition. For example, to measure how robustly a module is defined in a given correlation network, one can use resampling techniques to create reference and test sets from the original data and evaluate module preservation across the resulting networks. Thus, any module preservation statistic naturally gives rise to a module quality statistic by applying it to repeated random splits (interpreted as reference and test set) of the data. By averaging the module preservation statistic across multiple random splits of the original data one arrives at a module quality statistic.
We briefly point out situations when alternative procedures may be more appropriate. To identify modules that are present in multiple data sets it can be preferable to consider all data sets simultaneously in a consensus module detection procedure. For example, the consensus module approach described in application 6 results in modules that are present in multiple networks by construction. To identify individual genes that diverge between two data sets, one can use standard discriminative analysis techniques. For example, differentially expressed genes can be found with differential expression analysis and differentially co-expressed genes can be found using differential co-expression analysis [17].
While cluster analysis and network analysis are different approaches for studying high-dimensional data, there are some commonalities. For example, it is straightforward to turn a network adjacency matrix (which is a similarity measure) into a dissimilarity measure which can be used as input of a clustering procedure (e.g., hierarchical clustering or partitioning around medoids) [25]. If a module is defined using a clustering procedure, one can use cluster preservation statistics as module preservation statistics. Conversely, our adjacency based module preservation statistics give rise to cluster preservation statistics since a dissimilarity measure (used for the cluster definition) can also be transformed into a network adjacency matrix. In some of our applications where modules are defined as clusters, we find that is highly correlated with the IGP cluster validation statistic [24] across modules. In our simulations, we observe that IGP and tend to be highly correlated when modules correspond to clusters with varying extents of preservation. This illustrates that leads to sensible results in the special case when modules are defined as clusters. When modules are not defined via a clustering procedure (e.g. in our KEGG pathway application), we find pronounced differences between and the IGP statistic.
The proposed composite preservation statistics and outperform (or tie with) the IGP statistic in all simulation scenarios (see Table 4). More comprehensive comparisons involving additional simulation scenarios and other cluster preservation statistics are needed but lie beyond our scope.
Although not the focus of this work, we mention that a major application of density-based statistics is to measure module quality in the reference data (for example, to compare various module detection procedures). Module quality measures can be defined using density-based and separability-based module preservation measures: the density and separability of a module in the reference network measures its homogeneity and separateness, respectively. In contrast, connectivity based measures (which contrast the reference adjacency matrix with the test network adjacency matrix) are not directly related to module quality measures (unless a data splitting approach is used in the reference data). Module quality measures based on density and separability measures can be used to confirm that the reference modules are well defined. A section in Supplementary Text S1 describes module quality measures that are implemented in the R function modulePreservation.
The proposed preservation statistics have several limitations including the following. First, our statistics only apply to undirected networks. Generalization of our statistics to directed networks is possible but outside of our scope.
A second limitation concerns statistics of connectivity preservation that are based on correlating network adjacencies, intramodular connectivities, etc, between the reference and the test networks. Because Pearson correlation is sensitive to outliers, it may be advantageous to use an outlier-resistant correlation measure, e.g., the Spearman correlation or the biweight midcorrelation [47], [48] implemented in the WGCNA package [46]. Robust correlation options have been implemented in the R function modulePreservation.
A third limitation is that a high value of a preservation statistic does not necessarily imply that the module could be found by a de novo module detection analysis in the test data set. For example, if a module is defined using cluster analysis, then the resulting test set modules may not have significant overlap with the original reference module in a cross-tabulation table. As explained before, this potential limitation is a small price to pay for making a module preservation analysis independent from the vagaries of module detection.
A fourth limitation is that it is difficult to pick thresholds for preservation statistics. To address this issue, we use permutation tests to adjust preservation statistics for random chance by defining Z statistics (Equation 29). The R function modulePreservation also calculates empirical p-values for the preservation statistics. A potential disadvantage of permutation test based preservation statistics (compared to observed statistics and ) is that they typically depend on module sizes. The choice of thresholds is discussed in the Methods section.
A fifth limitation is computational speed when it comes to calculating permutation test based statistics (e.g. ). When only and observed preservation statistics are of interest, we recommend to avoid the computationally intensive permutation test procedure by setting nPermutations = 0 in the modulePreservation function.
A sixth limitation is that the different preservation statistics may disagree with regard to the preservation of a given module. While certain aspects of a module may be preserved, others may not be. In our simulation studies, we present scenarios where connectivity statistics show high preservation but density measures do not and vice versa. Since both types of preservation statistics will be of interest in practice, our R function modulePreservation outputs all preservation statistics. Although we aggregate several preservation statistics into composite statistics, we recommend to consider all of the underlying preservation statistics to determine which aspects of a module are preserved.
While we describe situations when cross-tabulation based preservation statistics are not applicable, we should point out that cross-tabulation statistics also have the following advantages. First, they are often intuitive. Second, they can be applied when no network structure is present. Third, they work well when module assignments are strongly preserved and the modules remain separate in the test network. In the first section of Supplementary Text S1, we describe cross-tabulation based module preservation statistics which we have found to be useful.
We note that the interpretation of gene co-expression relationships depends heavily on biological context. For example, in a dataset consisting of samples from multiple tissue types, co-expression modules (that is, modules defined by co-expression similarity) will often distinguish genes that are expressed in tissue-specific patterns (e.g., [32], [49]). In a dataset consisting of samples from a single tissue type, co-expression modules may distinguish sets of genes that are preferentially expressed in distinct cell types that comprise that tissue (e.g., [30]). In a dataset consisting of samples from a homogeneous cellular population, co-expression modules may correspond more directly to sets of genes that work in tandem to perform various intracellular functions. In many cases, co-expression modules may not present immediate functional interpretations. However, previous work has shown that many co-expression modules are conserved across phylogeny [4], [21], [32], [50], enriched with protein-protein interactions [7], [21], [30], and enriched with specific functional categories of genes, including ribosomal, mitochondrial, synaptic, immune, hypoxic, mitotic, and many others [7], [21], [30], [33].
Although elucidating the functional significance of identified co-expression modules requires substantial effort from biologists and bioinformaticians, the importance of co-expression modules lies not only in their functional interpretation, but also in their reproducibility. Because transcriptome organization in a given biological system is highly reproducible [30], co-expression modules provide a natural framework for comparisons between species, tissues, and pathophysiological states. This framework can reduce dimensionality by approximately three orders of magnitude (e.g., moving from say 40,000 transcripts to 40 modules) [29], [33], while simultaneously placing identified gene expression differences within specific cellular and functional contexts (inasmuch as the cellular and functional contexts of the modules are understood). The co-expression modules themselves are simply summaries of interdependencies that are already present in the data. Preservation statistics can be used to address an important question in co-expression module based analyses: how to show whether the modules are robust and reproducible across data sets.
Given the above-mentioned limitations, it is reassuring that the proposed module preservation statistics perform well in 6 real data applications and in 7 simulation scenarios. Although it would be convenient to have a single statistic and a corresponding threshold value for deciding whether a module is preserved, this simplistic view fails to realize that module preservation should be judged according to multiple criteria (e.g., density preservation, connectivity preservation, etc). Individual preservation statistics provide a more nuanced and detailed view of module preservation. Before deciding on module preservation, the data analyst should decide which aspects of a module preservation are of interest.
Due to space limitations, we have moved our description of cross-tabulation based preservation statistics to the first section of Supplementary Text S1. We briefly mention related measures reported in the literature. Our co-clustering statistic (in the first section of Supplementary Text S1) is similar to the cluster robustness measure [23], [51] and the accuracy based measures are conceptually related to a cluster discrepancy measure proposed in [23]. Cluster validation measures and approaches are reviewed in [52]. Many cross-tabulation based methods have been proposed to compare two clusterings (module assignments), e.g., the Rand index [53] or prediction based statistics [26], [27].
Our methods are applicable to weighted or unweighted networks that are specified by an adjacency matrix , an matrix with entries in . The component encodes the network connection strength between nodes and . In an unweighted network, the nodes , can be either connected () or disconnected (). In a weighted network, the adjacency takes on a value in that encodes the connection strength between the nodes. Networks do not have to be defined with regard to correlations. Instead, they may reflect protein binding information, participation in molecular pathways, etc. In the following, we assume that we are dealing with an undirected network encoded by a symmetric adjacency matrix: . But several of our module preservation statistics can easily be adapted to the case of directed network represented by a non-symmetric adjacency matrix.
To simplify notation, we introduce the function that takes a symmetric matrix and turns it into a vector of non-redundant components,(2)We assume that the diagonal of the matrix is fixed (for example, if is an adjacency matrix, the diagonal is defined to be 1), so we leave the diagonal elements out. Thus, the vector contains components.
A network represented by its adjacency matrix can be characterized by a number of network concepts (also known as network indices) [29], [45]. The network density is the mean adjacency,(3)Higher density means more (or more strongly) interconnected nodes.
The connectivity (also known as degree) of node is defined asThe connectivity of node measures its connection strength with other nodes. The higher the more centrally located is the node in the network.
The Maximum Adjacency Ratio (MAR) [29] of node is defined as(4)The is only useful for distinguishing the connectivity patterns of nodes in a weighted network since it is constant () in unweighted networks.
The clustering coefficient [54] of node is defined as(5)While the clustering coefficient was originally defined for unweighted networks, Equation 5 can be used to extend its definition to weighted networks [5]: one can easily show that implies .
Many network analyses define modules, that is subsets of nodes that form a sub-network in the original network. Modules are labeled by integer labels , and sometimes by color labels. Color labels can be convenient for visualizing modules in network plots. For module with nodes, the dimensional adjacency matrix between the module nodes is denoted by . Denote by the set of node indices of the nodes in module . Network concepts (such as the connectivity, clustering coefficient, MAR etc) defined for are defined as intramodular network concepts. For example, the density of module is defined as the mean adjacency of :(6)The intramodular connectivity of node in module is defined as the sum of connection strengths to other nodes within the same module,(7)Nodes with high intramodular connectivity are referred to as intramodular hub nodes.
Here we describe module preservation statistics that can be used to determine whether a module that is present in a reference network (with adjacency ) can also be found in an independent test network (with adjacency ). Specifically, assume the vector encodes the module assignments in the reference network. Thus () if node is assigned to module . We reserve the label (and color grey) for nodes that are not assigned to any module. For a given module with nodes, the module adjacency matrices are denoted and in the reference and test networks, respectively. We propose network concepts that can be useful for determining whether a module (found in the reference network) is preserved in the test network.
Intuitively, one may call a module preserved if it has a high density in the test network. We define the mean adjacency for module as the module density in the test network,(8)Some of the density statistics such as the mean adjacency are similar to previously described methods based on within-cluster and between-cluster dissimilarities [22]. For example, the mean intramodular adjacency (Equation 8) is oppositely related to the within-module scatter used in assessing the quality of clusters based on a dissimilarity [55]. The network density measure can be considered a generalization of the cluster cohesiveness measure [28] to (possibly weighted) networks.
Other network concepts may be used to obtain a summary statistic of a module. For example, our R function modulePreservation also calculate preservation statistics based on the mean (Equation 5):(9)and mean MAR (Equation 4):(10)in the test network.
Connectivity preservation statistics quantify how similar connectivity of a given module is between a reference and a test network. For example, module connectivity preservation can mean that, within a given module , nodes with a high connection strength in the reference network also exhibit a high connection strength in the test network. This property can be quantified by the correlation of intramodular adjacencies in reference and test networks. Specifically, if the entries of the first adjacency matrix are correlated with those of the second adjacency matrix then the adjacency pattern of the module is preserved in the second network. Therefore, we define the adjacency correlation of the module network as(11)High indicates that adjacencies within the module in the reference and test networks exhibit similar patterns.
If module is preserved in the second network, the highly connected hub nodes in the reference network will often be highly connected hub nodes in the test network. In other words, the intramodular connectivity in the reference network should be highly correlated with the corresponding intramodular connectivity in the test network. Thus, we define the correlation of intramodular connectivities,(12)where and are the vectors of intramodular connectivities of all nodes in module in the reference and test networks, respectively. Analogously, we define the correlation of clustering coefficients and maximum adjacency ratios,(13)(14)
The specific nature of correlation networks allows us to define additional module preservation statistics. The underlying information carried by the sign of the correlation can be used to further refine the statistics irrespective of whether a signed or unsigned similarity is used in network construction. To simplify notation, we define(18)We will use the notation for the correlation matrix restricted to the nodes in module . We define the mean correlation density of module as(19)Thus the correlation measure of module preservation is the mean correlation in the test network multiplied by the sign of the corresponding correlations in the reference network. We note that a correlation that has the same sign in the reference and test networks increases the mean, while a correlation that changes sign decreases the mean. Because the preservation statistic keeps track of the sign of the corresponding correlation in the reference network, we call it the mean sign-aware correlation.
To measure the preservation of connectivity patterns within module between the reference and test networks, we define a correlation-based measure similar to the statistic (Equation 11):(20)In our applications we find that the correlation-based preservation statistic is preferable to its general network counterpart ; therefore, we only report .
Typical values of module preservation statistics depend on many factors, for example on network size, module size, number of observations etc. Thus, instead of attempting to define thresholds for considering a preservation statistic significant, we use permutation tests. Specifically, we randomly permute the module labels in the test network and calculate corresponding preservation statistics. This procedure is repeated times. For each statistic labeled by index we then calculate the mean and the standard deviation of the permuted values. We define the corresponding statistic as(29)where is the observed value for the statistic . Under certain conditions, one can prove that under the null hypothesis of no preservation the statistic asymptotically follows the standard normal distribution . Thus, under the assumption that the number of permutations is large enough to approximate the asymptotic regime, one can convert the statistics to p-values using the standard normal distribution. Our R function modulePreservation outputs the asymptotic p-values for each statistic. But we should point out that it would be preferable to use a full permutation test to calculate permutation test p-values. We often report Z statistics (instead of p-values) for the following two reasons: First, permutation p-values of preserved modules are often astronomically significant (say ) and it is more convenient to report the results on a Z scale. The second reason is computational speed. The calculation of a Z statistic only requires one to estimate the mean and variance under the null hypothesis, for which fewer permutations are needed. To estimate a permutation test p-value accurately would require computational time far beyond practical limits.
In the sixth section of Supplementary Text S1, we describe when close relationships exist between many of the preservation statistics presented above. This suggests that one can combine the individual preservation statistics into a composite preservation statistic. We propose two composite preservation statistics. The first composite statistic (Equation 1) summarizes the individual Z statistic values that result from the permutation test. The second composite statistic (Equation 34) summarizes the ranks of the observed preservation statistics.
The relationships derived in Supplementary Text S1 suggest to summarize the density based preservation statistics as follows:(30)Similarly, the connectivity based preservation statistics can be summarized as follows:(31)When density and connectivity based preservation statistics are equally important for judging the preservation of a network module, one can consider the composite summary statistic (Eq. 1)Alternatively, a weighted average between and can be formed to emphasize different aspects of module preservation. Future research could investigate alternative ways of aggregating preservation statistics. While our simulations and applications show that works well for distinguishing preserved from non-preserved modules, we do not claim that it is optimal. In practice, we recommend to consider all individual preservation statistics.
Our simulated as well as empirical data show that the separability tends to have low agreement (as measured by correlation) with the other preservation statistics (Figure 8). Since the statistic often performs poorly, we did not include it in our composite statistics.
Since is not a permutation statistic but rather the median of other statistics, we do not use it to calculate a p-value. Instead, the R function modulePreservation calculates a summary p-value () as follows. For each permutation Z statistic, it calculates the corresponding p-value assuming that, under the null, the Z statistic has a normal distribution . The normal distribution can be justified using relatively weak assumptions described in statistics textbooks. As a caveat, we mention that we use preservation p-values as descriptive (and not inferential) measures. On the other hand, we cannot assume normality for . Hence, instead of calculating a p-value corresponding to , we calculate a summary log-p-value instead, given as the median of the log-p-values of the corresponding permutation statistics. Because of the often extremely significant p-values associated with the permutation statistics, it is desirable to use logarithms (here base 10). We emphasize that the summary log-p-value is not directly associated with ; rather, it is a separate descriptive summary statistic that summarizes the p-values of the individual permutation statistics.
It seems intuitive to call a module with preserved, but our simulation studies argue for a more stringent threshold. We recommend the following threshold guidelines: if , there is strong evidence that the module is preserved. If there is weak to moderate evidence of preservation. If , there is no evidence that the module preserved. As discussed below, these threshold values should be considered rough guidelines since more (or less) stringent thresholds may be required depending on the application.
The modulePreservation R function calculates multiple preservation statistics and corresponding asymptotic p-values. Similar to the case of statistics, a threshold that is appropriate in one context may not be appropriate in another. The choice of thresholds depends not only on the desired significance level but also on the research question. When several preservation statistics are analyzed individually for any indication of module preservation then the threshold should correct for the these multiple comparisons. Since several “tests” for preservation are considered, an obvious choice is to use one of the standard correction approaches (e.g., Bonferroni correction) for determining the threshold that should be put on multiple tests. Toward this end, one can use the uncorrected, individual preservation statistics and p-values output by the modulePreservation function. A Bonferroni correction would be a conservative but probably too stringent approach in light of the fact that many of the preservation statistics are closely related (see the 6th section in Supplementary Text S1). Given the strong relationships among some preservation statistics, we have found it useful to aggregate the statistics (and optionally the empirical p-values) in a statistically robust fashion using the median but many alternative procedures are possible. To provide some guidance, we recommend thresholds for that we have found useful in our simulations studies (Supplementary Text S6) and in our empirical studies.
In some applications such as the human vs. chimpanzee comparison described above, one is interested in ranking modules by their overall preservation in the test set, i.e., one is interested in a relative measure of module preservation. Since our simulations and applications reveal that (Equation 1) strongly depends on module size, this statistic may not be appropriate when comparing modules of very different sizes. Here we define an alternative rank-based measure that relies on observed preservation statistics rather than the permutation statistics. For each statistic , we rank the modules based on the observed values . Thus, each module is assigned a rank for each observed statistic. We then define the median density and connectivity ranks(32)(33)Analogously to the definition of , we then define the statistic as the mean of and ,(34)Alternatively, a weighted average of the ranks could be formed to emphasize different aspects of module preservation. It is worth repeating that a composite rank preservation statistic is only useful for studying the relative preservation of modules, e.g., we use for studying which human brain co-expression modules are least preserved in chimpanzee brain networks.
While all examples in this article relate to correlation (in particular, co-expression) networks, we have also implemented methods and R function that can be applied to general networks specified only by an adjacency matrix. For example, this function could be used to study module preservation in protein-protein interaction networks. We also define a composite statistic , which is defined for a general network specified by an adjacency matrix (Eq. 35).(35)where and . Note that is only computed with regard to a subset of the individual statistics. To invoke this preservation statistic, set dataIsExpr = FALSE in the modulePreservation R function.
A detailed description of the methods is provided Supplementary Text S1 which contains the following sections. In the first section of Supplementary Text S1, we describe standard cross-tabulation based module preservation statistics. Specifically, we present three basic cross-tabulation based statistics for determining whether modules in a reference data set are preserved in a test data set. These statistics do not assume that a test network is available. Instead, module assignments in both the reference and the test networks are needed.
In the second section, we briefly review a hierarchical clustering procedure for module detection. Many methods exist for defining network modules. In this section, we describe the method used in our applications but it is worth repeating that our preservation statistics apply to most alternative module detection procedures.
In the third section, we review the definition of signed and unsigned correlation networks. Correlation networks are a special case of general undirected networks in which the adjacency is constructed on the basis of correlations between quantitative variables.
In the fourth section, we present module quality statistics, which we are implemented in the modulePreservation R function. While our main article focuses on statistics that measure preservation of modules between a reference and a test network, we briefly discuss the application of some of the preservation statistics to the related but distinct task of measuring module quality in a single (reference) network. More precisely, the density and separability statistics can be applied to the reference network without a reference to a test network. The results can then be interpreted as measuring module quality, that is how closely interconnected the nodes of a module are or how well a module is separated from other modules in the network.
In the fifth section, we review the notation for the singular value decomposition and for defining a module eigennnode. The section describes conditions when the eigenvector is an optimal way of representing a correlation module. It also reviews the definition of (the proportion of the variance explained by the eigennode). We derive a relationship between and the module membership measures , which will be useful for deriving relationships between preservation statistics.
In the sixth section, we investigate relationships between preservation statistics in correlation networks.
The KEGG database and many textbooks describe these fundamental pathways in more detail but the following terse descriptions may be helpful. The Wnt signaling pathway describes a network of proteins most well known for their roles in embryogenesis and cancer, but also involved in normal physiological processes in adult animals. The Hedgehog signaling pathway is one of the key regulators of animal development conserved from flies to humans. The apoptosis pathway mediates programmed cell death. Endocytosis is the process by which cells absorb molecules (such as proteins) from outside the cell by engulfing them with their cell membrane. The Transforming growth factor beta (TGF-) signaling pathway is involved in many cellular processes in both the adult organism and the developing embryo including cell growth, cell differentiation, apoptosis, cellular homeostasis and other cellular functions. The Phosphatidylinositol signaling system facilitates environmental information processing and signal transduction. The mitogen-activated protein kinase (MAPK) cascade is a highly conserved pathway that is involved in various cellular functions, including cell proliferation, differentiation and migration. The Calcium signaling pathway describes how calcium can act in signal transduction after influx resulting from activation of ion channels, or as a second messenger caused by indirect signal transduction pathways such as G protein-coupled receptors.
|
10.1371/journal.ppat.1003134 | Hospital-Community Interactions Foster Coexistence between Methicillin-Resistant Strains of Staphylococcus aureus | Methicillin-resistant Staphylococcus aureus (MRSA) is an important cause of morbidity and mortality in both hospitals and the community. Traditionally, MRSA was mainly hospital-associated (HA-MRSA), but in the past decade community-associated strains (CA-MRSA) have spread widely. CA-MRSA strains seem to have significantly lower biological costs of resistance, and hence it has been speculated that they may replace HA-MRSA strains in the hospital. Such a replacement could potentially have major consequences for public health, as there are differences in the resistance spectra of the two strains as well as possible differences in their clinical effects. Here we assess the impact of competition between HA- and CA-MRSA using epidemiological models which integrate realistic data on drug-usage frequencies, resistance profiles, contact, and age structures. By explicitly accounting for the differing antibiotic usage frequencies in the hospital and the community, we find that coexistence between the strains is a possible outcome, as selection favors CA-MRSA in the community, because of its lower cost of resistance, while it favors HA-MRSA in the hospital, because of its broader resistance spectrum. Incorporating realistic degrees of age- and treatment-structure into the model significantly increases the parameter ranges over which coexistence is possible. Thus, our results indicate that the large heterogeneities existing in human populations make coexistence between hospital- and community-associated strains of MRSA a likely outcome.
| One of the most notorious cases of antibiotic-resistant bacteria is methicillin-resistant Staphylococcus aureus (MRSA), which causes diseases ranging from skin and soft-tissue infections to pneumonia and septicemia. Traditionally, MRSA was mainly hospital-associated, but in the past decade community-associated strains have spread widely. Typically drug-resistant bacteria have lower reproduction or transmission rates, called a fitness cost. Because this cost is estimated to be significantly lower for community-associated strains, it has been predicted that these will eventually replace the hospital-associated strains. However, hospital-associated strains are resistant against a greater variety of antibiotics, which may compensate for the higher fitness cost. Here, we integrate realistic data on drug-usage, resistance profiles, contact, and age structures into a mathematical model of MRSA transmission to predict the competition between hospital- and community-associated strains. We find that for a realistic degree of population structure it is likely that both strains of MRSA will coexist in the long term. This results from significantly different hospitalization and antibiotic consumption rates between age groups. In particular, elderly individuals have much higher rates of antibiotic usage and hospitalizations than other age groups. This generates a situation where community-associated strains can predominate in the community but are outcompeted in the hospital, resulting in coexistence in the population.
| Over the past ten years community-associated strains of methicillin-resistant Staphylococcus aureus (CA-MRSA) have emerged and spread rapidly, accounting for large increases in disease both in the community and in the hospital [1], [2]. While originally thought to be primarily a hospital-associated pathogen (HA-MRSA), the emergence of a community-associated strain, which has a different genetic background [2] and drug susceptibility profile [3], [4], has raised questions about how the epidemiology and the ecology of the disease will evolve, particularly with respect to which strain will predominate.
MRSA resistance is mediated by the integration of a staphylococcal cassette chromosome mec (SCCmec) in a site-specific manner into the staphylococcal genome [5]. Eight different SCCmec allotypes, as well as numerous subtypes, which encode varying levels of resistance to multiple antibiotics, have been described to date [2]. In vitro growth assays have demonstrated an inverse correlation between resistance level and growth rate [6], [7], which presumably limits the spread of HA-MRSA strains – characterized by high levels of resistance to multiple antibiotics –beyond hospitals [8], [9]. On the other hand, the maintenance of resistance in the most common community-associated strains, which are characterized by resistance to only a limited set of antibiotics, seems to cause only a negligible reduction in the growth rate relative to non-resistant strains [10], [11]. This in turn accounts for its wide dissemination in the community and the assumption that it is more easily transmitted than HA-MRSA strains [2].
Given the presumed relatively lower cost of resistance in CA-MRSA, it has been speculated that CA-MRSA strains may eventually replace HA-MRSA strains in the hospital [12], [13]. However, the high costs of resistance for HA-MRSA strains might be offset by the fact that they are resistant to a much broader range of antibiotics than CA-MRSA strains [3], [4]. If this were the case, coexistence between the two strains might be a plausible long-term scenario. In fact, a recent empirical study [14] indicates that despite a substantial increase in patients infected with CA-MRSA strains, the frequency of HA-MRSA strains in the hospital has remained remarkably stable.
Intuitively, both outcomes, coexistence and replacement, are possible. On the one hand, replacement, or competitive exclusion, is the standard outcome expected by ecological theory for two strains occupying the same ecological niche. Accordingly, explaining observed coexistence in other bacterial pathogens has proven challenging [15], [16]. On the other hand, structural differences in hospital and community populations may impose sufficiently different selective advantages to allow coexistence. Thus, HA-MRSA with its broad resistance spectrum may be better adapted to the hospital environment where antibiotic use is common, whereas CA-MRSA may be better adapted to the community, where antibiotic use is less frequent and hence a broad resistance spectrum cannot compensate for the reduced transmissibility of HA-MRSA. One factor, which may counteract this mechanism, is that the average length of stay of a patient in the hospital (4–5 days) is much shorter than the time until a patient colonized with MRSA clears the bacteria (several 100 days [17]). Accordingly, a given colonization (bacterial population within a host) almost never experiences only the hospital environment, hence making local adaptation difficult. It follows from the above reasoning that addressing the question of coexistence is not possible from the hospital perspective alone, but instead it is necessary to take both hospital and community populations (and their frequent exchange) into account.
While this question of coexistence is an interesting ecological problem, it is also an important question for public health as the outcome of the interaction between CA-MRSA and HA-MRSA may have epidemiological and clinical consequences: HA-MRSA has a much broader resistance spectrum than CA-MRSA [18] and may therefore be more difficult to treat. Moreover, the two strains also differ with respect to their pathogenicity. While CA-MRSA has primarily been associated with skin- and soft-tissue infections, there have also been suggestions that it may be more invasive and virulent than HA-MRSA [2]. If CA-MRSA completely replaces HA-MRSA, hospitals might be confronted with a more virulent but also more treatable pathogen. Moreover, their ability to replicate and transmit in the community may mean significantly more infections as well. In this analysis, we use epidemiological models which integrate realistic data on drug-usage frequencies, resistance profiles, contact, and age structures to assess the impact of competition between HA- and CA-MRSA strains. In particular, we examine the likelihood of coexistence as an outcome.
We considered three epidemiological models of increasing complexity and assessed how the interaction between hospital and community populations could lead to stable coexistence between HA- and CA-MRSA. The basic model assumes that all individuals, regardless of age, have similar hospital admission and discharge rates as well as antibiotic usage rates. In this case the only difference between the hospital and the community is the usage frequency of antibiotics, which may lead to selection favoring one strain in the community and the other in the hospital. Next we considered two extensions of this model. First we examined how heterogeneity between age-classes with respect to hospitalization rates and antibiotic usage impacted coexistence between strains in each setting. Second, we explicitly distinguished between treated and untreated patients, thereby capturing the prophylactic effect of treatment, which is likely to be much stronger in the hospital than in the community.
The epidemiological dynamics of CA- and HA-MRSA were described by set of ordinary differential equations. Based on estimates from the literature and an analysis of public-use data in the United States, we assume differing dynamics of colonization, infection, and antibiotic use between the hospital and the community and consider how differing implementations of the host population structure impacts the dynamics of each strain and examine the parameter ranges over which coexistence occurs. In each of these models, the possibility of coexistence between HA- and CA-MRSA for a given parameter-combination was determined by an invasibility analysis: First, the system is allowed to reach the equilibrium with one strain only (burn-in time: 5×104 days); then the other strain is introduced at low abundance; if the introduced strain increases in frequency after the introduction, we say that it can invade the equilibrium determined by the resident strain. If both strains (HA-MRSA and CA-MRSA) can invade the equilibrium of the other strain, this indicates coexistence.
In the basic model we assume that populations in the hospital and community are homogenous. The two populations are connected through admission into the hospital population (with rate a) and discharge into the community with rate (d) and are each subdivided into individuals that are uncolonized (SC and SH), colonized with CA-MRSA (CC,CA and CH,CA), and colonized with HA-MRSA (CC,CA and CH,CA). In the community, colonized individuals infect uncolonized individuals with a transmission rate βC,CA for CA-MRSA and βC,HA for HA-MRSA. In the hospital the corresponding rates are βH,CA and βH,HA. We assume that regardless of location, in the absence of treatment HA-MRSA suffers a fitness cost (s) against CA-MRSA, which is assumed to be due to a reduced transmission rate; i.e. βX,HA = βX,CA(1-s). Colonization can be cleared either spontaneously, with rate cBL, or through treatment. Since colonization is significantly more common than infection, we assume that antibiotic use is independent of the colonization status of the patient. Antibiotic use occurs at rate τC and τH in the hospital and community, respectively. As treatment is in most cases not specific for S. aureus, we assume that the probability (fC,X and fH,X) that a given course of treatment is effective against either strain corresponds to the proportion of drugs to which each is susceptible among all prescribed drugs (see parameters section). We further assume that even if the drug consumed is effective, there are reasons other than antibiotic resistance that a bacterial colonization may not be cleared, i.e. effectively treated patients clear the bacterial population only with a probability cT<1. The basic model is described by the following set of ordinary differential equations:
The age-structured model is derived from the basic model by sub-dividing each compartment into 18 different age classes (five-year bins for the ages from 0 to 85 and one bin for 85+). For instance, the compartment SC (uninfected individuals in the community) is subdivided into the compartments SC1, SC2… SC18, where SC1 covers susceptible individuals of age 0–5, SC2 of 5–10, etc. Admission rates, aj, and discharge, dj, rates are assumed to depend on the age class j (see parameters section). The impact of age structure on contact rates is captured by assuming that the transmission rates in the community, βC,Xj,k, are proportional to the frequency of physical contacts (divided by the number of people in that age class). Because detailed contact data from the US were not available, we used data on contact rates measured in the UK [19]. Finally, the treatment rate τCj in the community also varies with age class j. Similar data on the age-dependency of contact rates and treatment rates in the hospital were not available to our knowledge. Because contact rates are likely to be more uniform in the hospital, since most of transmission is indirect, we made the conservative assumption that contact and treatment rates in the hospital are uniform regardless of age. The age-structured model is described by the following set of ordinary differential equations:A diagram of the age-structured model is found in Figure 1.
The treatment-structured model (Equation S1 in the online supplementary material) is derived from the basic model by subdividing each compartment according to treatment status. Specifically, we distinguish between four treatment classes: 1) Untreated (U); 2) treated with a drug that is effective against neither CA-MRSA nor HA-MRSA (treatment T(1)); 3) treated with a drug that is effective against CA-MRSA but not HA-MRSA (treatment T(2)); and 4) treated with a drug that is effective against both CA-MRSA and HA-MRSA (treatment T(3)). Treatment with a drug from class j is initiated at rate τCj and τHj. Upon treatment initiation with an effective drug, the infection is either cleared immediately with a probability cT or alternatively remains colonized (this is an approximation to the real dynamics in which the patient would clear after a given amount of time). Finally, treated individuals stop treatment at a rate ρ (i.e. the inverse duration of antibiotic use). A diagram of the model is found in Figure S1.
The treatment-and-age-structured model is derived from the treatment-structured model in the same way the age-structured model is derived from the basic model.
Our models integrate realistic values for drug-usage frequencies, resistance profiles, age structure, age-dependent contact patterns, and hospitalization rates. Usage frequencies, age distribution, hospitalization rates, and the mean length of stay in the hospital were estimated from publicly available data by five-year age groups. Figure 2 summarizes the age-dependency of population size, hospitalization rates, length of stay in the hospital, and antibiotic usage rates in the community. As data on the age-dependency of treatment and contact rates are available for the community only, we make the conservative assumption that these rates are homogenous in the hospital (see discussion). In contrast to these parameters there is large uncertainty concerning the magnitude of transmission rates (particularly in the hospital) and especially concerning the degree to which the transmission rate of HA-MRSA is reduced compared to that of CA-MRSA (i.e. the selective cost of HA-MRSA). Therefore we vary these parameters over broad ranges. A summary of the parameter values and ranges can be found in Table 1.
The number of hospitalizations and average length of stay for each age group was estimated from the Nationwide Inpatient Sample (NIS), Healthcare Cost and Utilization Project, Agency for Healthcare Research and Quality for the year 2008. The NIS contains data on ∼8 million records of hospital stays annually from about 1,000 hospitals, approximating a 20% stratified sample of US community hospitals, and includes all nonfederal, short-term, general, and specialty hospitals, such as obstetrics-gynecology, ear-nose-throat, orthopedic, and pediatric institutions. The NIS includes public hospitals and academic medical centers but excludes long- and short-term acute rehabilitation facilities, psychiatric hospitals, and alcoholism and chemical dependency treatment facilities. Hospitalization rates were calculated as the average number of hospitalizations per-person per-day by age group. The numbers of individuals for each age group were obtained from the US Census bureau's annual estimates of the resident population by five-year age groups (www.census.gov).
Antibiotic usage in the community was estimated based on data from the National Ambulatory Medical Care Survey (NAMCS) and the National Hospital Ambulatory Medical Care Survey (NHAMCS) for 2008. NAMCS is an annual national survey of visits to non-federally employed office-based physicians who are primarily engaged in direct patient care, and NHAMCS is designed to collect data on the utilization and provision of ambulatory care services in hospital emergency and outpatient departments and in ambulatory surgery centers. Weighted patient level data was used to estimate the annual number of prescriptions for antibiotics that were written for each age group. The usage rate was calculated as the average number of prescriptions written per person per day per age group. Antibiotic usage in the hospital was estimated based on the data from [20], which reported antibiotic use from 130 US hospitals (see Table 1).
To calculate the approximate effectiveness of community antibiotic usage on CA- and HA-MRSA, we calculated the number of prescriptions for each antibiotic class and, based on assumptions about the effectiveness of each antibiotic against CA- and HA-MRSA, we estimated the percentage of drug usage that was effective against each pathogen (Supplementary Table S1). The effectiveness of drugs used in the hospital is similar to the community though skewed towards some of the more effective drugs [20], [21]. Thus, we assume that the effectiveness of antibiotic usage on CA- and HA-MRSA is slightly more effective in the hospital than in the community (see Table 1).
In order to explore the possibility of coexistence between HA-MRSA and CA-MRSA, we consider a series of epidemiological models of increasing complexity. The simplest, basic model ignores all population structure beyond the distinction between hospital and community. We then extend this basic model by incorporating age- and treatment-heterogeneities in accordance with published data (see Methods).
For the basic model, which ignores both age- and treatment-structure, we find that the interaction between the hospital and the community can in principle generate coexistence between HA- and CA-MRSA. However, we observed this outcome only for a relatively narrow band of fitness-costs for HA-MRSA (Figure 3). Moreover, the width of this band decreases with decreasing transmissibility in the hospital, which we quantified as the average number of secondary cases caused by the admission of one patient to the hospital containing only susceptible patients (y-axis in Figure 3). Of note, the parameter range for which coexistence occurs is much narrower than the parameter range for which HA-MRSA is fitter than CA-MRSA in the hospital but less fit than CA-MRSA in the community (see red bars in Figure 3). Thus opposite directions of selection are not a sufficient condition for coexistence. This is due to the significant linkage between the two systems caused by the high-turnover (admission/discharge) in the hospital, which results in the two strains being cycled into and out of the community, making it harder for the two strains to coexist. If the cost of HA-MRSA exceeds the values in the coexistence range, the equilibrium in which only CA-MRSA is present becomes stable; if the cost is smaller, the equilibrium in which HA-MRSA can exclude CA-MRSA becomes stable. Such a narrow range of coexistence is not unexpected for a homogeneous model. However, the epidemiology of MRSA exhibits several heterogeneities which can stabilize coexistence over a broader range of conditions.
We included age-dependent transmission rates for the community by assuming that transmission rates are proportional to the rate of physical contact [19]. Including age structure in this manner substantially broadens the parameter range over which HA- and CA-MRSA can coexist (Figure 4). This increase is due in part to relative differences in hospitalization between younger and older individuals, which changes the relative difference in selection between HA- and CA-MRSA strains. Because the hospital admission rate and the average length of stay increases as individuals age, older individuals are more likely to spend time in the hospital, where MRSA is favored, and consequently they are more likely to be colonized with HA-MRSA which increases the range over which HA-MRSA is able to persist despite an influx of CA-MRSA from the community. Moreover, the number of physical contacts an older person has in the community is considerably lower than the corresponding number for young persons. This in turn further reduces the selective advantage of CA-MRSA in old age classes. As physical contact occurs preferentially between members of the same or neighboring age classes [19], this further contributes to maintaining the association between age and strain.
An additional source of heterogeneity is treatment itself. We take this heterogeneity into account by explicitly tracking the treatment status of patients and assuming that individuals receiving a given antibiotic cannot be colonized with strains that are susceptible to this drug. Including treatment heterogeneity in this way leads to an additional, substantial extension of the parameter range over which coexistence occurs (Figure 5). This is because antibiotic prophylaxis of colonization creates a substantial additional selective advantage for HA-MRSA (which has the broader resistance spectrum) in the hospital. However, whereas the fraction of protected patients is large in the hospital, it is negligible in the community and hence prophylaxis does not substantially increase the fitness of HA-MRSA in the community. Treatment heterogeneity and age-structure act synergistically to increase coexistence, such that the broadest coexistence range can be observed for the treatment- and age-structured model (see Figure 5).
HA-MRSA is most likely adapted to the hospital environment in other ways than by its broad antibiotic resistance spectrum (e.g. tolerance to disinfectants, smaller requirements of invasibility, etc.) [2], [22]. Accordingly, it is likely better able to compete against CA-MRSA (in the absence of therapy) in the hospital as opposed to the community. Taking this effect into account, we find that as the fitness-cost of HA-MRSA in the hospital decreases relative to CA-MRSA, the maximal fitness cost of HA-MRSA for which coexistence occurs is strongly increased. By contrast the minimal cost for coexistence changes only weakly, because reducing the cost of resistance in the hospital does not affect relative fitness in the community (Figure 6). Thus, context specific fitness costs further facilitate coexistence between HA- and CA-MRSA. It is remarkable that decreasing the cost of HA-MRSA in the hospital has a much stronger effect in the presence of age structure than in its absence (which indicates that age structure helps separate the hospital from the community). Thus there is a synergistic effect between age-structure and hospital specific reduction of fitness costs.
The above analysis was based on the ability of one strain to invade the equilibrium defined by the presence of the other strain. This method indicates where the two strains can coexist at equilibrium and therefore allows one to assess the main ecological forces underlying coexistence and competitive exclusion. However, it has three disadvantages: First, the equilibrium might be attained only very slowly: for instance two strains might coexist for a transient period which can extend over decades even though the equilibrium analysis indicates that one strain should exclude the other. Second, even if the two strains coexist one of them might attain only very low levels (i.e. even though the two strains can coexist in theory, almost all infections are caused by one single strain). Third, the pairwise-invasibility approach only allows an analysis of the competitive interaction of two strains, whereas, in reality, several S. aureus strains compete with each other: Notably, HA-MRSA and CA-MRSA compete with methicillin sensitive S. aureus (MSSA), which could modify their interaction.
In order to address these issues, we considered a more pragmatic definition of coexistence: We initiated the population either with HA-MRSA as the only resident strain or with two resident strains (HA-MRSA and MSSA) and ran the simulation for 30 years. Then we add the new strain (CA-MRSA) and examined how the frequencies of each changed over time. Specifically, we tested after 10, 20, 50, 100, and 200 years, which strains still exist in substantial frequency (using a threshold of 5%). Note that we focused here only on the invasion of HA-MRSA/MSSA equilibrium by CA-MRSA rather than the opposite, since the former describes the current epidemic development (whereas the latter is merely of theoretical interest).
We first considered the interaction between CA-MRSA and HA-MRSA (in analogy to the above analyses). We find that the two strains can coexist during a long transient phase (10–50 years) for a broad range of conditions, which do not support coexistence at equilibrium (see Figure 7). Moreover, we can substantially reduce the range of realistic parameters by considering the interaction between HA-MRSA and MSSA: as we know that HA-MRSA has attained substantial frequencies (in the USA at least) after <50 years of methicillin use, the model is only consistent with reality for those parameter-combinations for which this is the case. Figure S2 shows that an invasion of HA-MRSA into the MSSA equilibrium is only possible if the fitness cost of HA-MRSA is below a threshold that is dependent upon the average number of secondary cases caused by the admission of one patient to the hospital containing only susceptible patients (R0HA,H). This threshold is indicated in Figure 7 by the dashed orange line. As it is also a fact that CA-MRSA was able to invade HA-MRSA, the realistic parameter range in Figure 7 is delimited by the dark grey area (corresponding to parameter values where the CA-MRSA invasion is impossible) to the left and the orange line to the right. Thus, Figure 7 indicates that we would expect a long-term coexistence between HA-MRSA and CA-MRSA for most realistic parameter combinations.
When we consider the interaction between all three strains by including MSSA as one of the initial resident strains, we find that the parameter range in which all three strains can coexist shrinks successively with increasing time (see Figure 8) and eventually vanishes (results not shown). This is not unexpected, as the model structure assumes that the hospital and the community are two different ecological niches, which can thus maximally support the coexistence of only two strains over the long-term. However, we do find that all three strains can coexist for a broad range of conditions during a long transient time-span of several decades. Overall, these results indicate that transient effects can strongly extend the range of coexistence, and even allow for long-term de-facto coexistence where this would not be expected at equilibrium.
We examined how differences in age-structured patterns of antibiotic use and hospitalization rates can promote coexistence of CA- and HA-MRSA. Overall, our results show that hospital and community-associated strains of MRSA can coexist if the broader resistance spectrum of the hospital-associated strains is balanced by intermediate fitness-disadvantages in the absence of treatment. For such intermediate fitness costs, the hospital-associated strains have higher fitness in the hospital, where treatment rates are high, whereas community-associated strains have a higher fitness in the community were treatment rates are low. Despite opposite directions of selection, both strains are present in both environments if there is coexistence at all (see Figure S3 for example runs). This occurs because of the high rates of discharge and hospitalization, which cycle individuals between the hospital and the community. Moreover, our results also indicate that opposite directions of selection are not sufficient for maintaining coexistence. This is especially true for our basic model describing well-mixed populations in the hospital and community, in which we found coexistence only for a very narrow range of HA-MRSA fitness-costs.
Including heterogeneity in the form of realistic age- and treatment-structures into the model significantly increases the range of parameters over which coexistence can occur, making it a likely outcome. Furthermore, the fitness cost of HA-MRSA in the absence of treatment is presumably weaker in the hospital than in the community because of factors such as easier invasion due to open wounds, catheters, etc., as well as increased use of antiseptics to which the hospital strain might be better adapted. Taking this possibility into account leads to an additional, substantial increase in the range over which coexistence is likely. Thus, coexistence between HA-MRSA and CA-MRSA is a likely outcome due to the combined effect of hospital-community interactions, age-structure, treatment-structure, and possibly setting dependent fitness costs in the absence of treatment.
Coexistence is mainly dependent upon the cost of HA-MRSA being neither too high nor too low. It should be noted, however, that the upper bound for resistance costs is, in this context, more informative than the lower bound. For costs of HA-MRSA below the lower bound, we would expect that CA-MRSA could not invade the HA-MRSA equilibrium. However, such an invasion is exactly what occurred during the 1990s. Thus, we know that fitness-costs of HA-MRSA are high enough to allow the invasion of CA-MRSA. The crucial question is whether they are low enough for this invasion to stop before CA-MRSA has completely replaced HA-MRSA.
The width of the coexistence range depends strongly on how effectively MRSA can transmit in the hospital. In our simulations we quantified this transmissibility as the average number of secondary cases caused by the admission of one patient to the hospital containing only susceptible patients (RA). If this value is considerably smaller than one (i.e. hospitals cannot maintain the spread of MRSA on their own), then the coexistence range becomes very narrow. This is because coexistence relies on opposite directions of selection in the hospital and community environment. If however, one of these environments does contribute only very weakly to transmission, this balancing effect cannot take place. The only published estimate for RA we are aware of found values of 0.68 (0.47–0.95) and 0.93 (0.71–1.21) for two Dutch hospitals; one implies a broad and one a narrow coexistence range (The same study also reported an RA value of 0.16, which however corresponded to an animal derived strain) [23]. Because the Netherlands has been exceptionally successful in reducing nosocomial spread of MRSA [23], [24], RA values are likely to be higher (and hence coexistence ranges broader) in most other settings. The sensitivity on RA also implies that in regions with better infection control in hospitals (and hence lower RA) one would expect CA-MRSA to completely replace HA-MRSA and hence also to cause most MRSA infections in hospitals.
Even though our model realistically includes several levels of population structure, our analysis might still underestimate the range over which CA-MRSA and HA-MRSA can coexist. First, other types of heterogeneity might promote coexistence in a similar way as the ones discussed here. Examples include spatial heterogeneities like rural vs. urban areas, small vs. large hospitals (which would impose different levels of stochastic effects and thereby affect strain abundances [25]), the cycling of older patients into long-term care facilities [26], or the highly variable length of time individuals remain colonized [27]. We also neglected (due to the absence of data) age- or department-structured antibiotic usage rates in hospitals, though this could further promote coexistence. Temporal heterogeneity, such as the seasonal use of antibiotics might be an additional factor contributing to coexistence in MRSA [28]. We have also broadly categorized the multitude of different MRSA strains as either CA- or HA-MRSA. This diversity could also contribute to coexistence, as different strains may have different resistance phenotypes (It should be noted however that explaining the coexistence of such individual strains is an additional challenge). In addition to such heterogeneities, coexistence might be facilitated by co-infection with different strains [29], [30], either through co-colonization of the nares [30] or specialization of different strains to different anatomical sites. For instance, CA-MRSA primarily causes infections of the skin, whereas HA-MRSA infections are generally more invasive [2], [14]. However, it is not clear to what extent different MRSA strains can co-infect a host, and it has also been shown in a different context that co-infection leads only under very specific conditions to coexistence [15], [16]. Moreover, other studies have shown that colonization with MSSA can be protective from MRSA [31], [32], suggesting that competition may limit the extent of co-colonization with different strains.
Even though our model can explain the coexistence between HA-MRSA and CA-MRSA, we did not find any parameter combination that supports coexistence at equilibrium between more than two strains (HA-MRSA, CA-MRSA and MSSA). This suggests that the system as described by our model corresponds to only two ecological niches. This implies that the maintenance of the diversity within HA-MRSA and CA-MRSA has to be explained by mechanisms not included in our model (such as the geographical and temporal variation mentioned above). Moreover, Figures 7, 8, and S3 also indicate that the system approaches equilibrium only very slowly, such that a long transient maintenance of this diversity is conceivable even if it would not persist in an equilibrium state.
Our model also describes a static situation in which the properties of the strains and the age structure do not change over time. However, both demographic change in the human population and evolutionary change of the MRSA strains are likely to occur and their impact on coexistence between competing strains is an interesting question for future studies. Demographic change will most likely increase the proportion of old people in the US and most western countries. In the context of our model this means that selection will tend to favor hospital adapted strains, as the hospitalization rates are considerably higher for the old age classes. However, the direction of evolutionary change depends very strongly on the physiological constraints underlying antibiotic resistance. For instance, if CA-MRSA can increase its resistance spectrum while maintaining a high transmissibility, it could eventually out-compete HA-MRSA. If on the other hand a higher fitness cost is the inevitable consequence of a broad resistance spectrum, then such a replacement is unlikely to occur. Such evolutionary changes may be particularly important given the very long transient phases during which CA- and HA-MRSA can coexist. These long transient phases provide the opportunity for evolutionary adaptation of the inferior strain (by way of compensatory mutations or extension of the resistance spectrum), which could allow it to persist, even though coexistence is not expected on the basis of the current pathogen fitness.
The classical ecological paradigm of niche overlap states that two species can coexist if their resource usage differs sufficiently [33]. The present study represents an application of those concepts to the important public health question of whether hospital- and community-associated strains of MRSA are expected to coexist in the long-term. An eventual replacement of HA-MRSA by CA-MRSA could cause important changes in the epidemiology of S. aureus. CA-MRSA can more readily cause infections in healthy individuals than HA-MRSA [18] and hence symptomatic MRSA infections could extend to a broader class of patients. CA-MRSA strains have also been associated with a higher virulence and invasiveness than HA-MRSA strains [34], as well as worse clinical outcomes [35], [36]. This higher virulence and invasiveness has been associated with an increased expression of several cytolytic toxins (such as PVL). However, the exact mechanisms underlying the higher virulence of CA-MRSA are still uncertain [34]. An increase in virulence is also not universal, as other studies have described better clinical outcomes associated with CA-MRSA infections [37], [38]. This may be because CA-MRSA infections are largely associated with skin and soft-tissue infections, which generally have favorable outcomes [38], [39]. In addition, CA-MRSA strains have a narrower resistance spectrum which makes it easier to provide effective treatment. Overall, while the empirical evidence is mixed, there does seem to be some indication that CA-MRSA differs from HA-MRSA with regards to virulence, the range of resistance, and transmissibility (see Table 1 in [18], and [34]). Accordingly, a replacement of HA-MRSA with CA-MRSA in hospitals would entail a change in these important properties of nosocomial MRSA infections.
More fundamentally, the transmission route of MRSA in hospitals might change. The current view is that MRSA in hospitals is mainly transmitted indirectly through short-term contaminated health-care workers [40]. This dynamic could change should the more invasive and more transmissible CA-MRSA replace HA-MRSA in hospitals. Accordingly, prevention efforts that focus currently on hand-hygiene among health-care workers could lose their effectiveness in reducing the spread of MRSA. Interestingly, our results suggest that a replacement of HA-MRSA by CA-MRSA is especially likely in those locations in which infection control in hospital is currently successful and hence transmission rates in the hospital are low. However, our results also indicate that due to the large heterogeneities characterizing human populations, coexistence between hospital- and community-associated strains of MRSA is overall a likely outcome.
|
10.1371/journal.pmed.1002134 | A Comparison of Midwife-Led and Medical-Led Models of Care and Their Relationship to Adverse Fetal and Neonatal Outcomes: A Retrospective Cohort Study in New Zealand | Internationally, a typical model of maternity care is a medically led system with varying levels of midwifery input. New Zealand has a midwife-led model of care, and there are movements in other countries to adopt such a system. There is a paucity of systemic evaluation that formally investigates safety-related outcomes in relationship to midwife-led care within an entire maternity service. The main objective of this study was to compare major adverse perinatal outcomes between midwife-led and medical-led maternity care in New Zealand.
This was a population-based retrospective cohort study. Participants were mother/baby pairs for all 244,047 singleton, term deliveries occurring between 1 January 2008 and 31 December 2012 in New Zealand in which no major fetal, neonatal, chromosomal or metabolic abnormality was identified and the mother was first registered with a midwife, obstetrician, or general practitioner as lead maternity carer. Main outcome measures were low Apgar score at five min, intrauterine hypoxia, birth-related asphyxia, neonatal encephalopathy, small for gestational age (as a negative control), and mortality outcomes (perinatal related mortality, stillbirth, and neonatal mortality). Logistic regression models were fitted, with crude and adjusted odds ratios (ORs) generated for each outcome for midwife-led versus medical-led care (based on lead maternity carer at first registration) with 95% confidence intervals. Fully adjusted models included age, ethnicity, deprivation, trimester of registration, parity, smoking, body mass index (BMI), and pre-existing diabetes and/or hypertension in the model. Of the 244,047 pregnancies included in the study, 223,385 (91.5%) were first registered with a midwife lead maternity carer, and 20,662 (8.5%) with a medical lead maternity carer. Adjusted ORs showed that medical-led births were associated with lower odds of an Apgar score of less than seven at 5 min (OR 0.52; 95% confidence interval 0.43–0.64), intrauterine hypoxia (OR 0.79; 0.62–1.02), birth-related asphyxia (OR 0.45; 0.32–0.62), and neonatal encephalopathy (OR 0.61; 0.38–0.97). No association was found between lead carer at first registration and being small for gestational age (SGA), which was included as a negative control (OR 1.00; 0.95–1.05). It was not possible to definitively determine whether one model of care was associated with fewer infant deaths, with ORs for the medical-led model compared with the midwife-led model being 0.80 (0.54–1.19) for perinatal related mortality, 0.86 (0.55–1.34) for stillbirth, and 0.62 (0.25–1.53) for neonatal mortality. Major limitations were related to the use of routine data in which some variables lacked detail; for example, we were unable to differentiate the midwife-led group into those who had received medical input during pregnancy and those who had not.
There is an unexplained excess of adverse events in midwife-led deliveries in New Zealand where midwives practice autonomously. The findings are of concern and demonstrate a need for further research that specifically investigates the reasons for the apparent excess of adverse outcomes in mothers with midwife-led care. These findings should be interpreted in the context of New Zealand’s internationally comparable birth outcomes and in the context of research that supports the many benefits of midwife-led care, such as greater patient satisfaction and lower intervention rates.
| New Zealand adopted an autonomous midwife-led model of maternity care in 1990. There has never been a detailed review examining what effect, if any, midwife-led care has on adverse outcomes for unborn and newly born infants in the New Zealand setting.
A review of the safety of New Zealand’s midwife-led maternity system is of relevance to other countries that are considering adopting this model of care.
We examined data on all full-term births in which no serious abnormalities were detected in the baby that occurred in New Zealand over a 5-y period (total sample size 244,047).
We compared rates of adverse outcomes for unborn and newly born infants among women who were in midwife-led versus medical-led care at first registration during antenatal care.
Overall rates of adverse outcomes in the New Zealand setting were low and comparable to international rates.
We found that, among mothers with medical-led care compared with midwife-led care, there were lower odds of some adverse outcomes for infants. These included oxygen deprivation during the delivery (birth-related asphyxia) (55% lower odds), neonatal encephalopathy—a condition that can result in brain injury (39% lower odds), and low Apgar score, which is a measure of infant well-being immediately postdelivery, with a low score being indicative of an unwell baby (48% lower odds).
Despite New Zealand having overall internationally comparable maternity outcomes, the findings of this study suggest that avoidable adverse outcomes may still be occurring.
Further research that examines the potential reasons for an apparent excess in adverse outcomes in midwife-led care is required.
| The organization of maternity systems varies internationally. Most systems involve a level of collaboration between medical and midwifery practitioners. Many countries have medical-led maternity care systems in which midwives provide a significant amount of care but are not autonomous in their practice [1]. However, in two countries, the Netherlands and New Zealand, midwife-led continuity of care is the typical model and has been defined as one in which “the midwife is the lead professional in the planning, organization and delivery of care given to a woman from initial booking to the postnatal period” [2]. This definition is consistent with how midwife-led care operates in New Zealand. There are movements in other countries to adopt such a system [3–5].
New Zealand’s maternity system experienced major reform in 1990 when legislation resulted in substantial changes to the way maternity care was funded, allowing for complete midwifery autonomy and removal of the requirement for a nursing background for midwives (i.e., direct-entry education) [6]. As a result, currently four out of five mothers have a midwife as their lead maternity carer [7]. For these women, a doctor is only generally involved if the mother has or develops obstetric risk factors. In New Zealand, each care provider type is paid (by the New Zealand government) equally based on the services provided; only medical-led carers are able to charge fees on top of the government subsidy. Thus, midwifery care is provided at no cost to the patient. Obstetric-led care is substantially less frequent than midwife-led care (5.5% versus 81.1% in 2012) [7]. Obstetricians tend to have two distinct subgroups of patients: those identified as high risk, for whom obstetric care is provided at no cost (by hospital-employed practitioners), and those who choose private fee-for-service obstetric care, which while subsidized, still comes at a substantial cost to the patient. General practitioner (GP) obstetricians, who prior to 1990 were the most common lead maternity carers, are now increasingly rare (1.1% in 2012) [7]. The lead maternity carer is responsible for overseeing and carrying out all aspects of maternity care. If the care required falls outside of the care provider’s scope of practice, it is their responsibility to refer the woman to an appropriate practitioner; however, the original care provider typically remains as the lead maternity carer in a shared care arrangement. Thus, women with a midwife as their lead maternity carer may have any level of medical input into their care, and those with an obstetrician are also likely to receive midwifery support. The changes to the maternity system in New Zealand occurred rapidly, and there has been little in the way of systematic evaluation that specifically investigates safety-related outcomes within the new system.
A 2016 Cochrane systematic review on 17,642 births across four countries reported numerous benefits associated with midwife-led models of care compared to other (standard) models of care [2]. Benefits included less fetal and neonatal loss, fewer preterm births (before 37-wk gestation), fewer interventions such as regional anesthesia and instrumental delivery, a higher chance of being attended at birth by a known midwife, and a higher chance of vaginal delivery. The Cochrane group reported no differences in rates of adverse outcomes that included neonatal convulsions, a 5 min Apgar score of less than or equal to seven, and neonatal admission to an intensive care unit [2]. However, the midwife-led interventions assessed in this review were heterogeneous, in many cases comparing highly coordinated care with routine care. Further, the midwife-led interventions frequently involved routine medical input. It was also noteworthy that in the majority of studies, the midwife-led interventions were carried out by a small number of midwives (in most cases ten or fewer). It is thus difficult to generalize the findings of this review to a population-based setting with completely autonomous midwives.
This study aimed to investigate whether the frequency of adverse perinatal outcomes differed between midwife-led and medical-led births within the New Zealand setting. The specific outcomes we investigated included mortality outcomes (perinatal related mortality, which includes stillbirth and neonatal mortality); morbidity outcomes, particularly those associated with perinatal care (Apgar score of less than seven at 5 min, intrauterine hypoxia, birth related asphyxia, and neonatal encephalopathy); and small for gestational age (SGA) as a negative control outcome.
This was a population-based retrospective cohort study in which mother/baby pairs were identified from routinely collected maternity (live-born births) and mortality (stillborn births) data and followed up for mortality and morbidity outcomes using mortality and hospitalization data.
The study was granted ethical approval by New Zealand’s Multi-Region Ethics Committee (MEC/11/EXP/131). All data were nonidentifiable, and individual informed consent from participants was not required.
This study is reported as per Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (S1 Checklist).
Maternity data are routinely collected in New Zealand from maternity claim forms submitted by lead maternity carers. Data include information on maternal characteristics (including body mass index [BMI] and smoking status at registration), obstetric characteristics (including parity, gestation at delivery, and mode of delivery), service provision (including lead maternity care type at first registration and trimester of registration), and state of the neonate at birth (including Apgar scores) [8]. These data are routinely linked with data from the National Minimum Dataset (NMDS), which records hospital discharge data. The NMDS includes demographic data and diagnostic and procedural coded data based on the International Statistical Classification of Diseases and Related Health Problems for both mother and baby [9]. These data include details of maternal conditions such as pre-existing hypertension and diabetes, fetal/neonatal abnormalities, and peri- and postnatal diagnoses including intrauterine hypoxia, birth-related asphyxia, and neonatal encephalopathy. The mortality dataset holds data on all deaths, including stillbirths, occurring in New Zealand [10].
Participants were included in the study if the delivery occurred within the study period of 1January 2008 to 31 December 2012, the longest time period in which all data were available. Participants met the following inclusion criteria: a singleton pregnancy, gestation equal to or greater than 37 wk, and no identification of major fetal or neonatal congenital, chromosomal, or metabolic abnormalities. These criteria were designed to exclude pregnancies with high rates of unavoidable adverse outcomes not related to care. Because our aim was to compare midwife-led and medical-led births, only women registered with either a midwife, obstetrician, or GP as their lead maternity carer were included. Those with no or an unknown lead maternity carer, a hospital as lead maternity carer (whereby the mother receives care from a team of maternity care providers working within a hospital setting), or those who registered with a lead maternity carer after the delivery were excluded.
Fetal and neonatal abnormalities were categorized based on identification of specific ICD-10-AM diagnostic codes that are included in the NMDS. The list of ICD-10-AM codes was defined by a maternal fetal medicine specialist without reference to study data (thus, exclusions were made blinded to exposure and outcomes). Codes were considered relevant if the condition increased the risk of fetal or neonatal mortality (such as anencephaly) or when the condition was considered serious in that tertiary-level neonatal admission would likely be required (such as a heart defect).
Exposure groups were based on the participant’s lead maternity carer at first registration. Those with a midwife as their registered lead maternity carer were allocated to the midwife-led group, and those with either an obstetrician or GP as lead maternity carer were allocated to the medical-led group.
Three types of variables were assessed: (1) mortality outcomes, (2) morbidity outcomes, and (3) SGA. These were defined as follows.
Variables were chosen based on a priori knowledge of their plausible associations with the exposures and outcomes measured in this study. A simplified directed acyclic graph is provided (S1 Fig).
Age was calculated on the basis of age at the time of delivery. BMI was recorded on the lead maternity carer claim form, with height and weight as measured by the lead maternity carer or reported by the mother. For both these variables, there is a potential U-shaped association with adverse outcomes in which both very young or underweight and older or overweight mothers may have higher risk, so these variables were treated in two ways in multivariable models: as categorical variables (age categories <20 y, 20–35 y, and >35 y, BMI categories <19, 19–24, and ≥25) and as continuous variables. In both cases, the results of the study were almost identical. In descriptive analyses, these variables were treated as categorized variables, whilst in multivariable models, both BMI and age were treated as continuous covariates.
Deprivation was based on the mother’s place of residence using the New Zealand Deprivation Index (NZDep 2006), a small-area-based index calculated using aggregated census data based on residents’ socioeconomic characteristics (such as benefit receipt, earning under an income threshold, and access to car or phone) [14]. Scores were grouped into deciles and treated as continuous in models.
Ethnicity was self-nominated by mothers and derived from the maternity dataset. When more than one ethnic group was nominated, women were assigned according to standard prioritization procedures in the following order: Māori, Pacific, Asian (Indian), Other, and European [15]. Indian would normally be included in the Asian group, but because of their higher risk of adverse outcomes, we treated them as a separate ethnic group.
Smoking status (yes/no), parity, and trimester at registration were recorded on the LMC claim form at the time of registration.
The presence of pre-existing diabetes and/or hypertension were based on ICD-10-AM diagnostic codes recorded in the NMDS. These codes included all ICD-10-AM codes that identify the condition pre-existing to pregnancy—e.g., code “O240” for pre-existing diabetes mellitus, Type 1, in pregnancy and code “O100” for pre-existing essential hypertension complicating pregnancy, childbirth, and the puerperium.
Data on gestation at delivery and delivery type were also collected, but these were not included in models because they are considered potential mediators between the model of care and perinatal outcomes (S1 Fig).
In all cases, those with missing data (missing data were <1% for each covariate) were excluded from the multivariable analyses. This led to a complete case analysis that excluded 1.3% of pregnancies.
Prevalence estimates of maternal and obstetric variables for the overall cohort and for each LMC exposure group were calculated and compared. For the exposure groups, crude and age-standardized prevalence estimates were calculated with prevalence standardized to the age structure of the total study population. The Cochrane-Mantel-Haenzel statistic (p-value) was used to compare the distribution of variables between exposure groups, adjusted for age.
Rates of PRM and stillbirth were calculated for the total population and for each exposure group, with the total number of births in the relevant population as the denominator. For all other outcomes, pregnancies in which a stillbirth occurred were excluded from the denominator. For low Apgar score and being SGA, those with missing outcome data were excluded (both >90% complete). Logistic regression models were fitted to assess the crude associations of mother’s age (<20, 20–35, and >35 y), deprivation (deciles 1–2; 3–4; 5–6; 7–8, and 9–10), mother’s ethnicity (European, Māori, Pacific, Asian, Indian, and Other), BMI (<19;19–24; 25+), smoking status at first LMC registration (yes/no), parity (nulliparous versus other), trimester of LMC registration (trimester one versus other), and the presence of pre-existing diabetes and/or hypertension (yes/no) with each outcome. Adjusted odds ratios (ORs) were calculated for each variable by fitting further logistic models for each outcome, with each model containing all of the other confounding variables listed above. Hypothesis tests for each factor are Type III tests from the adjusted logistic regression model, testing for overall association of each given factor with the outcome (e.g., any differences in perinatal mortality outcome by age group.)
Finally, logistic regression models were fitted to calculate the odds of each outcome relative to the model of maternity care at first registration, with midwife-led care as the reference group. Crude ORs for the model of maternity care were calculated, followed by adjusted ORs in which models included age (continuous), ethnicity (European, Māori, Pacific, Asian, Indian, and other), deprivation (continuous deciles), BMI (continuous), smoking at registration (yes/no), parity (nulliparous and other), trimester of registration (trimester one and other), and pre-existing conditions (yes/no). Reported p-values are Type III hypothesis tests from the adjusted logistic regression models (as noted above). To account for the potential impact of clustering by region with associated potential differences in services available in different hospitals, we conducted supplementary analysis examining the above birth outcomes by model of maternity care. These models adjusted for the potential confounders as specified above and included District Health Board of residence as a random intercept term (using a generalized linear mixed model).
All data analysis was performed using SAS statistical software package version 9.3.
We were not able to further stratify the midwife-led group into those who had received medical input during pregnancy from those who had not because of data limitations. We were also not able differentiate stillbirths into antepartum and intrapartum deaths, which may have been helpful in that intrapartum deaths are more likely to be related to model of care. All analytical and other data management steps relating to inclusions and exposure categorizations were done prior to linking the data on model of care and covariates to the outcome data.
There were 286,572 singleton, term pregnancies in New Zealand in the period of the study. Of these, 3,766 were excluded because the baby was identified as having a serious congenital anomaly, and 37,691 were excluded because their LMC was not a midwife, GP, or obstetrician. Of these births without a midwife or medical LMC, 99% were explicitly noted as not having an LMC (data field entered as “No LMC,” so the mother likely had no antenatal care and presented in labor), with the remainder having a hospital or District Health Board as their registered LMC. Only 11 mothers had missing LMC data. An additional 1,068 were excluded because they registered with their LMC after delivery of the baby, in total leaving 244,047 pregnancies meeting the inclusion criteria. Of these women, 223,385 (91.5%) were first registered with a midwife as their LMC, and 20,662 (8.5%) with a medical LMC.
Table 1 shows the characteristics of mothers overall and for mothers with midwife-led and medical-led care. Compared with the medical-led group, the midwife-led group had greater proportions of younger mothers, mothers of a non-European ethnicity, and mothers from higher deprivation areas. They were also more often overweight, registered with their carer later in pregnancy, and were more often smokers. Compared with the midwife-led group, the medical-led group were more likely to be older, nulliparous, and to deliver their babies at earlier gestations. Rates of pre-existing diabetes and/or hypertension were similar between groups (0.7% for midwife-led and 0.9% for medical-led) (Table 1). Missing data were relatively rare for any given variable (<1%).
Table 2 presents the number and rates of adverse outcomes in total and by exposure group.
In the logistic regression model for perinatal mortality, after adjustment, the odds of death were higher for mothers who were smokers at registration, had a BMI ≥25, registered at a later trimester, or were nulliparous (Table 3). Adjusted odds of each of the morbidity-related outcomes were higher for mothers who were smokers at registration or were nulliparous, with smoking being particularly strongly related to SGA (OR 2.56; 95% CI: 2.46–2.66), and nulliparity with birth asphyxia/hypoxia/encephalopathy (OR 2.44; 2.22–2.67). Late trimester at registration was not independently associated with the two morbidity outcomes (included in these analysis) but was weakly associated with SGA (OR 1.11; 1.08–1.14). High BMI, by contrast, was associated with higher odds of morbidity outcomes (adjusted ORs were 1.27; 1.16–1.39 for birth asphyxia/hypoxia/encephalopathy and 1.29; 1.18–1.41 for low Apgar) but lower adjusted odds of SGA (OR 0.72; 0.70–0.74) (Tables 3 and 4).
Compared with mothers aged 20–35 y, age under 20 y was not associated with the morbidity-related outcomes but was associated with SGA. Compared with mothers aged 20–35 y, age greater than 35 y was associated with higher adjusted odds of the morbidity outcomes (ORs were 1.15; 1.01–1.30 for birth asphyxia/hypoxia/encephalopathy and 1.12; 0.99–1.26 for low Apgar) and SGA (OR 1.12; 1.07–1.16). Compared with other mothers, Māori and Pacific mothers were at increased crude odds of PRM; however, this association was not apparent after adjusting for the other covariates. Pacific and Asian mothers had lower adjusted odds of birth asphyxia/hypoxia/neonatal encephalopathy. Māori, Pacific, Asian, Indian, and Other mothers all had higher odds of SGA compared with other mothers. High deprivation (NZDep 9–10) was associated with higher adjusted odds of the morbidity-related outcomes and SGA, with ORs ranging from 1.19; (95% CI 1.03–1.38) to 1.60 (1.36–1.88). Mothers with pre-existing conditions were more likely to have SGA babies (OR 1.36; 1.16–1.60) (Tables 3 and 4).
Compared with midwife-led births, after adjusting for age, ethnicity, deprivation, BMI, smoking, parity, trimester of registration, and pre-existing conditions, there were nonstatistically significant lower odds of stillbirth (OR 0.86; 0.55–1.34), neonatal mortality (OR 0.62; 0.25–1.53), and PRM (OR 0.80; 0.54–1.19) among medical-led births—in all cases, confidence intervals were wide, and it was not possible to definitively say whether one model of care was associated with fewer deaths than the other. For medical-led births compared with midwife-led births, after adjusting for covariates, there were 48% lower odds of an Apgar score of less than seven at 5 min (OR 0.52; 0.43–0.64), 55% lower odds of birth-related asphyxia (OR 0.45; 0.32–0.62), 39% lower odds of neonatal encephalopathy (OR 0.61; 0.38–0.97), and 21% lower odds of intrauterine hypoxia (OR 0.79; 0.62–1.02) (Table 5). There was no difference in adjusted odds of being SGA between groups (OR 1.00; 0.95–1.05) (Table 5).
When we limited the analysis to low-risk mothers with a BMI of less than 25, nonsmokers at registration, and no pre-existing hypertension and/or diabetes, without exception the associations seen were slightly stronger than those for the whole cohort. For example, after adjusting for age, ethnicity, deprivation, parity, and trimester of registration, ORs for medical- compared with midwife-led births were as follows: for low Apgar score at 5 min, 0.48; 0.36–0.65, and for combined intrauterine hypoxia, birth-related asphyxia, and neonatal encephalopathy, 0.54; 0.40–0.73 (see S1 Table). Accounting for the potential impact of clustering of pregnancy outcomes by hospital (District Health Board [DHB]) resulted in failure to converge for mortality outcomes and results that were largely consistent with the main findings but with some movement towards the null for low Apgar (OR 0.60; 0.48–0.74) and combined intrauterine hypoxia, birth-related asphyxia, and neonatal encephalopathy (0.76; 0.62–0.94) (see S2 Table).
Results from our study showed that after adjusting for likely confounding, medically led (i.e., having a medical LMC at first registration) births were associated with substantially lower odds of an Apgar score of less than seven at 5 min, birth-related asphyxia, and neonatal encephalopathy when compared with midwife-led births. Medical-led births were also associated with somewhat lower odds of stillbirth and neonatal mortality and intrauterine hypoxia; however, confidence intervals in these instances were wide and included the null. There was no association between model of care and being SGA after adjustments were made. The associations between other risk factors and the outcomes assessed in this study were all largely as expected [11,16–20].
This is one of very few population-based studies based on individual data that have been able to compare rare adverse fetal and neonatal outcomes among births with autonomous midwife-led care compared with medical-led care. This study uses national level data and includes around 85% of all births occurring in New Zealand over the period of the study.
Given the different demographic and risk factor distribution between the study groups, the potential for confounding needs to be carefully considered. We are reassured that confounding was unlikely to account for the key findings for the following reasons. First, we used being SGA as a negative control to assess for residual and unmeasured confounding [13]. SGA was included as a negative control because it was likely to share a similar confounding structure as the other outcomes (see S1 Fig) but is unlikely to be influenced by model of care. Because of the independent and strong relationships between the potentially confounding covariates and this end point, we would expect to observe an association between model of care and SGA if that association was due solely to confounding by those covariates. In fact, once we adjusted for the covariates, there was no association between model of care and SGA (OR 1.00; 0.95–1.05), suggesting that either these covariates are not acting as confounders in an important way or that they have been adequately adjusted for.
Because BMI has a different relationship with SGA than the other adverse outcomes, we stratified the cohort by BMI (≥25 and <25) and found that the results in relation to model of care and adverse outcomes were similar for both BMI groups. One exception was PRM, in which ORs comparing medical-led to midwife-led births showed a greater protective effect for medical-led births in the normal BMI cohort; however, the estimates were imprecise on both counts, and the confidence intervals were substantially overlapping and included the null (S1 Table). We also reran the main analyses for lower-risk women (BMI less than 25, nonsmokers, and no pre-existing hypertension and/or diabetes) and found that in all cases the associations were stronger than for the full cohort (S1 Table). Finally, we estimated the magnitude of residual confounding that would be necessary to completely explain the associations demonstrated using a formula from Greenland [21]. This formula allows one to calculate the impact of confounding on an estimated association given the strength of association between the putative confounder and the dependent and independent variables. Assuming the prevalence of the unmeasured confounder(s) would be such to maximize the magnitude of confounding, to completely explain the findings relating to low Apgar score at 5 min or the combined intrauterine hypoxia/birth-related asphyxia/neonatal encephalopathy outcome, the ORs between the putative confounder(s) and the exposure or the outcome would have to be very strong (greater than five and greater than four, respectively). It is difficult to think of an unmeasured confounder or confounders that would meet these criteria.
The second potential limitation was the use of retrospective data that in some instances lacked detail and possibly accuracy and completeness. Our exposure groups were classified based on the registered LMC at first registration. We were unable to differentiate between midwives who provided all care themselves from those who had a level of medical input. Thus, our study assesses midwife-led care, not (necessarily) sole midwife care compared with medical-led care. There were a small number of women who had a different registered LMC at delivery than the one identified at registration, i.e., there were a small number of participants in the midwife-led group who were registered with a hospital or an obstetrician at time of delivery (n = 4,727, 1.9%). There was no means of ascertaining when the transfer of care occurred. To assess the impact of this, we removed all these participants and reran our logistic regression models; the results were unchanged.
It is possible that there was an under-reporting of some outcomes in the data. However, it is difficult to conceptualize any mechanism in which this would be differential in relation to the care received, so the effect of this would be to decrease study power rather than to introduce bias. The possible exception to this is Apgar score after 5 min, for which there were data missing in 8.3% of the midwife-led group and 11.8% of the medical-led group. If these missing data were more likely to be low Apgar scores, then the ORs would exaggerate the positive effect of medical-led births. However, it seems more likely (rather than less) that babies with low Apgar scores would have them recorded because of their clinical significance, so this seems an unlikely explanation for this specific finding.
While our ORs for mortality outcomes indicated a reduced risk for medical-led births, consistent with morbidity outcomes, confidence intervals were wide because of the rarity of this outcome. However, we used all available data, so these results still provide the best estimates given the data and sample size constraints that we had.
There have been very few individual studies published that compare midwife-led care with other models of maternity care that are sufficiently powered to look at major adverse events such as those included in this study. As mentioned earlier, a recent systematic review comparing midwife-led maternity care with other models of care concluded that women with midwife-led care were “less likely to experience intervention and more likely to be satisfied with their care with at least comparable adverse outcomes” [2]. However, for reasons outlined previously, the findings of this review cannot be generalized to a population context of completely autonomous midwives. Many of the analytical observational studies comparing midwife-led versus standard care models do not have sample sizes that are adequately powered for investigating rare, adverse outcomes [22–24]. One of the larger studies was performed by the Birthplace in England Collaborative Group, which compared outcomes of women with singleton, term pregnancies who had planned to receive care at home, in freestanding midwifery units, in alongside midwifery (joined to an obstetric unit) units, and in obstetric units [25]. They found that there was no difference in a composite end point of mortality and morbidity-related outcomes between groups, although there was an increase in risk of the composite mortality and morbidity outcome for women who planned a home birth compared with women planning on birthing in an obstetric unit (OR 1.59; 95% CI 1.01–2.52), which was higher still for low-risk nulliparous women (OR 2.80; 1.59–4.92). Similarly, a study from Stockholm that compared women birthing in a hospital birth center (with midwife-led care) with those birthing in hospitals under standard care found no overall difference in rates of perinatal mortality among those using the birthing center [26]. However, perinatal mortality rates among primiparous women were higher for those who delivered at a birth center compared with those who birthed in hospitals under standard care (OR 2.2; 1.3–3.9). A New Zealand-based analysis of outcomes of babies according to their planned place of birth showed that women who planned home births tended to be older, European, and multiparous and to have a BMI in a normal range; crude analysis showed that these women had lower rates of adverse outcomes compared with other women, but no adjustment for confounding characteristics was made [27].
Two studies have been conducted in the Netherlands that compare adverse perinatal outcomes between autonomous midwife- and medical-led care [28,29]. The first study concluded that babies of low-risk mothers under midwife-led care had increased risk of delivery-related death compared with babies of high-risk mothers under medical-led care (relative risk 2.33, 95% CI 1.12–4.83) [28]. The authors of this study were unable to adjust for known risk factors because of the use of aggregated data in which case information was collected without recording personal characteristics of the woman (to guarantee anonymity). A more recent study in the Netherlands (but one that used data from a similar period) found no difference in relative risk of intrapartum and neonatal mortality between births that started in midwife-led versus secondary obstetric-led care (relative risk 0.88, 95% CI 0.52–1.46) [29].
The results of the current study may raise questions about some aspects of the safety of the midwife-led model of care, at least in the New Zealand context. However, it is also very important that the findings are interpreted in context of research supporting the many positive aspects of midwife-led care, such as lower intervention rates and greater patient satisfaction [2]. It is also important to interpret the findings in the context of New Zealand’s overall birth outcomes in comparison to other countries. The 2011 perinatal mortality rate (occurring at >24 wk gestation) in New Zealand (6.7) was similar to that of the United Kingdom, lying between the rates in Ireland (6.1) and England and Wales (7.5). Using a slightly different definition (deaths occurring at >20 wk gestation), New Zealand had a largely similar perinatal mortality rate at 10.6 deaths per 1,000 births compared to Australia’s at 9.8 deaths per 1,000 births [17]. This is reassuring in that it suggests that in absolute terms New Zealand’s maternity system is still internationally comparable in terms of adverse outcomes; however, it does not preclude the possibility that avoidable adverse outcomes are occurring, potentially both in New Zealand and in other countries.
There is a need to understand the reasons for the apparent excess of adverse outcomes in midwife-led deliveries in New Zealand. Despite a radical change in the way maternity care was delivered, there has never been a full and proper evaluation to ensure the maternity system in New Zealand is safe. This should be a major priority both for New Zealand and other countries with current or prospective midwife-led systems. Given the positive findings of the Cochrane review relating to midwife-led care, it may well be that midwife-led care is optimal within the context of well-organized systems [2]. However, there is an urgent need to establish which aspects of those systems potentially make that care more, or less, safe. This might include an evaluation of different approaches to training maternity care professionals, the impact of the level of collaboration between midwives and doctors, the triaging process in allocating midwife-led versus medical-led care, if there are any differences in risk of adverse outcomes relating to rurality or access to secondary services in midwife-led births, and the level of organization around antenatal care.
|
10.1371/journal.ppat.1006626 | Binding of a C-type lectin’s coiled-coil domain to the Domeless receptor directly activates the JAK/STAT pathway in the shrimp immune response to bacterial infection | C-type lectins (CTLs) are characterized by the presence of a C-type carbohydrate recognition domain (CTLD) that by recognizing microbial glycans, is responsible for their roles as pattern recognition receptors in the immune response to bacterial infection. In addition to the CTLD, however, some CTLs display additional domains that can carry out effector functions, such as the collagenous domain of the mannose-binding lectin. While in vertebrates, the mechanisms involved in these effector functions have been characterized in considerable detail, in invertebrates they remain poorly understood. In this study, we identified in the kuruma shrimp (Marsupenaeus japonicus) a structurally novel CTL (MjCC-CL) that in addition to the canonical CTLD, contains a coiled-coil domain (CCD) responsible for the effector functions that are key to the shrimp’s antibacterial response mediated by antimicrobial peptides (AMPs). By the use of in vitro and in vivo experimental approaches we elucidated the mechanism by which the recognition of bacterial glycans by the CTLD of MjCC-CL leads to activation of the JAK/STAT pathway via interaction of the CCD with the surface receptor Domeless, and upregulation of AMP expression. Thus, our study of the shrimp MjCC-CL revealed a striking functional difference with vertebrates, in which the JAK/STAT pathway is indirectly activated by cell death and stress signals through cytokines or growth factors. Instead, by cross-linking microbial pathogens with the cell surface receptor Domeless, a lectin directly activates the JAK/STAT pathway, which plays a central role in the shrimp antibacterial immune responses by upregulating expression of selected AMPs.
| The JAK/STAT pathway mediates the effects of a large number of cytokines and growth factors. It is activated following binding of a cytokine or growth factor to its respective receptor. To date, over 50 cytokines and growth factors have been shown to utilize the pathway to regulate cell growth, survival differentiation, motility, and immune responses. The JAK/STAT pathway is ubiquitous in vertebrates but can also be found as an intact pathway in some invertebrates, including shrimp. However, few cytokines and growth factors like molecules are identified in invertebrates and the function of the pathway in invertebrates is seldom studied. In this study, we identified core components of JAK/STAT pathway in shrimp and found the pathway had an important function in antibacterial immunity. Bacterial pathogens directly activate the JAK/STAT pathway through a secreted C-type lectin containing a coiled coil domain and a C-type lectin domain (MjCC-CL) in shrimp. Working as a cytokine like ligand, the MjCC-CL binds to polysaccharides from bacteria and the ILR domain of Domeless, induces STAT phosphorylation and translocation into the nucleus, and expression of several AMPs. The MjCC-CL, both as the pattern recognition receptor of bacteria and the ligand of Dome mediates activation JAK/STAT pathway.
| Like other invertebrates, the shrimp’s defense against microbial pathogens relies on innate immune responses, which generally encompass their recognition, killing and disposal through humoral and cellular mechanisms. Humoral responses are rapid and effective, and include recognition factors such as lectins, as well as effector mechanisms, including hemolymph clotting, production of antimicrobial peptides (AMPs) and oxygen reactive intermediates, and, melanization of the microbe [1]. The cellular immune responses include pathogen recognition, killing and clearance by phagocytosis, immobilization by hemocyte extracellular traps, or nodulation or encapsulation of larger microorganisms [2, 3].
Initiation of the innate immune response is triggered by pathogen sensors called pattern recognition receptors (PRRs) such as Toll-like receptors and lectins. More than 10 different types of PRRs are found in shrimp [4]. Among these PRRs, C-type lectins (CTLs) are important in the recognition of carbohydrate moieties (microbe-associated molecular patterns, MAMPs) displayed on microbial surfaces. CTLs are a structurally diverse family of carbohydrate-binding proteins of wide taxonomic distribution in both vertebrates and invertebrates, characterized by calcium-dependent ligand recognition through a C-type carbohydrate recognition domain (CTLD) that displays a unique sequence motif and structural fold. By recognizing microbial glycans, the CTLD is responsible for the CTLs’ roles as pattern recognition receptors (PRRs) in the innate immune response to bacterial infection of both vertebrates and invertebrates. In addition to the CTLD, however, most CTLs display additional domains that can carry out effector functions. In CTLs from vertebrates some of these effector functions have been well established and the mechanisms involved have been described in detail. One of the most thoroughly characterized example, the mannose-binding lectin (MBL) a prototypical CTL, displays a collagenous domain that by associating with an MBL-associated serine protease (MASP) can activate the complement cascade [5, 6]. In contrast, the effector functions of CTLs from invertebrates remain poorly understood. During the past few years, we have focused on the roles of CTLs in the shrimp immune response to bacterial infection [7–9]. Based on their domain organization, at present time the shrimp CTLs can be classified in three distinct groups: (1) CTLs that contain only one CTLD, (2) CTLs with two CTLDs, and (3) CTLs with one CTLD and an additional domain [7] such as a low-density lipoprotein receptor class A domain [10], and an immunoglobulin-like domain [11]. The shrimp CTLDs can recognize viral or bacterial glycans, and thus immobilize the invading microorganisms, CTLs can also induce downstream immune responses aimed at their killing, destruction, and elimination. For some shrimp CTLs, their effector innate immune functions such as phagocytosis, prophenoloxidase activation, and promotion of respiratory burst have been described [7, 8], although the mechanisms involved still remain unclear. We recently identified in the kuruma shrimp Marsupenaeus japonicus a novel CTLs (MjHeCL) that in addition to recognizing microbial glycans via the CTLD, displays unique functional properties, as the inhibition of proliferation of the hemolymph microbiota by maintaining the homeostatic expression of antimicrobial peptides [7]. The mechanistic aspects of this regulation of AMP expression in shrimp by CTLs, however, remained to be elucidated.
In this study, we identified in M. japonicus a novel CTL that we designated MjCC-CL, and that like MjHeCL, can regulate the expression of selected AMPs. MjCC-CL has a chimeric structure comprising the canonical CTLD that can recognize bacterial glycans and a coiled-coil domain (CCD). Our study revealed that the mechanism by which upon bacterial challenge, MjCC-CL regulates AMP expression is based on signaling by the JAK/STAT (Janus Kinase/Signal Transducer and Activator of Transcription) that leads to the transcription of five AMPs. The JAK/STAT pathway, which is common to both invertebrates and vertebrates, controls multiple biological processes, including cell development, growth and survival, tissue homeostasis and immune responses [12, 13]. In vertebrates, the JAK/STAT pathway is indirectly activated by cytokines or growth factors released upon stress signals and cell death. In striking contrast, the present study revealed that activation of the JAK/STAT pathway and AMP upregulation in shrimp takes place by direct MjCC-CL-mediated recognition of the microbial pathogen glycans by the CTLD, and binding of the CCD to the hemocyte surface Domeless, the receptor that activates JAK/STAT pathway.
From our transcriptome analysis of the kuruma shrimp (M. japonicus) we identified a unique CTL which we named MjCC-CL (GenBank accession no. KU213612), and comprises a signal peptide, a coiled-coil domain (CCD), and a CTLD (Fig 1A). A phylogenetic analysis of MjCC-CL and other CTLs from invertebrate and vertebrate species showed that in contrast with MjHeCL, a shrimp CTL that we recently characterized [7], MjCC-CL clusters with CTLs from vertebrates (Fig 1B). Further SMART analysis indicated that the N-terminal region (including the CCD) of MjCC-CL is similar to the interleukin (IL) 10 domain from mammals at the threshold level. Multiple alignment of the N-terminal region of MjCC-CL with Hs-IL10 (Homo sapiens interleukin 10, CAG46790.1) and Ms-IL10 (Mus musculus interleukin 10, EDL39722.1) showed that they share low similarity (Fig 1C). A comparison of homology models of the CCD of MjCC-CL and IL10 revealed an α-helix-rich structure for the CCD (Fig 1C1), with overall similarity to IL10 (Fig 1C2). The tissue distribution of MjCC-CL transcripts and protein in the control and challenged animals were analyzed by qRT-PCR and western blotting, respectively. The results in control shrimp showed that MjCC-CL was expressed in all tissues tested, with a relatively high expression in the intestine (Fig 1D). In the challenged shrimp, a time-course expression analysis showed that MjCC-CL was increased significantly in hemocytes at 3 and 6 h after V. anguillarum injection (Fig 1E), and with no obvious changes in the relatively high level of expression in intestine (Fig 1F). To analyze in vivo the potential immune function of MjCC-CL, we knocked down MjCC-CL expression by RNAi (MjCC-CL-RNAi shrimp) (Fig 1G) and assessed the bacterial clearance and survival rate of shrimp relative to the control animals with normal MjCC-CL expression. The results showed that in the experimentally challenged MjCC-CL-RNAi shrimp the number of bacteria increased significantly (Fig 1H), while the survival rate of the shrimp decreased significantly (Fig 1I) as compared to the controls. These results strongly suggested that MjCC-CL participates in the antibacterial response in shrimp.
To gain further insight into the role(s) of MjCC-CT in antibacterial immunity, we analyzed the potential binding to both Gram-positive (GP) and Gram-negative (GN) bacteria of: (a) the intact recombinant MjCC-CL (S1B Fig); (b) the separate recombinant CC and CTL domains (S1C and S1D Fig); and (c) the authentic MjCC-CL purified from shrimp intestine. The results showed that the intact rMjCC-CL and the rCTLD bound to the GP bacteria S. aureus and B. subtilis, and to the GN bacteria V. anguillarum and E. coli, whereas the rCCD showed no binding activity to any of the bacterial species tested (Fig 2A). Like the rMjCC-CL, the native MjCC-CL purified from shrimp intestine (Fig 2B) also bound to four types of bacteria (Fig 2C). Direct binding analysis with enzyme-linked immunosorbent assay (ELISA) revealed that the rMjCC-CL and CTLD bound to lipopolysaccharide (LPS) from E. coli and peptidoglycan (PGN) from S. aureus or B. subtilis, whereas no binding activity was detected for the rCCD (Fig 2D–2F and S1 Table). We then examined the potential binding of MjCC-CL to the shrimp hemocyte surface by immunocytochemical analysis using an anti-GST antibody. The results showed that rMjCC-CL or rCC could bind to the hemocyte surface (Fig 2G). To detect if the LPS/rMjCC-CL complexes could bind to the hemocyte surface, rMjCC-CL that had been pre-incubated with LPS was injected into shrimp, and the hemocytes were collected for immunocytochemical analysis. The control animals received LPS-incubated GST instead of LPS/rMjCC-CL complexes. The results revealed that the injected LPS/rMjCC-CL complexes localized on the hemocyte external surface, whereas in the GST control animals the label localized to the hemocyte cytoplasm (Fig 2G and 2H). Taken together, the results indicate that MjCC-CL not only binds to both GN and GP bacteria by recognizing their cell surface polysaccharides, but also the “self” ligands on the shrimp hemocyte surface.
As MjCC-CL binds to the bacterial surface, we next examined the possibility that this shrimp lectin could have direct bacteriostatic or bacteriocidal activity. Exposure of bacterial cultures to intact rMjCC-CL at increasing concentrations ranging from 10 μg/ml to 100 μg/ml showed no differences in bacterial growth with control cultures that received no MjCC-CL (Fig 3A). In light of these results we investigated if the antibacterial activity of MjCC-CL could be indirect, such as via the upregulation of AMP expression. For this, we first evaluated the expression of six AMPs, including 4 antilipopolysaccharide factors (ALFs) (GenBank accession nos. ALF-A1 KU213607, ALF-C1 KU213608, ALF-C2 KU160498, and ALF-D1 KU160499) and 2 Crustins (Crus) (CruΙ-1 KU160502, and CruΙ-5 KU213606), in shrimp intestine after V. anguillarum or LPS challenge; the results showed that the expression of all six AMPs was significantly upregulated (Fig 3B). Then, we performed RNA interference (RNAi) of MjCC-CL and the expression of these six AMPs upon LPS challenge was analyzed. After transfection with RNAi targeting MjCC-CL in shrimp (Fig 3C) following challenge with LPS, no upregulation of expression of ALF-A1, ALF-C1, ALF-C2, CruΙ-1 and CruΙ-5 was observed. In contrast, expression of ALF-D1 was equally upregulated in the RNAi transfected animals and in the control animals in which MjCC-CL was normally expressed (Fig 3D). To confirm the above results, the RNAi-transfected shrimp were first injected with rMjCC-CL protein, and subsequently challenged with LPS as above, and the expression of six AMPs was analyzed. The results showed that ALF-A1, ALF-C1, ALF-C2, CruΙ-1 and CruΙ-5 expression was increased significantly by LPS challenge in the rMjCC-CL-rescued shrimp, as compared to the control animals (Fig 3E). To confirm the specificity of the rMjCC-CL-mediated rescue we also injected a group of RNAi-transfected shrimp with recombinant CTL2, another shrimp CTL, and carried out the LPS challenge as above. The results showed that AMP expression was not induced significantly after rCTL2 injection, as compared to the control (Fig 3F). All the above results indicated that MjCC-CL specifically upregulates the expression of five different AMPs.
Next, we investigated the signaling pathway through which MjCC-CL regulates AMP expression. In Drosophila, AMP expression is mainly regulated by the Toll and IMD pathways [14]. Additionally, the cytokine-activated JAK/STAT pathway is key for antiviral responses in both Drosophila and mammals [13, 15–18]. Since in mammals the JAK/STAT signaling pathway is activated by different cytokines, our observation that MjCC-CL contains an IL10-like domain led us to hypothesize that its function might be also related to this signaling pathway. Therefore, we examined the potential role(s) of MjCC-CL in activation of the three signaling pathways, Toll, IMD, and JAK/STAT by assessing the translocation into the hemocyte nucleus of their transcription factors Dorsal, Relish and STAT (GenBank accession no. KU213611), respectively, upon increasing circulating MjCC-CL levels, and using LPS challenge as a positive control. The results of immunocytochemical analysis of hemocytes from experimental and control shrimp showed that while LPS challenge induced Dorsal or Relish translocation (Fig 4A and 4B) no translocation of Dorsal (Fig 4A, a) or Relish (Fig 4B, b) was detected in rMjCC-CL-injected shrimp. In contrast, STAT did translocate into the hemocyte nucleus in both rMjCC-CL-injected and LPS-injected shrimp (Fig 4C, c). Western blot analysis of proteins extracted from cytoplasm and nucleus of intestinal cells yielded results similar to those from the immunocytochemical study (Fig 4D and 4E). To further confirm the results, LPS was first incubated with rMjCC-CL and upon which any remaining free LPS was washed off. The LPS-rMjCC-CL mixture was injected into shrimp and translocation of Dorsal, Relish and STAT into the hemocyte nucleus was examined. The results showed that the LPS-rMjCC-CL complex could induce STAT translocation into the hemocyte nucleus (Fig 4F-f) but not translocation of Dorsal (Fig 4G) or Relish (Fig 4H-h). To confirm the role of MjCC-CL in activation of the JAK/STAT pathway, we analyzed STAT phosphorylation in MjCC-CL-RNAi shrimp. The results showed that in the MjCC-CL knockdown shrimp (Fig 5A) the LPS challenge reduced STAT phosphorylation (Fig 5B) and inhibited STAT translocation into the nucleus (Fig 5C and c). When rMjCC-CL protein was injected into the MjCC-CL knockdown shrimp, the LPS challenge induced STAT phosphorylation (Fig 5D) and translocation into the nucleus (Fig 5E and e) as in the control shrimp. However, if in the rescue experiments rMjCC-CL was replaced by rCTL2, STAT phosphorylation (Fig 5F) and translocation did not markedly change (Fig 5G, g). Taken together, our results suggest that MjCC-CL regulates AMP expression via the JAK/STAT pathway, without involving the Toll or IMD pathways.
We subsequently investigated whether the JAK/STAT pathway regulates AMP expression in shrimp. The shrimp STAT contains an N-terminal domain (NTD), a coiled-coil domain (CC), a DNA-binding domain (DB), a linker domain (LD), an SH2 domain (SH2) and a transactivation domain (TAD) (S2A Fig), and in a phylogenetic analysis, it clusters with other invertebrate STATs (S2B Fig). Analysis of the spatiotemporal distribution of shrimp STAT revealed that it was expressed in all tested tissues, but had relatively low expression in the hemocytes and stomach (Fig 6A). Expression of STAT was upregulated in hemocytes and the intestine at 3 h after bacterial challenge and gradually returned to normal levels from 6 to 24 h (Fig 6B and 6C). To investigate whether expression of ALF-A1, ALF-C1, ALF-C2, CruΙ-1 and CruΙ-5 is regulated by the JAK/STAT pathway, we knocked down STAT expression by RNAi (Fig 6D and 6E), and analyzed AMP expression. The results showed that the expression of the above five AMPs was reduced significantly in the intestine of STAT-RNAi shrimp (Fig 2F), a finding that is similar to the effect observed in MjCC-CL-RNAi shrimp. To confirm that the JAK/STAT pathway regulates AMP expression, we injected the shrimp with a STAT inhibitor prior to the LPS challenge, and analyzed STAT translocation and AMP expression. The results showed that in the STAT inhibitor-injected shrimp STAT did not translocate into the nucleus of the hemocytes upon LPS challenge (Fig 6G and g), and STAT phosphorylation in the intestine was also inhibited (Fig 6H). Further, in the STAT inhibitor-injected shrimp the LPS challenge failed to upregulate AMP expression (Fig 2I). These results suggest that STAT phosphorylation and translocation are functionally related to the increased expression of ALF-A1, ALF-C1, ALF-C2, CruΙ-1 and CruΙ-5.
To identify putative STAT-binding sites in the promoter sequences of the AMPs of interest we used a genome walking approach. We succeeded in identifying putative NF-κB and STAT-binding sites (Fig 6J) in the CruΙ-1 promoter sequence (S3 Fig). Next, we conducted a chromatin immunoprecipitation (ChIP) assay with anti-pSTAT, purified and analyzed the DNA fragment obtained, and amplified by RT-PCR the CruI-1 sequence of interest (Fig 6K). Subsequently we carried out an electrophoretic mobility shift assay (EMSA) with a Dig-labeled CruΙ-1 probe containing the predicted STAT binding site, and purified the recombinant GST-STAT protein and native STAT protein from LPS-injected shrimp to confirm whether STAT directly binds to the predicted STAT binding site in the promoter sequence of CruΙ-1. These results showed that STAT could bind to the predicted STAT binding site in the CruΙ-1 promoter sequence (Fig 6L and 6N) and that the binding ability was increased in LPS-injected shrimp (Fig 6N). All these results suggested that the JAK/STAT pathway regulates the expression of AMPs.
To investigate the antibacterial function of the JAK/STAT pathway in shrimp, we first evaluated signaling activation by detecting STAT phosphorylation with an antibody specific for phosphorylated STAT (anti-pSTAT) after challenge with bacteria (V. anguillarum, Escherichia coli, Staphylococcus aureus or Bacillus subtilis), LPS (E. coli), or PGN (S. aureus or B. subtilis). The results showed that they all could induce STAT phosphorylation in the intestine 3 h post challenge, whereas no pSTAT could be detected in the intestine of control shrimp challenged with PBS (Fig 7A). As both GN and GP bacteria induced STAT phosphorylation, only the shrimp pathogen V. anguillarum and purified LPS were used in subsequent experiments. To expand the above results, we conducted a time-course analysis by assessing STAT phosphorylation in the intestine of shrimp at 1h and 3 h upon V. anguillarum challenge. The results revealed that the STAT phosphorylation increased in the intestine of V. anguillarum-challenged shrimp from 1 to 3 h after challenge (Fig 7B). We also examined STAT phosphorylation and translocation in hemocytes by immunocytochemical analysis, and the results indicated that pSTAT translocated from the cytoplasm into the nucleus 1 to 3 h after bacterial challenge (Fig 7C). We then extracted cytoplasmic and nuclear proteins from the intestine, and examined the subcellular distribution of STAT by western blotting using an anti-STAT antibody. The results showed that levels of total STAT in the intestinal tissue remained unchanged in the untreated, PBS-challenged or bacteria-challenged shrimp (Fig 7D). However, when STAT levels were analyzed separately in the cytoplasm and nucleus of the intestinal cells, when compared with the STAT levels in the cytoplasm of intestinal cells from PBS-challenged and untreated control shrimp, the STAT level at 3 h post-bacterial challenge was relatively lower (Fig 7E). Consistently, STAT was only detected in the nucleus of intestinal cells in the bacteria-challenged shrimp (Fig 7E). Taken together, these results suggest that challenge with bacteria and bacterial polysaccharides induce STAT phosphorylation and translocation into the nucleus at 3 h post-challenge in both hemocytes and intestinal cells, and indicate that bacterial challenge can activate the JAK/STAT signaling pathway in shrimp. To confirm that the JAK/STAT pathway is involved in the antibacterial response, we knocked down STAT by RNAi (Fig 7F), and comparatively analyzed bacterial clearance and the survival rate of the STAT-RNAi and GFP-RNAi control shrimp. Injection of V. anguillarum into the STAT-RNAi shrimp resulted in impaired bacterial clearance (Fig 7G), and their survival rate declined significantly compared with the GFP-RNAi control shrimp (Fig 7H). These results suggest that the JAK/STAT pathway plays an important role in antibacterial immunity in shrimp.
To further investigate the mechanism by which MjCC-CL mediates activation of the JAK/STAT pathway, we examined in vivo the potential role of the CCD on STAT phosphorylation and translocation into the nucleus of shrimp hemocytes. For this, the full length CCD (GST-CCD) expressed from E. coli, and two synthetic truncated forms, sCC1 (3–39 aa) and sCC2 (47–119 aa) (Fig 8A), were injected into shrimp and STAT phosporylation and translocation were analyzed as described above for the whole MjCC-CL. Injection of GST-CCD, resulted in significantly increased STAT phosphorylation at 2 and 3 h (Fig 8B-b), whereas no change in STAT phosphorylation levels at 1 h, 2 h, and 3 h after injection of sCC1 or sCC2 were detected (Fig 8C). Subsequently, cytoplasmic and nuclear proteins were extracted from hemocytes and the subcellular distributions of total STAT and pSTAT were assessed by WB using an anti-STAT and anti-pSTAT antibodies, respectively. The results showed that total STAT and pSTAT in the nucleus were increased at 3 h in hemocytes of the GST-CC-injected shrimp (Fig 8D-d), whereas no changes were observed in the sCC1- and sCC2 -injected shrimp (Fig 8E). We also assessed STAT translocation in hemocytes by immunocytochemical analysis, and the results showed that STAT translocated into the nucleus of the GST-CC-injected shrimp (Fig 8F and f), but no changes in sCC1- and sCC2-injected shrimp were observed (Fig 8G and g). The results strongly suggest that the CCD of MjCC-CL is responsible for activation of the JAK/STAT signaling pathway, and that the intact CCD structure is required for activity.
It has been established in Drosophila that Domeless (Dome) is the type I cytokine cell surface receptor involved in activation of the JAK/STAT pathway [19]. Thus, we investigated the possibility that in shrimp the Dome orthologue could also function as a lectin cell surface receptor, and therefore be involved in the MjCC-CL-mediated activation of the JAK/STAT pathway, upon an initial interaction of MjCC-CL with microbial pathogens. For this, we first cloned the kuruma shrimp Dome, (GenBank accession no. KX358405). Dome comprises a signal peptide, an interleukin 6 receptor (ILR) alpha domain, five fibronectin-type 3 (FN3) domains, and a transmembrane (TM) region (Fig 9A). We analyzed the potential interaction of MjCC-CL with Dome and cross-linking of bacteria by co-immunoprecipitation (co-IP), pulldowns, and bacterial binding assays. The pIEx-4-RFP plasmid with MjCC-CL or its CC and CTL domains was constructed and co-transfected into HaEpi cells [20] with the pIEx-4-RFP containing ILR domain of Dome, and a co-IP assay was performed to study the interaction between MjCC-CL and Dome. The results showed that MjCC-CL interacts with the ILR domain of Dome via its CCD (Fig 9B and 9B1) and that the CTL domain does not interact with ILR (Fig 9B2). There was no interaction between RFP and the His-tagged protein (Fig 9B3). Next, MjCC-CL and its CC and CTL domains, as well as the ILR domain of Dome, were expressed in E. coli (S1A–S1E Fig). A GST pulldown assay was performed to verify the interaction. The results showed that full-length MjCC-CL (Fig 9C) and the CCD of MjCC-CL (Fig 9C1) interacted with the ILR domain of Dome, but the CTL domain (Fig 9C2) and GST could not interact with ILR (Fig 9C3). The same results were obtained with His pulldown analysis (Fig 9D and 9D1-3). We then performed the Co-IP assay using antibodies specific for Dome or MjCC-CL (Fig 9E), and the results showed that upon Vibrio challenge Dome and MjCC-CL interacted with each other (Fig 9F). Taken together, the above results suggest that MjCC-CL functions both as a PRR of bacteria and a ligand of Dome, binding to bacteria with its CTL domain and cross-linking them to the Dome receptor with its CCD, thereby activating the of the JAK/STAT pathway.
JAK, the principal component of the JAK/STAT pathway, is present in the kuruma shrimp (GenBank accession no. KU213610). The architecture of JAK consists of a Band 4.1 homolog (B41) domain and an Src homology 2 (SH2) domain (S2A Fig). To investigate in vivo the potential role(s) of JAK and Dome in the shrimp immune response to bacterial challenge, we first examined the distribution and expression patterns of JAK and Dome. The results showed that Dome was mainly expressed in heart, gill and intestine, with lower expression in hemocytes, hepatopancrease and stomach. In contrast, JAK was similarly expressed in all tissues, with heart showing the lowest levels (S2B and S2C Fig). To confirm the role of the JAK/STAT pathway and specifically the functions of the key components Dome and JAK, we knocked them down by RNAi and analyzed STAT phosphorylation and AMP expression upon LPS challenge. Upon LPS challenge of the Dome-RNAi shrimp (Fig 10A), STAT phosphorylation decreased significantly in the intestine (Fig 10A1), STAT translocation into the nucleus was inhibited in the hemocytes (Fig 10A2 and a2), and the expression of ALF-A1, ALF-C1, ALF-C2, CruΙ-1 and CruΙ-5 was also significantly impaired (Fig 10A3). Similar results were obtained after treatment with RNAi targeting JAK (Fig 10B–10B3). The above results suggest that JAK and the MjCC-CL cell surface receptor Dome are key components of the JAK/STAT pathway and are involved in STAT phosphorylation and nuclear translocation, and AMP expression.
To investigate the possibility that MjCC-CL could activate the JAK/STAT pathway in a heterologous system, we examined the MjCC-CL mediated JAK/STAT pathway activation in mouse macrophages, using STAT3 phosphorylation as an indicator, and IL6 as a positive control for pathway activation. After treatment with rMjCC-CL, the cells were collected and examined by immunocytochemical and WB analysis. The results showed like IL6, rMjCC-CL could induce STAT3 phosphorylation in mouse macrophages (Fig 11A and 11B). The results suggested that in a mammalian system MjCC-CL functions as a cytokine to activate the JAK/STAT pathway.
In this study we identified in the kuruma shrimp a novel chimeric CTL, MjCC-CL, which by a unique “self” protein-protein interaction at the cell surface plays a central role in the shrimp antibacterial immune response. The MjCC-CL protein comprises a typical CTLD that function as a PRR for “non-self” microbial glycans, and a CCD that by interacting with the Dome receptor at the hemocyte surface, directly activates the JAK/STAT signaling pathway to upregulate the expression of five AMPs.
Upon recognition of glycans on the surface of invading microbes via the canonical CTLD, soluble CTLs may participate in antibacterial responses in several ways, such as agglutinating and immobilizing the potential pathogens, functioning as opsonins to promote their phagocytosis, and by direct microbicidal activity or indirectly by activating enzymatic pathways leading to complement activation in vertebrates, or melanization by activation of the prophenoloxidase pathway in invertebrates [8]. We recently reported that the shrimp CTL MjHeCL directly regulates the AMP levels in plasma, which in turn maintain the homeostasis of the hemolymph microbiota [7]. MjHeCL only contains a CTLD, is structurally similar to CTLs from other invertebrate species, and is mainly expressed in hemocytes at relatively high levels that are not affected by experimental microbial challenge. In contrast, in addition to the CTLD, MjCC-CL displays a CCD rich in α-helix content with overall similarity to the mammalian IL10, and that a phylogenetic analysis finds clustering with vertebrate CTLs. Further, MjCC-CL was detected in all tested tissues and its expression was significantly upregulated by microbial infection. Although both MjHeCL and MjCC-CL regulated AMP expression, given the structural differences between the two CTLs, this likely to take place through different mechanisms, and aimed at very different functional outcomes: while the former maintains homeostasis of the internal microbiota, the latter is key to immune responses for exogenous infectious challenge. Most importantly, our findings in this study revealed a novel mechanism by which a CTL activates the JAK/STAT signaling pathway and regulates AMP expression.
The JAK/STAT pathway was originally identified as a cytokine signaling pathway in mammals [21], and its relevance in the regulation of both innate and adaptive immunity has been widely recognized [13]. The JAK/STAT pathway consists of three main components: cytokine receptors at the cell surface, Janus kinases (JAKs), and signal transducers and activators of transcription (STATs). In mammals, many cell surface cytokine receptors, four JAKs (JAK1, JAK2, JAK3 and TYK2) and seven STATs (STAT1, STAT2, STAT3, STAT4, STAT5A, STAT5B, and STAT6) have been identified [22]. Further, over fifty cytokines and growth factors, including interferons, interleukins and colony-stimulating factors, have been identified as indirect activators of the JAK/STAT pathway via cell surface receptors and mediate various immune responses to infection [23–25]. Although the JAK/STAT signaling pathway is ubiquitous in vertebrates, it can also be found as an intact pathway in some invertebrate taxa [26]. Thus, an evolutionarily conserved function of the JAK/STAT signaling pathway in immune responses in humans and insects, including fruit flies and mosquitos, has been suggested [15–18]. A complete JAK/STAT pathway is found in Drosophila and mosquito [12, 18, 27]. In mollusks [28] and echinoderms [29], only some components have been identified and it remains unclear whether they have a complete JAK/STAT pathway. Other organisms, such as Caenorhabditis elegans and Dictyostelium, do not have a completely functional JAK/STAT cassette, but homologs of some JAK/STAT pathway proteins have been found [30, 31]. However, there are no JAKs in Dictyostelium; instead, STAT activation is mediated by G protein-coupled receptors [32].
In Drosophila, Toll signaling and IMD pathways control the humoral immune response to bacterial or fungal infections, leading to the production of several AMPs [33–35], and the primary function of the JAK/STAT pathway is control of the cellular immune response [36]. The Drosophila JAK/STAT signaling pathway comprises two receptor-like molecules: one is Dome, which shows weak similarities to the cytokine-binding modules of the vertebrate IL6 receptor family and functions as the receptor of the pathway. Another is Latran (or Eye transformer), which has similarity to Dome and is encoded by a predicted gene, CG14225. Latran was reported to be a negative regulator of Drosophila JAK/STAT signaling [37, 38]. Cytokine-like proteins called Upds are ligands that bind to Dome and consequently activate the JAK/STAT pathway [19]. In contrast to the well-characterized Toll and IMD pathways in Drosophila immunity, however, relatively little is known about the transcriptional responses induced by the JAK/STAT pathway in humoral immune response.
In shrimp, the main components of the JAK/STAT pathway have been identified. Only one STAT, the key component of the pathway, is present in various shrimp species, including Penaeus monodon [39], Fenneropenaeus chinensis [40] and Marsupenaeus japonicus [41]. The shrimp STAT is similar to the mammalian STAT5. In mammals, STAT5 participates in the regulation of a wide range of physiological processes of the cell, including proliferation, differentiation, survival, apoptosis, and others [42, 43]. A single receptor domeless (Dome), which shares functional and sequence similarity with the mammalian cytokine class I receptors, has been identified in Litopenaeus vannamei [44]. A JAK has also been reported in L. vannamei [45]. Unlike the antiviral function of the JAK/STAT pathway in mammals and insects, white spot syndrome virus (WSSV) uses a shrimp STAT as a transcription factor to enhance viral gene expression in host cells [46], and the pathway is helpful and beneficial for WSSV replication [44, 47]. However, another report has shown that silencing of the shrimp JAK causes higher mortality and increased viral load in L. vannamei, and the pathway has antiviral function [45]. In contrast to the function of JAK/STAT pathway in antiviral responses, relatively little is known about the function of the pathway in antibacterial defense.
In our study, the ligand of the JAK/STAT pathway receptor, MjCC-CL, was identified and found to contain a CCD with sequence similarity to IL10 and a CTLD. Through protein secondary structure prediction and SWISS-MODEL analysis, the CCD was also predicted to have a highly α-helical nature, which is highly similar in overall structure with cytokine, IL10 in mammals. The fundamental property of coiled coils is their stability. The biological functions resulting from coiled-coil stability are related with membrane fusion, transmembrane signal transduction, and interaction with different proteins [48]. The CCD binds to the Dome receptor to activate the JAK/STAT pathway. The CTL domain of the MjCC-CL binds both Gram-positive bacteria and Gram-negative bacteria by binding to different polysaccharides, including LPS and PGN, on the bacterial surface. Numerous C-type lectins have been identified in shrimp and other invertebrates, and these lectins have functional diversity [8, 49]. Therefore, this study identifies an example of direct activation of the JAK/STAT pathway and provides novel ways to identify new ligands of the JAK/STAT pathway in invertebrates. MjCC-CL from shrimp also induces STAT3 phosphorylation in mouse macrophages, suggesting that MjCC-CL function as a cytokine in mammals. Although the principal components of the JAK/STAT pathway have been identified in other invertebrate species, the ligands capable of activating the pathway have been described only in Drosophila. Three Upds that activate the JAK/STAT pathway in flies were identified, including Upd and Upd3 [15, 50]. Although the Upds bear no sequence similarity to cytokines, the predicted highly α-helical nature is consistent with an overall structure that could be similar to cytokines. In our study, the lectin MjCC-CL not only activates the JAK/STAT pathway in shrimp hemocytes, but can also function as a cytokine by activating JAK/STAT pathway in mammalian macrophages.
In the present study, significant changes in the principal components of the shrimp JAK/STAT signaling pathway were observed after challenge with bacteria or their surface glycans, such as LPS and PGN. These challenges induced the phosphorylation and translocation of STAT in shrimp, and up regulation of expression of five AMPs via JAK/STAT signaling. Further, our study revealed a putative STAT-binding site in the promoter sequence of an AMP, CruI-1. The JAK/STAT pathway is associated with various immune responses, including anti-viral and anti-bacterial responses [12, 51]. The first evidence that the JAK/STAT pathway is involved in invertebrate immunity came from studies performed in Anopheles mosquitoes, which indicated that the STAT protein accumulates in the nucleus after immune challenge [52]. A subsequent study found that the JAK/STAT pathway participates in hematopoiesis and the cellular immune response in Drosophila [12]. Gene expression profile studies have identified a subset of Drosophila immune-responsive genes that are regulated by the JAK/STAT pathway, namely the genes encoding the complement-like protein Tep2 and the Turandot stress genes, but the exact function of these genes in immunity remains elusive [53]. In shrimp, STAT is upregulated in shrimp challenged with bacteria, WSSV, poly IC and PGN [54]. STAT can enhance the replication of WSSV in Penaeus monodon [46].
In conclusion, the components of the shrimp JAK/STAT pathway were identified, including a ligand, MjCC-CL; a receptor, Dome; a kinase, JAK; and a transcription factor, STAT, and the mechanistic aspects of their interactions upon immune challenge were elucidated, as schematically illustrated in Fig 12. Upon bacterial challenge (including Gram-positive or Gram-negative bacteria), MjCC-CL can recognize the pathogens’ surface glycans, via the CTLD. MjCC-CL then interacts via the CCD with the receptor Dome to activate the JAK/STAT pathway and induce the phosphorylation and translocation of STAT, which in turn regulates the expression of ALF-C1, ALF-C2, ALF-D1, CruΙ-1 and CruΙ-5. Taken together, our results suggest that MjCC-CL functions both as a PRR for recognition of infectious pathogens, and as the ligand of Dome to activate JAK/STAT signaling, and that the MjCC-CL/Dome/JAK/STAT pathway plays an important role in the anti-bacterial response by regulating AMP expression in shrimp. Thus, our study of the shrimp MjCC-CL revealed a striking functional difference with vertebrates, in which the JAK/STAT pathway is indirectly activated by cell death and stress signals through cytokines or growth factors. Instead, by cross-linking microbial pathogens with the cell surface receptor Domeless, a lectin directly activates the JAK/STAT pathway, which plays a central role in the shrimp antibacterial immune responses by upregulating expression of selected AMPs.
The full-length cDNA sequence of MjCC-CL was obtained from shrimp intestine by transcriptomic sequencing. The sequence of MjCC-CL was amplified by RT-PCR using corresponding primers (S2 Table) and re-sequenced for confirmation. Similarity analysis was conducted using BLASTx (http://www.ncbi.nlm.nih.gov/). The corresponding cDNA was conceptually translated, and the deduced proteins were predicted using ExPASy (http://www.expasy.org/). Domain architecture prediction of the proteins was performed using SMART (http://smart.embl-heidelberg.de/). Homology modeling of MjCC-CL was performed using SWISS-MODEL (https://swissmodel.expasy.org/). MEGA 5 was used for phylogenetic analysis.
Healthy kuruma shrimp (M. japonicus; about 10 g each) were obtained from a fish market in Jinan, Shandong Province, China, and were acclimated in a laboratory aquarium tanks with aerated seawater at 23°C for 3 days before the infectious challenge. For this, each shrimp was injected in the abdomen with either 20 μl of bacteria (Vibrio anguillarum, 1 × 107 CFU in PBS, 1 ml) or PBS alone. For hemocyte collection, hemolymph was extracted from three challenged or control shrimp with a syringe containing 1 ml cold anticoagulant buffer at 4°C (0.45 M NaCl, 10 mM KCl, 10 mM EDTA and 10 mM HEPES, pH 7.45) and immediately centrifuged at 800 g for 10 min (4°C). Organs (heart, hepatopancreas, gills, stomach and intestine, were collected, and total RNA was extracted with TRIzol reagent (Cwbio, Beijing, China).
The full-length cDNA sequences of Dome, Jak and Stat were obtained from shrimp intestine by transcriptomic sequencing. All sequences were amplified by RT-PCR using specific primers (S2 Table) and the amplicons sequenced and analyzed by BLASTx (http://www.ncbi.nlm.nih.gov/) for sequence confirmation. Prediction of the domain architecture of the corresponding proteins was performed using SMART (http://smart.embl-heidelberg.de/).
The mRNA tissue distribution of five genes was analyzed using semi-quantitative RT-PCR with the primers RT-F and RT-R for MjCC-CL, Dome, JAK and STAT (S2 Table). RNA from the hemocytes, heart, hepatopancreas, gills, stomach, and intestine was used in this assay. β-actin was used as the control with the primers ActinF and ActinR (S2 Table). qRT-PCR was used to detect the expression profiles of five genes after V. anguillarum infection following a previously described method [7]. The qRT-PCR was programmed at 95°C for 10 min, followed by 40 cycles at 95°C for 10 s and 60°C for 60 s. The plate was read at 78°C for each cycle. The expression profiles of MjCC-CL, Dome, JAK and STAT were detected in the hemocytes and intestine challenged with V. anguillarum. All experiments were repeated at least three times using individual templates. The obtained data were evaluated using the 2-ΔΔCt method, as described previously [55], and statistically analyzed; significant differences in the unpaired sample t-test were accepted at p< 0.05.
The 3’-terminal sequence (about 500 bp) for siRNA of MjCC-CL, Dome and STAT were amplified by the primers Fi and Ri linked to the T7 promoter (S2 Table) and were used as templates for the synthesis of dsRNA. The cDNA fragment of GFP used for dsGFP synthesis was amplified using the primers GFP-Fi and GFP-Ri (S2 Table). The dsRNA was synthesized using T7 polymerase (Fermentas, USA) based on the method of Wang et al [56]. The RNA interference (RNAi) assay was performed as described in previous reports [57]. dsRNA (20 μg) for MjCC-CL, Dome or STAT was injected into the abdominal segment of each shrimp. To enhance the RNAi effect, a second injection was performed 12 h after the first injection. dsGFP was used as a control. The intestine was collected from the shrimp 24 h after the second injection, and total RNA was extracted and assessed by qRT-PCR using the primers RT-F and RT-R (S2 Table) to evaluate the efficacy of the RNAi. To screen potential AMPs regulated by JAK/STAT signaling, AMP expression in challenged shrimp after receiving STAT-RNAi plus challenge were analyzed. The RNA interference was followed previous mentioned method. The shrimp were divided into four groups: one control group received PBS injection and the other RNAi groups were injected with V. anguillarum, lipopolysaccharides (LPS; E. coli, Sigma), peptidoglycan (PGN, S. aureus, Sigma; PGN, B. subtilis, Sigma). AMPs, including ALF-A1, ALF-C1, ALF-C2, ALF-D1 [58], CruΙ-1 and CruΙ-5, were detected by qRT-PCR with specific primers (S2 Table) after 6 h of bacterial challenge. The above AMPs were detected after STAT-knockdown and then challenged with LPS. dsGFP was used as the control.
After the RNAi method was established, a bacterial clearance assay was performed. Shrimp were separated into three groups and injected with dsMjCC-CL, dsSTAT or dsGFP as a control. Then, the three groups were injected with V. anguillarum (20 μl, 1 × 109 CFU). Thirty minutes after bacteria injection, shrimp hemolymph was collected, diluted and then cultured on solid LB plates overnight. The numbers of bacterial colonies were counted. The assay was repeated three times.
Shrimp (30 shrimp per group, about 10 g each) were divided into three groups to evaluate the shrimp survival rate after STAT and MjCC-CL knockdown and V. anguillarum infection. dsGFP was used as a control. After STAT and MjCC-CL were knocked down by dsRNA injection, all shrimp were injected with V. anguillarum (20 μl, 1 × 1010 CFU in PBS, 1 ml). The number of dead shrimp was monitored every day, and the survival rates of the three groups of shrimp were calculated. The experiments were repeated three times. The data were statistically analyzed by t-test, and a difference was considered to be significant at p< 0.05.
MjCC-CL, the CCD and CTLD of MjCC-CL, CTL2 (as a control C-type lectin), and the ILR domains of Dome and STAT were recombinantly expressed in Escherichia coli. The sequences of the above five proteins were amplified from shrimp hemocytes using the primers ExF and ExR (S2 Table). The PCR procedure was as follows: one cycle at 95°C for 3 min; 35 cycles at 94°C for 30 s, 55°C for 45 s, and 72°C for 45 s; and one cycle at 72°C for 10 min. The PCR products were then cloned into the pET32a (Novagen) or pGEX4T-1 (GE Healthcare) vectors. The recombinant proteins were purified by affinity chromatography using His-Bind resin (Ni2+-resin; Novagen, Darmstadt, Germany) or GST-resin (GenScript, Nanjing, China) following the manufacturer’s instructions. MjCC-CL, STAT antiserum preparation was performed as previously described [59]. The truncated forms of CCD (3–39 aa) and (47–119 aa) from MjCC-CL (sequences were shown in Fig 6A) were synthesized by DgPeptidesCo., Ltd (Hangzhou, China), and named as synthesized truncated CC1, sCC1 (3–39 aa) and synthesized truncated CC2, sCC2 (47–119 aa). Dorsal and Relish expression and antiserum preparation were carried out as previously described [60, 61].
Gram-positive bacteria (B. subtilis and S. aureus) and Gram-negative bacteria (E. coli and V. anguillarum) were used to test the binding activity of recombinant MjCC-CL and the CTL and CCDs of MjCC-CL. Bacteria were cultured in 2–4 mL of Luria-Bertani (LB) medium (1% tryptone, 0.5% yeast extract, and 1% NaCl) overnight and then gathered by centrifugation at 6000 g for 5 min. After washing three times with TBS, the bacteria were resuspended in TBS and adjusted to an OD600 of 1.0. The bacteria (400 μL) in TBS were incubated with purified rMjCC-CL, rCTLD and rCCD (100 μg) for 60 min at 28°C with rotation, collected by centrifugation, and then washed four times with TBS. Finally, the bound proteins were eluted with 7% SDS for 1 min and subjected to 12.5% SDS-PAGE. The proteins in the gel were transferred to a nitrocellulose membrane for western blot analysis. An anti-histidine antibody (ZSGB Bio, Beijing, China, 1:3000 dilution in TBS containing 5% nonfat milk) was used as the primary antibody, and secondary antibody was alkaline phosphatase-conjugated horse anti-mouse IgG (ZSGB Bio, Beijing, China, 1:10,000 dilution in TBS containing 5% nonfat milk). Native MjCC-CL protein was purified from shrimp intestine according to a previous report [7] and also incubated with four bacteria. The mixture was gently rotated at 28°C for 1 h. Bacterial pellets were collected by centrifugation at 6000 g for 5 min, washed three times with Tris buffer (pH 8.0) and then analyzed by western blotting.
An enzyme-linked immunosorbent assay (ELISA) was used to test the direct binding activity of rMjCC-CL, rCTLD and rCCD to different bacterial cell wall components. LPS from E. coli and PGN from S. aureus and B. subtilis separately were chosen for the assay. Each well of the microplate was coated with 2 μg of the polysaccharide and incubated at 37°C overnight. The microplate was incubated at 60°C for 30 min, blocked with BSA (1 mg/mL, 200 μL) at 37°C for 2 h, and washed with TBS (200 μL). Purified rMjCC-CL, rCTLD and rCCD (final concentration 0–20 μg/mL in TBS with 0.1 mg/mL BSA) was added to each well of the coated plates and incubated at room temperature for 3 h. The plate was then washed four times with TBS, and alkaline phosphatase-conjugated horse anti-mouse IgG (1:3000 dilution in binding buffer containing 0.1 mg/mL BSA) was added (100 μL per well) and incubated at 37°C for 2 h. After the plate was washed four times with TBS, the color was developed with p-nitro-phenyl phosphate (1 mg/mL in 10 mM diethanolamine and 0.5 mM MgCl2) at room temperature for 30 min. The OD value was read at 405 nm. Each binding assay was performed three times. The dissociation constants (Kd) and maximum binding (Bmax) parameters were calculated by GraphPad Prism version 5.00 software for Windows (San Diego, CA, USA).
Gram-positive bacteria (B. subtilis and S. aureus) and Gram-negative bacteria (E. coli and V. anguillarum) were used to test the potential antibacterial activity of recombinant MjCC-CL. Each protein (10, 30, 50, 100 μg) was added to a 96-well culture plate which contained 180 μL of mid-log phase bacteria (2 × 105 CFU) cultured in Poor Broth (1% tryptone, 0.5% NaCl (w/v) and pH 7.5). The plate was incubated for 24 h at 28°C and the absorbance at 600 nm was measured using an ELX800 Universal Microplate Reader (Bio-Tek Instruments, INC) to evaluate the bacterial concentration. The assays were repeated thrice. GST protein was used as negative control.
Recombinant MjCC-CL was used for the “overexpression” assay. rMj-CC-CTL (20 μg per shrimp) was injected into shrimp for 1 and 2 h. GST was used as control. Hemocytes were then collected for immunocytochemical assay and intestine protein was extracted for western blotting to detect Dorsal, Relish and STAT translocation into nucleus. rMj-CC-CTL, sCC1 (3–39 aa) and sCC2 (47–119 aa) were also injected into shrimp to detect STAT phosphorylation and translocation into nucleus by western blotting assay and immunocytochemical assay. To confirm the function of MjCC-CL in JAK/STAT pathway activation, rMj-CC-CTL and rCTL2 incubated with LPS (4 μg) were injected into shrimp for 3 h. GST was used as control. Hemocytes were then collected for immunocytochemical assay and intestine protein was extracted for western blotting to detect Dorsal, Relish and STAT translocation into nucleus. The AMPs were also assessed in intestine in rMj-CC-CTL- and rCTL2-injection shrimp at 6 h.
Western blotting was used to detect STAT phosphorylation with the commercial anti-STAT5 (phospho Tyr694) antibody (Abcam, USA). Tissue proteins were obtained from the hemocytes and intestine of normal shrimp and bacterially challenged shrimp. Cytoplasmic proteins and nuclear proteins were extracted using a Nuclear Protein Extraction Reagent Kit (BioTeke, China) following the manufacturer’s instructions. The samples were separated by 12.5% SDS-polyacrylamide gel electrophoresis following the Laemmli method [62]. The proteins in the gel were then transferred onto nitrocellulose membranes. The membranes were blocked for 1 h with 3% non-fat milk in TBS (10 mM Tris-HCl, pH 7.5, 150 mM NaCl) and incubated with 1/100 diluted antiserum against STAT, phosphorylated STAT, Dorsal, Relish or β-actin in TBS with 3% non-fat milk for 2 h. Then, alkaline phosphatase-conjugated goat anti-rabbit IgG (1/10,000 diluted in TBS) was added after washing to remove the free, nonspecifically binding antiserum, and the membranes were incubated for 2 h. The membrane was dipped in the reaction system (10 ml of TBS with 45 μl of NBT and 35 μl of BCIP) in the dark for 5 min to visualize the signal.
Hemolymph obtained from shrimp was fixed with 1 ml of a mixture containing anticoagulant (pH 7.4) and 4% paraformaldehyde and then centrifuged at 600 g for 10 min at 4°C. The collected hemocytes were deposited onto a glass slide, washed with PBS (140 mM NaCl, 10 mM sodium phosphate, pH 7.4) and incubated in 0.2% Triton X-100 at 37°C for 5 min. After washing with PBS, the hemocytes on the glass slides were blocked with 3% BSA (30 min, 37°C) and incubated with anti-pSTAT, anti-STAT, anti-Dorsal or Anti-Relish (1:400 in 3% BSA) overnight at 4°C. The hemocytes were then washed with PBS and incubated with 3% BSA for 10 min; the Alexa Fluor 488-conjugated second antibody to rabbit (1:1,000 ratio, diluted in 3% BSA) was then added, and the samples were incubated for 1 h at 37°C in the dark. After being washed three times, the hemocytes were incubated with 4’-6-diamidino-2-phenylindole dihydrochloride (DAPI, AnaSpec Inc., San Jose, CA; 1 μg/ml in PBS) for 10 min at room temperature and washed six times. Fluorescence was observed under an Olympus BX51 fluorescence microscope (Shinjuku-ku, Tokyo, Japan). WCIF ImageJ software was used to analyze the colocalization of STAT and DAPI-stained nuclei in hemocytes.
As STAT in M. japonicus is clustered together with STAT5 in human (S1 Fig), STAT5 inhibitor, (573108-10MG, Merck) (2 μg) was used to inject into each shrimp and then challenged by LPS. DMSO injection was used as a control. The intestine was collected for protein and RNA extraction at 3 and 6 h after LPS challenge. The phosphorylation level of STAT was analyzed by western blotting with anti-pSTAT antibody, and the AMP expression was detected by qRT-PCR at 6 h after LPS challenge.
The promoter sequences of antimicrobial peptides regulated by the JAK/STAT pathway were cloned using Genome Walker Kits (Takara, Japan) following the manufacturer’s instructions. The specific primers against CruΙ-1 (G-CruΙ-1-1, G-CruΙ-1-2 and G-CruΙ-1-3) used for genome walking are listed in S2 Table. ChIP was performed following previously described methods [63, 64] using the primers CruΙ-1 RTF/R for CruΙ-1 (S2 Table).
The shrimp intestine was lysed, and STAT protein was purified using anti-STAT antibody- CNBr-activated Sepharose 4B (60 mg; Amersham Biosciences AB, Uppsala, Sweden). The digoxigenin-labeled probes (sense, 5'-GCGTAAGGTTTTCTTGGAATA-3'; antisense, 5'-TATTCCAAGAAAACCTTACGC-3') were synthetized and labeled by Sangon Company (China). Two micrograms of purified proteins was mixed with 3 μl of 5× binding buffer (Beyotime Institute of Biotechnology, Shanghai, China) for 10 min and incubated with 20 fmol of digoxigenin-labeled probe for 20 min. In competition experiments, unlabeled probe was pre-incubated with the relevant proteins for 10 min before the Dig-labeled probe was added and incubated for 20 min at room temperature. The reaction solution was run on a 6% polyacrylamide/0.5× TBE gel at 80 V, and the samples were transferred onto a nylon membrane (IMMOBILON-NY+, Millipore, Milford, MA, USA). The membrane was first blocked with blocking buffer for 30 min and then incubated with anti-Dig phosphatase antibody (1:10000 in blocking solution; Roche, Germany) for 1h. The signal was visualized with 5-bromo-4-chloro-3-indolyl phosphate and nitroblue tetrazolium chloride.
To analyze the interaction of MjCC-CL with Dome, co-immunoprecipitation was performed after these molecules were overexpressed in HaEpi cells. The appropriate cDNA sequences encoding whole MjCC-CL or the CC region and CTL domain of MjCC-CL were amplified with primers (S2 Table) and inserted into the pIEx-4-RFP plasmid (with a C-terminal red fluorescent protein tag); then, the ILR domain of Dome was also amplified and inserted into the pIEx-4 plasmid (with a His tag). HaEpi cells [20] were incubated in a 6-well tissue culture plate containing 2 ml of Grace’s medium with 10% FBS at a density of 70% to 90%. Before transfection, the cells were pre-incubated in Grace’s medium for 1 h. Afterward, 8 μg of vector DNA and 8 μg of DNAfectin transfection reagent (Tiangen, Beijing, China) were mixed, suspended in 200 μl of Grace’s medium and incubated for 20 min; then, this solution was added to the medium in the culture plate. After 12 h, the cells were re-fed in Grace’s medium containing 10% FBS and cultured for an additional 48 h. The cells were then harvested and washed twice with ice-cold 1× PBS. Afterward, the cells were re-suspended in SDS-lysis buffer (1% SDS, 10 mM EDTA, 50 Mm Tris-HCl, pH 8.1), and the lysates were pre-cleared with protein A resin at 4°C for 1 h and incubated with anti-GFP antibody at 4°C overnight. The mixture was then incubated with protein A resin at 4°C. After 2 h, the complex was washed three times and analyzed by western blotting with anti-His antibody.
To further confirm the interaction of MjCC-CL with Dome, whole MjCC-CL, the CC and CTL domains of MjCC-CL with GST tags and the ILR domain of Dome with a His tag were expressed in E. coli, and GST-pulldown and His-pulldown were performed. Recombinant proteins (30 μg) were added to 20 μl of glutathione resin (for GST-tagged proteins) or charged Ni-NTA beads (for His-tagged proteins) and incubated at room temperature for 2 h with slight rotation. The mixture (resin and binding proteins) was washed three times by centrifugation at 500 g for 3 min to remove the unbound proteins. The test protein with a His tag or GST tag was added into the mixture and was gently rotated at room temperature for 2 h. After the resin was washed three times, bound proteins were eluted and analyzed by SDS-PAGE.
Co-immunoprecipitation (Co-IP) assay. Proteins from shrimp intestine were extracted with lysis buffer (150 mM NaCl, 1.0% Nonident-P40, 0.1% SDS, 50 mM Tris [pH 8.0]) and incubated with protein A for 10–15 min to remove non-specific binding proteins. Then, proteins were incubated with antibodies specific for Dome or MjCC-CL for 3 h at room temperature, after which the mixture was incubated with protein A for 3 h at room temperature, and the pellet washed with PBS five times. The resulting pellet (bound protein, antibody and protein A) was analyzed by western blot.
The mouse primary peritoneal macrophages were obtained from Dr. Cheng-Jiang Gao laboratory in Medical School of Shandong University, the procedure of cell isolation following previous report [65]. The cells were cultured at 37°C under 5% CO2 in DMEM supplemented with 10% FCS (Invitrogen Life Technologies), 100 U/ml penicillin, and 100 μg/ml streptomycin. IL6 (ProSpec, Israel), rGST and rMjCC-CL (20 ng/ml) were added in the cell culture for 30 min. Then the cells were collected and used for immunocytochemical assay and western blotting. PSTAT3 antibody (Abcam, USA) was used as the first antibody to detect the STAT3 phosphorylation in mouse macrophages.
|
10.1371/journal.pntd.0002121 | A Cell-surface Phylome for African Trypanosomes | The cell surface of Trypanosoma brucei, like many protistan blood parasites, is crucial for mediating host-parasite interactions and is instrumental to the initiation, maintenance and severity of infection. Previous comparisons with the related trypanosomatid parasites T. cruzi and Leishmania major suggest that the cell-surface proteome of T. brucei is largely taxon-specific. Here we compare genes predicted to encode cell surface proteins of T. brucei with those from two related African trypanosomes, T. congolense and T. vivax. We created a cell surface phylome (CSP) by estimating phylogenies for 79 gene families with putative surface functions to understand the more recent evolution of African trypanosome surface architecture. Our findings demonstrate that the transferrin receptor genes essential for bloodstream survival in T. brucei are conserved in T. congolense but absent from T. vivax and include an expanded gene family of insect stage-specific surface glycoproteins that includes many currently uncharacterized genes. We also identify species-specific features and innovations and confirm that these include most expression site-associated genes (ESAGs) in T. brucei, which are absent from T. congolense and T. vivax. The CSP presents the first global picture of the origins and dynamics of cell surface architecture in African trypanosomes, representing the principal differences in genomic repertoire between African trypanosome species and provides a basis from which to explore the developmental and pathological differences in surface architectures. All data can be accessed at: http://www.genedb.org/Page/trypanosoma_surface_phylome.
| The African trypanosome (Trypanosoma brucei) is a single-celled, vector-borne parasite that causes Human African Trypanosomiasis (or ‘sleeping sickness’) throughout sub-Saharan Africa and, along with related species T. congolense and T. vivax, a similar disease in wild and domestic animals. Together, the African trypanosomes have significant effects on human and animal health and associated costs for socio-economic development in Africa. Genes expressed on the trypanosome cell surface are instrumental in causing disease and sustaining infection by resisting the host immune system. Here we compare repertoires of genes with predicted cell-surface expression in T. brucei, T. congolense and T. vivax and estimate the phylogeny of each predicted cell-surface gene family. This ‘cell-surface phylome’ (CSP) provides a detailed analysis of species-specific gene families and of gene gain and loss in shared families, aiding the identification of surface proteins that may mediate specific aspects of pathogenesis and disease progression. Overall, the CSP suggests that each trypanosome species has modified its surface proteome uniquely, indicating that T. brucei, T. congolense and T. vivax have subtly distinct mechanisms for interacting with both vertebrate and insect hosts.
| African trypanosomes (Trypanosoma spp. section Salivaria) are unicellular hemoparasites of vertebrates. They are transmitted by Tsetse flies (Glossina spp.) and cause endemic disease throughout sub-Saharan Africa. African trypanosomes include T. brucei which causes Human African Trypanosomiasis (‘sleeping sickness’) and, along with two related species T. congolense and T. vivax, a similar disease in domestic and wild animals (‘nagana’). Although the incidence of human disease has recently declined [1], there remains an estimated 30,000 cases per year [2]; while total losses in agricultural productivity due to animal disease across Tsetse-infested Africa are estimated to be US$4.75 billion per annum [3]. The combined effects of African trypanosomes on humans and livestock are a significant threat to public and veterinary health, and wider socio-economic development [4].
The first genomic comparisons between T. brucei and related trypansomatid parasites, T. cruzi and Leishmania major, which cause Chagas disease and leishmaniasis in humans respectively, showed that most genes are widespread and arranged into regions of conserved synteny [5]–[7]. By contrast, it was also apparent that the gene families likely encoding cell surfaces molecules were non-homologous and largely lineage-specific [8]–[9]. In the vertebrate host, the T. brucei surface is dominated by the Variant Surface Glycoprotein (VSG); serial replacement of VSG (i.e. antigenic variation) is a means of immune evasion and results in chronic infection [10]. African trypanosome genomes contain large VSG gene families [11]–[12], but mono-allelic expression of a single gene is ensured because transcription is restricted to telomeric VSG expression sites (ES) [13]–[15]. Several other Expression Site-Associated Genes (ESAG1-12; [16]–[18]) are located in the ES and are co-transcribed with the active VSG [19]–[20]; all but ESAG8 are predicted or known to be cell surface-expressed [21]. T. cruzi and L. major also possess multi-copy surface glycoprotein families (i.e. mucins and amastins respectively) but these are unrelated to VSG [8]–[9]. Indeed, Leishmania promastigotes have a largely non-proteinaceous, lipophosphoglycan-based surface coat [9].
Hence, while T. brucei, T. cruzi and L. major have physiological similarities associated with shared ancestry, the cell-surface architectures are highly divergent, reflecting the evolution of specific mechanisms for immune evasion and survival by each parasite [22]. A principal objective of comparative genomics is to identify taxon-specific features that may plausibly explain such phenotypic differences. Despite their similarities T. brucei, T. cruzi and L. major diverged long ago; so surface features that appear exclusive when their genomes are compared are not necessarily species-specific, or diagnostic of the diseases they cause. In particular, it remains to be determined if the T. brucei-specific surface features identified from these initial comparisons are truly species- or disease-specific, or general features of all African trypanosomes. Comparisons between more closely related species are essential to resolving this issue.
We recently reported the draft genome sequences for T. congolense, the closest known relative of T. brucei, and T. vivax, a more distantly related species, and described the evolution of VSG genes in African trypanosomes [12]–[23]. All species cause chronic animal trypanosomiasis characterized by recurrent parasitaemia and antigenic variation, but subtle differences are present in their pathology, life cycle and host range. For example, T. vivax can cause hyperacute hemorrhagic disease in cattle typically with much higher mortality than other species [24]. In the Tsetse, T. brucei and T. congolense infect the midgut but then migrate to the salivary glands and proboscis respectively prior to transmission to the vertebrate. In contrast, T. vivax avoids the insect midgut, a feature that seems to facilitate wholly mechanical transmission and its colonization of Tsetse-free areas [24]. Further, all three species infect a wide range of domestic animals but only T. brucei has evolved human infectivity, probably on at least two occasions in east (T. b. rhodesiense) and west Africa (T. b. gambiense) respectively [25].
Cell surface-expressed gene families encode abundant proteins at the forefront of host-parasite interactions [8]–[9], [22], [26]–[27]. The major surface protease (MSP, or gp63) has multiple isoforms, one of which (MSP-B) is responsible for cell-surface remodelling prior to transmission into the vector [28]–[29]. Papain-type cysteine peptidase B and C (also known as cathepsin-L and -B) are strongly associated with virulence phenotypes, degrading host proteins [30]–[31] and facilitating parasite transversal of the blood-brain barrier [32]. Other gene families encode diverse cell surface receptors, e.g. adenylate cyclases [33], and membrane transporters that are essential for normal cell physiology, e.g. transferrin receptors (TFR) [34]. Hence, the cell surface is an intuitive place to begin exploring species differences and here we present phylogenetic analyses of all gene families with predicted cell-surface roles in African trypanosomes. Although we do not include low-copy number features or non-protein cell-surface components, which may be equally important in function, our detailed analysis of the principal cell-surface gene families presents a global picture of evolutionary change on the trypanosome cell-surface.
The African trypanosome cell surface phylome is a collection of phylogenies for gene families with predicted cell surface expression. The approach is summarized in Figure S1. Phylogenies were estimated from sequence data accessed through the GeneDB portal [35] and extracted from four genome sequences: Trypanosoma brucei TREU927 [11], T. congolense IL3000 and T. vivax Y486 [12] and, to provide an outgroup in phylogenetic comparisons, T. cruzi CL Brener [5]. Genome sequencing and annotation methods have been described previously [6], [12].
All T. brucei genes with cell surface motifs, (i.e. a predicted signal peptide, a predicted GPI anchor or a trans-membrane helix) were extracted from the T. brucei 927 genome sequence. Genes annotated as ‘unlikely’ or with fewer than 100 codons were removed. Homologs to each T. brucei ‘surface’ gene were identified among all T. brucei, T. congolense, T. vivax and T. cruzi predicted genes using wuBLAST [36]. Where at least four homologs occurred in at least one species, this constituted a ‘family’ amenable to phylogenetic analysis. Surface-expressed genes with fewer than four homologs are recorded as singleton, paired and triplet sequences in tables available from the CSP webpage. After removing genes already identified as homologous to T. brucei genes (i.e. widespread gene families), the BLAST exercise was repeated for T. congolense and T. vivax genes to identify cases absent in T. brucei. Signal peptides were predicted using SignalP [37], GPI anchors were predicted using Fraganchor [38] and trans-membrane helices were predicted using TMHMM [39]. 205 ‘surface expressed’ families were reduced to 79 by removing cases of poor alignment (i.e. sequences that could not be aligned by eye), of mis-annotation (i.e. non-coding sequence), of redundancy (i.e. technical duplicates arising from alleles in the T. congolense genome that were separately assembled), of genes with known expression in mitochondrial, lysosomal or other internal membranes, and by combining families with overlapping homology. Surface-expressed families may have been omitted because they possess signal peptides, GPI anchors, or trans-membrane helices that cannot be reliably recognized by current methods, or because their 5′ or 3′ regions are mis-specified. Equally, spurious recognition of these domains in hypothetical proteins (mostly T. vivax families) cannot be excluded. Each family is given a ‘Fam’ number (0–81) as described in Table S1; note that for historical reasons, there is no Fam48 or 68.
Given that most species-specific genes are putative and encode hypothetical proteins, evidence in support of their coding status was gathered from three sources: i) transcriptomic studies of T. brucei [40]; ii) Expressed Sequence Tags (EST) in multiple life stages of T. congolense [41]; and iii) partial RNAseq data for bloodstream form T. vivax [12] mapped against the T. vivax genome using SMALT [42].
Translated nucleotide sequences for each family were aligned in ClustalW [43]; all multiple alignments were then manually edited in BioEdit 7.1.3. [44]. In most cases, the amino acid sequence alignment was used in phylogenetic analysis to reduce homoplasy, but nucleotide sequences were examined in cases of low sequence divergence. The rates of synonymous (ks) and non-synonymous substitutions (ka) per site were calculated for each alignment using KaKs Calculator 2.0 [45] to estimate within-family sequence diversity.
Bayesian phylogenies were estimated using MrBayes v3.2.1 [46] under these settings: Nruns = 4, Ngen = 5000000, samplefreq = 500 and default prior distribution. Nucleotide and amino acid sequence alignments were analyzed using GTR+Γ and WAG+Γ models respectively. Maximum likelihood phylogenies were estimated using PHYML v3.0 [47] under an LG+Γ model [48] for amino acid sequences or a GTR+Γ model for nucleotide sequences. Node support was assessed using 100 non-parametric bootstrap replicates in addition to Bayesian posterior probabilities. Trees were rooted using T. cruzi sequences, or otherwise mid-point rooted. VSG phylogenies were estimated using alignments of selected, full-length sequences representative of global diversity under different conditions, as described previously [12].
The CSP contains phylogenies of gene families drawn from multiple species. We can infer historical gene duplications and losses from comparison of gene family phylogenies with the overlying species evolution [49]–[50]. For each gene family, a fully binary, rooted gene tree was integrated across the species tree (i.e. [T. brucei, T. congolense], T. vivax], T. cruzi]) using NOTUNG 2.6 [51]. A parameter ρ, was calculated from the ratio of speciation duplications (i.e. nodes supporting orthologs in daughter species) to unilateral duplications (i.e. nodes supporting in-paralogs in the same species), adjusted for gene family size. ρ reflects the degree of gene family turnover (combined incidence of gene gain and loss); high values of ρ indicate a phylogeny with minimal turnover, in which most lineages are represented by orthologs in all species. Low values indicate a phylogeny with high turnover, in which ancestral genes are frequently lost and replaced by novel duplicates, resulting in clades of species-specific in-paralogs and minimal orthology.
Significant differences in evolutionary rate between two lineages were examined using relative rates tests (RRTs; [52]). Nucleotide sequence alignments combining a given lineage, its sister taxon and an out-group (as described in Tables 1 and 2) were created and evaluated with MEGA v5.05 [53]. Where a test lineage consisted of multiple paralogous genes, the average rate difference between all comparisons is reported.
Phylogenetic incompatibility describes the presence of multiple phylogenetic signals within a single sequence alignment and is the historical signature of recombination. The Pair-wise Homoplasy Index (PHI) detects incompatibility between sites and is robust in the presence of rate heterogeneity [54], which might otherwise simulate the effects of recombination. P<0.05 for PHI indicates significant incompatibility between sites within an alignment, consistent with recombination. For each ESAG family, the index was calculated using PhiPack [54] for separate alignments of ESAGs sensu stricto and of homologous sequences from non-ES loci (unless these are largely absent, i.e. ESAG6/7, 8 and 12).
To determine mRNA expression levels for a single Fam50 family member (Tb927.7.380), quantitative real-time polymerase chain reaction (qRT-PCR) was carried out on total RNA extracted using RNeasy Mini Kit (QIAGEN). cDNA was generated using SuperScript II reverse transcriptase according to the manufacturer's instructions. qRT-PCR was carried out using three different isolated mRNA samples from four life-cycle stages (in vitro cultured bloodstream-stage and procyclic forms; in vivo cultured short stumpy bloodstream-stage; and in vivo cultured T. brucei bloodstream-stage). T. brucei Rab11 was used as a control to determine relative quantities of mRNA. The relative abundance of specific RNA was subsequently determined.
Fam1 (i.e. Tb927.6.1310) and Fam50 (i.e. Tb927.7.380) genes were synthesized by Eurogentec. Tb927.6.1310 is the most divergent of all Fam1 gene copies, so it was selected for the benefit of targeting a single copy gene in localization experiments. Tb927.7.380 is also one of five tandem copies, and was selected because it was expressed to the greatest level for all paralogs in qPCR analyses. T. brucei single marker bloodstream line cells were cultured in HMI-9 medium as described previously [55]. Ectopic expression of haemagglutinin (HA) epitope-tagged Tb927.6.1310/Tb927.7.380 at the N-terminus (following the predicted signal peptide sequence) was carried out using pXS5/pDEX-577 [56] constitutive and inducible expression vectors respectively. For Western blotting, proteins were transferred onto Immobilon (polyvinyildene fluoride) membranes and incubated with primary mouse anti-HA antibody (1∶8,000) and subsequently with secondary rabbit anti-mouse peroxidase conjugate antibody (1∶10,000, Sigma). Immunofluorescence microscopy was carried out on permeabilised and non-permeabilised transfected cells harvested at log phase.
We estimated phylogenies using Maximum Likelihood and Bayesian methods for 79 gene families in African trypanosomes with known or predicted cell-surface location. This cell-surface phylome describes how these families have diversified during the evolution of African trypanosomes, and also identifies species-specific gene families, many of which remain uncharacterized. Taken together, the CSP shows that the cell-surface architecture evolved in the common ancestor of all African trypanosomatids, and has subsequently experienced subtle changes in individual lineages, suggesting the adaptation of their common inheritance. The CSP is described in a Venn diagram (Figure 1) and in Table S1. Throughout, gene families are referred to by their ‘Fam’ number (0–81; see methods). All sequence alignments, Hidden Markov Models (HMMs) and phylogenetic trees can be accessed at: http://www.genedb.org/Page/trypanosoma_surface_phylome.
The conserved elements of the CSP, at the centre of Figure 1, generally contain cell-surface features that have been well described, including most known principal parasite effectors (i.e. MSPs, cathepsins and trans-sialidases) [26]–[27]. By contrast, genes at the periphery of Figure 1 are species-specific and mostly uncharacterized, even when they have given names in T. brucei; only 8/45 species-specific families (Fam0, 2, 3, 8, 12, 14–16) are characterized to some extent (e.g. by cellular localization) and function is only well known for two (VSG and ESAG6/7). Naturally, many trypanosome cell-surface proteins perform basic functions that are constrained by selection, resulting in small species differences (e.g. Fam54-56, 59-60, 62-65, 69-76 and 78-81). However, a widespread family is not necessarily unchanged, and the phylogenies of several conserved families involved in host-parasite interaction indicate surface proteome differences between species that could have functional implications.
In T. vivax, whole lineages have been lost, and on multiple occasions; for example among trans-sialidase genes (see Fam47 CSP page), there are no T. vivax orthologs to basal-branching lineages represented in T. brucei by Tb927.5.440 and Tb927.2.5280, which are otherwise widespread. Similarly, there are only three Major Facilitator Superfamily (MFS) transporters loci in T. vivax compared with six in T. brucei (see Fam58 CSP page), and no orthologs to the Proteins Associated with Differentiation (PAD) genes, one of which encodes a carboxylate transporter implicated in differentiation from vertebrate to insect life stages in T. brucei (i.e. Tb927.7.5930; [57]). Such within-family losses may coincide with the expansion of the remaining lineages. For instance MSP-B (Fam46) is present in T. brucei, T. congolense and the outgroup T. cruzi, but is absent from T. vivax; (a result confirmed by searching T. vivax unassembled reads for reciprocal BLASTx matches to MSP-B). This coincides with the evolution of 11 MSP-C genes in T. vivax, a gene that is single-copy in all other species (see Fam46 CSP page, and Table S1).
The surface functional repertoire also diverges through gene gain, for example among Fam61 genes (nucleoside/nucleobase transporters), required to scavenge host purines and are functionally differentiated with respect to both parasite life stage and substrate [58]–[61]. The Fam61 phylogeny shows that multiple gene duplications have occurred in both T. brucei and T. congolense (see Fam61 CSP page). However, while T. brucei has elaborated its nucleoside transporter lineage, producing four species-specific loci from a single-copy ancestral locus (probably Tb09.160.5480), T. congolense instead diversified its nucleobase transporter lineage, with 18 gene copies compared with three in T. brucei and five in T. vivax. This is not simply a difference in gene dosage, or an artifact of sequence assembly, since seven of these T. congolense-specific transporters (e.g. TcIL3000.0.12740) have a highly derived predicted protein sequence, lacking ∼130 amino acids from the 3′ end and displaying only 39% amino acid identity with the T. congolense chromosome 11 isoform (54% similarity), and which itself displays 54% identity and 66% similarity with its T. brucei ortholog. Therefore, these genes are predicted to encode proteins with signal peptides and eight trans-membrane helices, but lack the canonical C-terminus of the conserved nucleobase transporter including its GPI-anchor signal.
The combined effect of gene gains and losses, i.e. gene family turnover, is reflected in the topology of phylogenies. Typically, gene families predate contemporary genomes, and orthologs in each species of each ancestral gene form a clade in the phylogeny. Examples of this familiar pattern in trypanosomes naturally include structural or metabolic gene families displaying little innovation [62]–[64], as well as some CSP families including Fam56 (ABC transporters) and Fam65 (aldehyde dehydrogenase), although the majority of these genes are likely intracellular. Many cell surface-expressed gene families similarly originate prior to contemporary species, but their tree topologies indicate greater post-speciation innovation. To investigate the extent to which species derive novel genes post-speciation, we calculated ρ for each family, the ratio of orthology (DIV) to paralogy (DUP), corrected for gene family size, and where DIV is the incidence of gene divergence through speciation and DUP is the incidence of gene duplication, inferred through phylogenetic reconciliation (Table S1). Families like Fam56 (ρ = 0.67) and Fam65 (ρ = 0.73) possess high ρ values, indicating that most loci are retained in all species; for example, across 22 ABC transporter loci there are no unilateral gene duplications and only 7 gene losses (2 in T. brucei/T. congolense, 1 in T. congolense and 4 in T. vivax). While these losses probably have functionally consequence, Fam56 and similar examples have a relatively constant gene complement.
Conversely, many familiar cell surface components have ρ<0.05, indicating that gene copies cluster more by species than by locus, i.e. recent paralogy rather than ancient orthology. Fam54 (amino acid transporters; ρ = 0.006), Fam58 (MFS transporters; ρ = 0.018) and Fam61 (nucleoside transporters; ρ = 0.01) all display low ρ values due to T. brucei-specific expansions (see individual CSP pages), which occurs against a general background of conservation. This cannot be said for phylogenies for other families, e.g. Fam46 (ρ = 0.013), Fam49 (ρ = 0.002), Fam50 (ρ = 0.003), Fam67 (ρ = 0.003), and Fam77 (ρ = 0.002), which consist of species-specific clades of highly similar, tandem duplicates, at one or a few conserved loci. Fam47 (trans-sialidase; ρ = 0.025) and Fam51 (adenylate cyclase; ρ = 0.002) provide examples intermediate between the first and second patterns, with T. brucei and T. congolense possessing orthologs to conserved loci, while all T. vivax genes are monophyletic and hence lack orthology with other species.
Gene family diversification is a product of both gene duplication and sequence divergence [65], so even where gene repertoire is constant, significant asymmetry in nucleotide substitution rates between ancestral and duplicated lineages may indicate that important functional change has occurred in either lineage. Previously, we have identified frequent rate asymmetry following gene duplication of amino acid transporters (Fam54) in T. brucei [66]. Further examples are evident in the CSP. For instance, branch lengths among cysteine peptidase B (Fam67) genes in T. congolense (average genetic distance = 0.092, n = 16) are significantly longer than in T. brucei (0.0037, n = 11, p<0.0001; t-test) or T. vivax (0.016, n = 6, p<0.0001). T. congolense cysteine peptidase B includes structural variants with distinct catalytic functionality [67], which is clearly absent from T. brucei. Table 1 records this and other cases of rate asymmetry involving species-specific expansions, further details of which are provided in each CSP family page.
A TFR is expressed in bloodstream form T. brucei and is required for iron uptake [68]. It is not homologous with its mammalian counterpart, and they function quite differently [68]. The trypanosome TFR is a GPI-anchored heterodimer encoded by paralogous gene families ESAG6 and 7 (Fam15; [68]–[70]). ESAG7 is 57 amino acids shorter than ESAG6 and encodes a protein without a GPI-anchor signal, but otherwise the genes are very closely related [71]. When present, ESAG6 and 7 are found in tandem immediately downstream of the ES promoter. Outside of the ES, genes homologous to ESAG6/7 in T. brucei 927 consist of a single ESAG6/7 tandem pair (Tb927.7.3250/3260) at a strand-switch region on chromosome 7, probably representing a secondary transposition from an ES, and the Procyclin-Associated Genes (PAG1, 2, 4 and 5; Fam14), which are adjacent to the procyclin loci [72]. ESAG6 and 7 and the PAGs are homologous to the a-type VSGs (a-VSG; [69], [71]), leading to the suggestion that the TFR derives from VSG [26], [73].
The T. congolense genome contains 45 genes (in Fam15) that are homologous to ESAG6/7, plus 31 genes (in Fam14) whose closest sequence match is to PAGs in T. brucei. We refer to both Fam14 and Fam15 as TFR-like genes. Figure 2 describes the phylogeny for TFR-like genes and shows that the T. congolense genes are paraphyletic, that is, there are two clades (Fam14 and 15) each more closely related to sequences in T. brucei (PAG and ESAG6/7 respectively) than to each other. Given the homology between a-VSG and TFR genes, this shows that Fam14 and 15 are not a-VSG (of which T. congolense has none; [12]) because they are much closer to T. brucei TFR than a-VSG. The T. vivax genome contains a-VSG-like genes but these have an equally distant relationship to both ESAG6/7 and a-VSG in T. brucei, and significantly are not part of the TFR gene family of T. brucei and T. congolense [12]. Therefore, genes that now encode TFR proteins, and others associated with procyclin expression sites in T. brucei, likely evolved before the speciation of T. brucei and T. congolense, but after the separation from T. vivax.
The essential difference between TFR genes in T. brucei and T. congolense is genomic distribution. While ESAG6/7 are almost exclusively found in ESs, T. congolense orthologs are distributed widely among subtelomeres and not usually close to telomeres. Nevertheless, phylogenetic and sequence comparisons suggest that TFR function is conserved in T. congolense. First, like ESAG6/7, Fam15 genes in T. congolense split into two equal-sized sister clades, encoding proteins that differ in the prediction of a GPI anchor (Figure 2). Second, just as ESAG6 and 7 are typically arranged in GPI+/GPI− tandem pairs in T. brucei, 28/45 of T. congolense genes are also arranged in tandem pairs at subtelomeric loci, each pair combining representatives from the GPI+ and GPI− clades. Finally, amino acid positions within the transferrin binding domain [71] are conserved in all ESAG6/7, PAG and their T. congolense orthologs (see Fam15 CSP page). These results suggest that an orthologous TFR is present in T. brucei and T. congolense but not T. vivax.
In addition to procyclin and VSG, T. brucei and T. congolense possess a third, highly abundant major surface glycoprotein expressed during the insect stage. These are known as Brucei Alanine-Rich Protein (BARP; [74]) and Glutamine Alanine-Rich Protein (GARP; [75]–[78]) respectively. Although GARP was initially thought analogous to procyclin in T. brucei [76], a procyclin ortholog was subsequently identified in T. congolense [78] and the CSP confirms a widespread procyclin family (Fam12). Structural affinities between GARP, which is expressed most strongly in epimastigotes [79], and BARP, which is epimastigote-specific [74], have been demonstrated [80], and the CSP confirms these two gene families as sister taxa (see Figure 3A). T. vivax contains 15 genes encoding BARP/GARP-like proteins that form three distinct subfamilies; each subfamily encodes proteins with distinctive repetitive domains towards the N-terminus that are absent in other species. Unfortunately, poor assembly in these regions prevents us from discerning their genomic organization, but at least some are arranged in tandem as in T. brucei and T. congolense.
BARP, GARP and their T. vivax homologs are part of a larger gene family (Fam50) in the CSP. We identified conserved sequence regions that unite these familiar families with other insect stage-specific genes and several hypothetical or uncharacterized genes (Figure 3B). The region at positions 262–274 contains a ubiquitous cysteine residue, followed four positions downstream by a VTxxSL motif in BARP and GARP, which is present with slight variations in all family members. A single-copy locus on chromosome 11 (i.e. Tb11.02.2370/TcIL3000.11.4860; marked ‘i’) is the sister clade to BARP/GARP and may be present in T. vivax also (TvY486_11149440). The Congolense Epimastigote-Specific Protein (CESP) gene family is expressed in T. congolense epimastigotes only, where it may have a role in adhesion to host surfaces [81]. Figure 3A shows that CESP has a sister clade in T. brucei comprising a tandem gene array on chromosome 8 (marked ‘ii); notably, these genes may be preferentially expressed in insect salivary glands [82], i.e. the location of T. brucei epimastigotes. An ortholog to CESP may be present in T. vivax (i.e. TvY486_0016400), although the position of this gene is not robust.
In addition to GARP and CESP, the CSP identified two related subfamilies encoding hypothetical proteins, (marked ‘iii’ and ‘iv’), which comprise subtelomeric tandem gene arrays. Analysis of stage-defined T. congolense EST (Figure 3C; [41]) proteomic analysis [83] found that subfamily ‘iii’ is preferentially expressed in T. congolense procyclic stage (see Figure 3C). Accordingly, the single-copy ortholog to subfamily ‘iii’ in T. brucei (Tb927.5.4020) is also preferentially expressed in procyclic cells based on transcriptome data [40], [84]; and a recent qRT-PCR analysis identified transcripts corresponding to Tb927.5.4020 in the insect midgut, although protein expression was not examined [82]. Subfamily ‘iv’ comprises sequences on chromosome 7 in T. brucei (i.e. Tb927.7.360) and a single-copy ortholog in T. congolense (i.e. TcIL3000.0.02370). In transcriptomic studies of T. brucei, expression data for these genes was weak and inconclusive [40]. However, qRT-PCR in various insect tissues suggests significant up-regulation of Tb927.7.360 (and paralogs) in the insect salivary gland and in metacyclic trypomastigotes [82]. Quantitative proteomic analysis in T. congolense indicated 13-fold higher expression of TcIL3000.0.02370 in epimastigotes over procyclics [83]. Hence, it seems likely that subfamily ‘iv’ genes are expressed during the insect-to-vertebrate transition.
To localize expression of a single gene copy of subfamily ‘iv’ in T. brucei, Tb927.7.380 was haemagglutinin (HA) epitope-tagged at the N-terminus (following the predicted signal peptide sequence) and expressed ectopically using a pDEX-577 inducible-expression vector. Protein expression was confirmed by Western blot (Figure 4A), and immuno-fluorescence microscopy indicates that Tb927.7.380 protein co-localizes with paraflagellar rod protein 2, consistent with specific expression at, or close to, the flagellar membrane (Figure 4B).
ESAGs have homologs outside of T. brucei [85]–[86], but these may only represent distant relationships within widely conserved protein families. With complete genome sequences for T. congolense and T. vivax we can now examine evidence for true orthology and therefore, the possibility that ESAG phylogenetic lineages predate T. brucei (Table 2). Orthologous lineages of TFR genes (i.e. ESAG6/7) are present in T. congolense and we have previously argued that ESAG2 belongs to a widespread lineage most closely related to b-type VSG in T. congolense [12]. Altogether, we find evidence that the T. congolense and T. vivax genomes contain homologous sequences to 9 of 12 ESAG families, while ESAG9 may have homologs in T. cruzi [87] (Table 2). Two trends emerge from phylogenies for each ESAG family shown in their individual CSP pages. First, ESAGs from multiple T. brucei strains are monophyletic and therefore, have a single origin; and second, with the exception of ESAG6/7, the sister clades to ESAGs are not orthologs in other species but chromosomal-internal genes in T. brucei. We interpret this as evidence for origins post-speciation, i.e. ESAGs are T. brucei-specific. Examining these closest relatives outside of the expression sites provides some indication of the origins of ESAGs, as demonstrated by Fam51, i.e. ESAG4 and the adenylate cyclases.
Trans-membrane adenylate cyclases are conserved across Trypanosomatids [88]–[89], and comprise a large gene family with diverse roles in T. brucei [85], [90]–[91]. ESAG4 is one lineage expressed specifically in the bloodstream stage, and instrumental in inhibiting host innate immunity [91]. The T. congolense and T. vivax genome sequences include 34 and 24 adenylate cyclase genes respectively. The adenylate cyclase phylogeny (Figure 5) shows that T. brucei and T. congolense lineages are paraphyletic, and in 10 cases T. brucei genes have orthologs in T. congolense that are positionally conserved. However, there are no orthologs of ESAG4 among T. congolense homologs. Indeed, the most closely related gene to ESAG4 is Tb11.01.8820, located at the subtelomeric boundary of chromosome 11. This gene has an ortholog in T. congolense (TcIL3000.11.16970), which is syntenic. Relative rates tests show that the substitution rate of ESAG4 has accelerated significantly compared with Tb11.01.8820 (p<0.0001; Table 2). Comparison of Tb11.01.8820 and ESAG4 sequences (Figure S2) shows that this remodelling has primarily affected the intracellular domains. 245 amino acid differences are distributed preferentially towards the C-terminal, with 69% occurring after the putative trans-membrane helix (a portion accounting for only 35% of total characters). Furthermore, of 54 sites where Tb11.01.8820 and TcIL3000.11.16970 are conserved, but ESAG4 is derived (i.e. unambiguous ESAG4 apomorphies), 41 occur in the intracellular domain. While the adenylate cyclase catalytic domain is intracellular, the evolution of ESAG4 has not altered the 8 residues identified as important for catalytic function [92]. Hence, ESAG4 represents a T. brucei-specific expansion of adenylate cyclase genes, most likely initiated through the transposition of a conserved locus to the ES, and coinciding with derivation of the protein structures associated with signal transduction within the cell but not catalysis.
The detail presented for other ESAGs in their CSP pages suggests that, like ESAG4 and Tb11.01.8820, ESAGs themselves are T. brucei-specific but descended from conserved genes, typically members of multi-copy families with subtelomeric distributions in several species. For example, ESAG2 and ESAG6/7 were, as previously noted, derived from VSG [12], [68]. ESAG3- and ESAG5-like loci are on T. vivax contigs containing telomeric repeats (GenBank accessions HE578915 and HE578917), but not VSG. ESAG8, although not surface-expressed, is most closely related to two leucine-rich repeat protein (LRRP) genes (i.e. Tb927.1.3670 and Tb927.3.580), that are chromosome-internal and include nuclear localization signal and RING motifs, which are diagnostic of ESAG8 [93]. While these two genes are T. brucei-specific, they are more closely related to conserved LRRP genes, suggesting that they may be progenitors of ESAG8. Finally, on the Fam3 CSP page, a structural comparison of ESAG11 and Invariant Surface Glycoprotein (ISG) sequences indicates that ESAG11 is homologous to ISG and so perhaps a highly modified derivative of these widespread surface proteins [94]. As Table 2 shows, only ESAG1 and ESAG12 appear to have no homology beyond T. brucei, suggesting that they have evolved de novo within the ES.
Besides ESAGs, the CSP contains various species-specific genes (Table S1). T. brucei-specific gene families include Fam4-7 encoding hypothetical proteins with predicted signal peptides but no similarity to known proteins. Fam4-7 genes are all adjacent to strand-switch regions and typically arrayed in tandem; transcriptomic studies suggest that they are expressed preferentially or solely in bloodstream forms [40], [84], [95]. The VSG-related (VR) genes previously identified in T. brucei [96] are also specific to T. brucei, although similar in structure to canonical VSG in T. congolense [12]. Finally, the CSP contains another family of VSG-like genes unique to T. brucei: Fam1.
In T. brucei 927, Fam1 comprises a polymorphic tandem array of 5 copies (0.1–3.2% nucleotide sequence divergence) at a strand-switch region on chromosome 6 (Figure 6A). Comparison with T. b. gambiense 972 indicates that Fam1 copy number may differ between strains because only a single gene (corresponding to the divergent 5′-most copy, Tb927.6.1310, Figure 6B) is present. The gene encodes a 347 amino acid protein with a predicted signal peptide and GPI-anchor. Fam1 genes are homologous to b-type VSG, but lack the typical C-terminal domain of canonical VSGs [12]. qRT-PCR analysis indicated that Tb927.6.1310 is predominantly expressed in bloodstream stages [12]. Enrichment of Tb927.6.1310 transcripts has been observed in metacyclic forms in the insect salivary gland [82], but this remains to be verified at the protein level.
We expressed the gene product of Tb927.6.1310 using a constitutive expression system (pXS5), and tagged at the N-terminus of the mature protein with an HA-9 epitope (Figure 6C). The HA epitope was placed two residues downstream of the predicted N-terminus of the mature protein following signal sequence processing. By Western analysis a single band was detected migrating at ∼45 kDa in whole cell lysates. However, the predicted molecular weight of the protein is ∼39 kDa, suggesting glycosylation at either or both predicted N-glycosylation sites. Cells were stained with a monoclonal antibody against HA and counterstained with FITC-concanavalin A. At 4°C, the fusion protein clearly colocalized with conA, conditions which block endocytosis and so retain the lectin exclusively within the flagellar pocket, a subdomain of the plasma membrane, and therefore demonstrating access to the cell surface. When cells were permeabilised with detergent, it was clear that Tb927.6.1310 protein was also present in additional internal compartments, and based on partial overlap of conA at 12°C (which retains conA in the flagellar pocket and early endosomes) and HA signals, these structures likely correspond to early and/or recycling endosomes. Hence, we conclude that the Tb927.6.1310 gene product is present at the parasite surface and may be restricted to the flagellar pocket, which is frequently observed for low abundance GPI-anchored proteins in this organism, and Tb927.6.1310 is also present within the endosomal apparatus.
In T. congolense, Fam22 is the most abundant species-specific gene family with >100 copies. Fam22 genes are distributed throughout putative subtelomeric regions and are typically situated immediately downstream of VSG. T. congolense VSG 3′UTR's are too short, (often only 15–30 bp; [41]) for Fam22 to fall within these regions. qRT-PCR analysis identified Fam22 sequences in all life stages except bloodstream forms (J. Donelson, unpublished data), but it is unclear whether Fam22 is a novel family of coding sequences or a non-coding, regulatory sequence. Nevertheless, Fam22 sequences are highly abundant. Trypanosoma vivax has substantially more species-specific gene families (19; Fam27-45) than either other species, which may be expected given that T. vivax is the natural outgroup to T. brucei and T. congolense. None have any significant similarity with known protein structures and more transcriptomic and proteomic surveys will be required to confirm that these sequence families genuinely encode T. vivax-specific proteins. However, many of these putative gene families are abundant (e.g. Fam31 and Fam34 have 38 and 34 members respectively) and transcripts corresponding to several gene families are among bloodstream-form RNA-seq data (Fam29-32, 34-35, 38-39; see Table S1).
The ancestor of T. brucei, T. congolense and T. vivax was very likely a hemoparasite of vertebrates, spread by Tsetse flies, and likewise fully exposed to the host immune response during its period in the mammalian host. Most familiar cell-surface features – both physiological regulators such as membrane transporters and disease effectors such as MSP and cathepsin – were already present in the ancestor. This is intuitive given that these features are typically present in T. cruzi. However, the CSP shows that the peculiar nature of the T. brucei cell surface, dominated by VSG [12], BARP/GARP-like genes and procyclin (Fam12) during various life-stages, also appears to have originated in the ancestral African trypanosome.
The role of the TFR on the ancestral cell-surface is more debatable. ESAG6/7 are thought to have evolved from a-VSG variant antigens [26], [73] but we show that the sister clade to ESAG6/7 are T. congolense Fam15 genes, which do not encode any known variant antigens [12]. Rather than originating from a-VSG in T. brucei, phylogenetic analysis of all VSG-like sequences [see Fam0 CSP pages] indicates that TFR-like sequences evolved from an a-VSG-like gene, (and further differentiated into ESAG6- and PAG-like genes), in the T. brucei/T. congolense ancestor, after separation from the lineage leading to T. vivax. While there are no TFR-like sequences in T. vivax, this does not preclude an analogous transferrin receptor in this species, since there is a large and structurally diverse a-VSG-like family (Fam23 [12]), the functional diversity of which is unknown. In short, we predict that Fam15 genes in T. congolense also encode a heterodimeric transferrin receptor, orthologous to the T. brucei TFR.
However, if the T. brucei/T. congolense ancestor possessed an orthologous heterodimeric TFR comprising GPI+ and GPI− monomers, we would expect GPI+ genes from T. brucei and T. congolense to be sister taxa reflecting their ancestry, and likewise for GPI−. Yet a literal interpretation of Figure 2 suggests separate expansions of Fam15 genes in each species, and thus independent origins of GPI+/− isoforms. Furthermore, branches separating ESAG6 and 7 (average genetic distance (p) = 0.114, n = 21) are much shorter than distances among the T. congolense genes (p = 0.604, n = 49), implying a recent origin for ESAG7 from ESAG6 through the deletion of its C-terminus. We consider this to reflect rapid turnover post-speciation of TFR-like genes that evolved in the ancestor, rather than independent origins, which is less parsimonious. Indeed, the same pattern of reciprocal monophyly between species is seen in other phylogenies (e.g. VSG, Fam50, Fam67), but it is clearly unparsimonious to suggest recent origins for these widely conserved families. Gene turnover replaces ancestral-type genes with more derived types post-speciation resulting in concerted evolution, a process exacerbated by recombination among tandem gene duplicates [97], and causing any signature of orthology to be ‘overwritten’ [98]. Such processes are known to affect ESAG6/7 routinely [20], [99] and frequent transposition of Fam15 genes between T. congolense subtelomeres is also apparent (data not shown). Given that this molecular evolution introduces phylogenetic artefacts, the Fam15 phylogeny need not refute the most parsimonious hypothesis that a TFR protein originated in the T. brucei/T. congolense ancestor.
While the essential character of the cell surface was established in the ancestral trypanosome, this common inheritance has been adapted subsequently. The evolution of ESAGs in T. brucei, uniquely linked to the telomeric VSG expression site, is a principal example of species-specific genomic adaptation. In some cases we can identify the likely origin of ESAG lineages among chromosome-internal loci; ESAGs 3, 4, 5 and 10 are derived from conserved loci that can be located precisely [85]–[86], [100]. ESAGs 2 and 6/7 are derived from variant antigen genes that evolved in the T. brucei/T. congolense ancestor [12]. ESAGs 8, 9 and 11 have more remote homology to conserved subtelomeric gene families, i.e. LRRP [101], MASP [87] and ISG (see Fam3 CSP page) respectively. This suggests a scenario in which genes with existing subtelomeric distributions (except ESAG10) and cell-surface roles (except ESAG8) were progressively compartmentalized into an independently-promoted telomeric locus, perhaps to provide a more precise regulatory environment.
Like the origin of Fam1 in T. brucei, the evolution of the ES demonstrates how novel cell-surface genes are repeatedly derived from existing major surface glycoproteins, whose abundance seems to provide a reservoir of raw material for neofunctionalization. Although ESAG functions are obscure, ESAG phylogenies suggest that they are distinct from those of conserved genes from which ESAGs evolved and indispensable on an evolutionary timescale. ESAGs from different T. brucei strains are monophyletic (except ESAG3), indicating no frequent transposition of sequences between ES and non-ES loci. ESAG-related genes at chromosome-internal loci are not observed in the ES and do not recombine with ESAGs, despite very frequent recombination among ES and non-ES copies respectively [20], [99], [102]. So although previous work has reported that ESAGs are not essential in the short term [101]–[102], the association between ESAG sequences sensu stricto and the telomeric ES has been preserved by selection over the long term, suggesting that ESAG and ESAG-like functions are distinct and non-redundant.
The CSP emphasizes dramatic cases of gene gain such as ESAGs in T. brucei, but significant phenotypic differences, such as life cycle variation, could be due to relatively subtle differences in conserved gene families such as Fam50. Given that BARP, GARP and CESP are preferentially expressed in the epimastigote stage [74], [79], [81] and that transcriptome data for both T. congolense and T. brucei indicate that subfamilies ‘iii’ and ‘iv’ are associated with insect mid-gut and salivary gland stages respectively [82], we suggest that Fam50 ranks alongside procyclin and VSG as a major surface glycoprotein, specifically related to the insect-to-vertebrate transition in multiple species. This is especially interesting because of the developmental variation among African trypanosomes during this transition. Unlike T. brucei and T. congolense, T. vivax remains within the insect mouthparts after feeding; this could reflect the basal-branching position of T. vivax in the species phylogeny (i.e. T. vivax is plesiomorphic and never evolved a mid-gut stage) or secondary loss (i.e. a mid-gut stage is the ancestral state). T. vivax also has a relatively small Fam50 repertoire, lacking orthologs to three clades: BARP/GARP and subfamilies ‘iii’ and ‘iv’. These genes might have evolved in the T. brucei/T. congolense ancestor if T. vivax is plesiomorphic, in which case all T. vivax genes should branch towards the root. Yet two of five Fam50 lineages in T. vivax, (i.e. TvY486_0016400 and TvY486_1114940), are nested among the would-be T. brucei/T. congolense gains. Reconciliation of this topology with the species tree indicates that if functionality is absent in T. vivax, this is due to secondary loss, rather than T. brucei/T. congolense gain.
Having systematically analyzed protein coding sequences for species differences, it is particularly important to remember that the cell-surface architecture comprises much more than the proteins encoded by the genes in the CSP and that non-proteinaceous elements, not least the surrounding glycocalyx composed of the carbohydrate moieties attached to membrane glycoproteins and glycolipids, might be equally important in determining phenotypic variation. Experimental studies of the cell-surface demonstrate that non-protein glycoconjugates could play an equal role in regulating host-parasite interactions, for example, a protease-resistant surface molecule (PRS) is known to dominate the surface of procyclic-stage T. conglolense [79]. T. brucei expresses various glycoconjugates on their surfaces that only become apparent in null mutants that cannot express the major surface glycoprotein [103]–[104]. Even considering the protein component, low abundance genes not considered in the CSP may still perform a vital role; for example, the haptoglobin-hemoglobin receptor (Tb927.6.440; [105]) responsible for resistance to trypanolytic factor by T. brucei is single-copy.
The essential character of genes expressed on African trypanosomes cell-surfaces was largely established in the common ancestor. Subsequently, prominent families have experienced rapid turnover of phylogenetic diversity, indicating both functional dynamism and redundancy. As we distinguish the functions of family members, we should be mindful of where orthology is absent and where it is retained; the latter, for example among MSP subtypes, cathepsin-L and B, or ESAG6-like and PAG-like TFR genes, is a strong indication of long-term functional differentiation and non-redundancy among paralogs. Truly species-specific genes represent adaptations of this shared inheritance and, in T. brucei, include almost all ESAGs as well as various GPI-anchored glycoproteins associated with strand-switch regions (Fam4-7). We anticipate that with improved genome assembly, species-specific genes, perhaps analogous to ESAGs, will be revealed in T. congolense and T. vivax also. To this extent, comparative genomics has met its objectives and the challenge now is to define how these unique genes and variants influence phenotypic differences in biology and disease.
|
10.1371/journal.ppat.1000203 | Cryoelectron Tomography of HIV-1 Envelope Spikes: Further Evidence for Tripod-Like Legs | A detailed understanding of the morphology of the HIV-1 envelope (Env) spike is key to understanding viral pathogenesis and for informed vaccine design. We have previously presented a cryoelectron microscopic tomogram (cryoET) of the Env spikes on SIV virions. Several structural features were noted in the gp120 head and gp41 stalk regions. Perhaps most notable was the presence of three splayed legs projecting obliquely from the base of the spike head toward the viral membrane. Subsequently, a second 3D image of SIV spikes, also obtained by cryoET, was published by another group which featured a compact vertical stalk. We now report the cryoET analysis of HIV-1 virion-associated Env spikes using enhanced analytical cryoET procedures. More than 2,000 Env spike volumes were initially selected, aligned, and sorted into structural classes using algorithms that compensate for the “missing wedge” and do not impose any symmetry. The results show varying morphologies between structural classes: some classes showed trimers in the head domains; nearly all showed two or three legs, though unambiguous three-fold symmetry was not observed either in the heads or the legs. Subsequently, clearer evidence of trimeric head domains and three splayed legs emerged when head and leg volumes were independently aligned and classified. These data show that HIV-1, like SIV, also displays the tripod-like leg configuration, and, unexpectedly, shows considerable gp41 leg flexibility/heteromorphology. The tripod-like model for gp41 is consistent with, and helps explain, many of the unique biophysical and immunological features of this region.
| The envelope (Env) spikes on the surface of HIV-1 and SIV virions facilitate target cell tropism, binding, and entry, and serve as the sole targets of humoral (antibody-mediated) immunity. X-ray crystallography has previously revealed the atomic structures of key core domains and peptides of the gp120 and gp41 Env spike subunits, but the manner by which these components are arranged in the Env spike is still speculative. Cryoelectron tomography (cryoET) affords a view of the entire Env spike in the context of the intact virion. We have previously published a cryoET model of the SIV Env spike which showed a unique tripod-like leg configuration for the solvent-exposed (external) gp41 stalk region. This model is consistent with, and helps explain, many of the unique biophysical and immunological features of this region. Subsequently another group using similar technology and virions reported a spike model displaying a compact gp41 stalk inconsistent with our splayed-leg spike model. In this report, we apply enhanced analytical cryoET procedures to show that HIV-1 also displays the tripod-like leg configuration, and shows considerable gp41 leg flexibility/heteromorphology. These results have implications for the design of effective vaccines targeting this region and may provide new insights into Env spike function.
| HIV-1 and the closely related SIV envelope (Env) spikes are composed of a trimer of heterodimers [1]–[6]. The base of the Env spike is comprised of three gp41 subunits, each of which possesses, from N-terminal to C-terminal, a fusion peptide, N-terminal heptad repeat, disulfide loop, C-terminal heptad repeat, membrane proximal external region (MPER), transmembrane domain, and cytoplasmic tail (CT). The relative positions of these various elements in the mature untriggered spike are largely unknown [7].
In contrast to gp41, the configuration of gp120 is better defined structurally. The CD4-liganded core structure consists of three subregions, the inner domain, the outer domain and the bridging sheet [8],[9]. The atomic structure of the unliganded SIV core has recently been described [10]. For both atomic structures, some of the more flexible elements, including V loops, N and C-terminal peptides and much of the glycan shield, were either deleted from the crystallization construct or were not resolvable due to flexibility [8]–[10].
The inherent flexibility of the V loops is a well recognized characteristic of HIV gp120 and has been suggested to be an important component of the viral defense against humoral immunity. Similarly, the CD4 binding site (CD4bs) components display flexibility, limiting the ability of most potential anti-CD4bs Abs to effectively bind, a process known as entropic masking [11].
Electron microscopy (EM) is an important adjunct to atomic structural studies and has the potential to allow the placement of the atomic structures of gp120 and gp41 core fragments and peptides, as well as the unresolved flexible components, into the global structural context of the Env spikes in situ [12]. Early work by Gelderblom and others showed virions covered with varying numbers of spikes [13]–[15]. A substantial fraction of purified spikes from HIV-1 and SIV were shown to display 3-fold symmetry though other forms were observed [1],[2]. By negative stain electron tomography, clear evidence for 3-fold symmetry was observed for a mutant form of SIV exhibiting Env with a truncated cytoplasmic tail [6]. The picture for HIV-1 is less clear in that the presumptive Env spikes appeared to display structural heterogeneity ([6] and unpublished data). Biochemical evidence for structural heterogeneity has also been published [1],[16],[17]. Because of the potential for morphological artifacts resulting from the use of the negative staining EM technique, including the attachment to a carbon substrate, pH changes, and drying, definitive analyses of the spike architecture could not be performed by this method [6].
More recently, we have further investigated the overall configuration of the SIV Env spike using cryoEM tomography (cryoET) wherein samples are preserved in a frozen hydrated state, free of the potential staining and drying artifacts common to negative staining [18]. Advantage was taken of the high level of Env spike incorporation and expression on the short-tailed mutants (SIVmac239/251 tail CEMx 174) (∼70 spikes/virion vs. ∼7–10 for wtSIV and wtHIV-1) thus aiding in data collection for the cryoET studies. HIV-1 variants with comparably high levels of Env spike expression are not available. The results from the SIV mutant show an Env spike in which each protomer of the presumptive trimeric gp120 head displays several morphological features and the solvent-accessible portion of gp41 forms a tripod-like set of legs. We were able to provide a tentative fit for the unliganded SIV gp120 core atomic structure [10] within the cryoET density volume and suggested that regions of unoccupied volume represented the masses of the V1/V2 and V3 loops missing from the atomic structure.
In an effort to determine the degree of structural heterogeneity within the spike population, the individual spike volumes were subjected to classification analysis in which the spikes were sorted into groups according to structural similarity. The results showed that most of the spikes were similar in form [18].
Subsequently, Zanetti et al. published a cryoET study showing an SIV Env spike average which differed from Zhu et al. in several important aspects [19]. For example, each gp120 subunit consisted of a simple globular mass adequate in volume to accommodate the atomic model of the core structure [10] but not large enough for the considerable mass of the V1/V2 and V3 loops. They also reported that the gp41 ectodomain formed a compact stalk rather than the flared tripod configuration that we observed. These differences are not easily reconciled since both groups took advantage of similar SIV short-tailed mutant virus.
In this report we have extend our cryoET analysis of the Env spike structure to include native (unmutated) Env spikes on wtHIV-1 and have now applied enhanced data collection and analysis techniques to generate 3D models. The data reveal that, as with our model of the short tailed SIV Env spike mutant, the wtHIV-1 displays tripod-like gp41 “legs”, at least in a significant percentage of the spikes. However, application of new approaches to search for structurally distinct morphological variations (i.e., a new classification algorithm) within the data suggests considerable conformational variability which likely reflects a more flexible structure than previously described.
The highly purified virus (HIV-1 BaL / SUPT1-CCR5 CL.30, lot p3955) used in this study was produced and provided by the AIDS Vaccine Program, SAIC Frederick, Inc., NCI, Frederick, MD. The production and purification procedures were as previously described [20]. The samples were treated with 2,2′-dithiodipyridine (Aldrithiol-2, AT-2), a process that eliminates viral infectivity while preserving Env structure and function [20],[21].
Fifteen µl of AT-2-treated viruses (∼2.8 mg/ml total protein) were added to 120 µl of PBS and pelleted at 25 psi for 15 min in an Airfuge centrifuge (Beckman Coulter) equipped with an A100/30 rotor. The pellets were resuspended in 10 µl of PBS of which 3.5 µl was placed on a 300 mesh R2/1 Quantifoil grid (Quantifoil, Jena, Germany) for 1 min. Excess virus and buffer was blotted with filter paper. The grid was then rapidly vitrified by plunging into liquid ethane in a liquid nitrogen bath using a homemade plunging apparatus.
The EM grids were transferred to a Gatan 626 cryoholder (Gatan, Pleasanton, CA) and examined under low dose conditions on a Philips (FEI, Eindhoven, Netherlands) CM300-FEG microscope operated at 300 kV. Single axis tilt series were recorded at 43,200× magnification using a Tietz Tem-Cam F224 slow scan CCD camera (2,048×2,048 pixels, Tietz Video, Gauting, Germany) and associated EM-MENU software. The pixel size at the specimen level is 5.56Å. Each tilt series consisted of 70–80 images recorded over an angular range of ±60° to ±70° at increments chosen according to the cosine rule [22]. The electron dose was estimated at 1–2 e−/Å2 per image.
In the absence of crystallized Env trimers, the analysis of intact virion-expressed Env spikes by cryoET may represent the best approach to determining important structural features of Env on the native virus. However, the use of cryoET for determining macromolecular conformations in situ is a rapidly evolving technique that has yet to reach its full potential and for which there is no generally agreed upon set of procedures. Our results and those of Zanetti et al. [19] have provided a first approximation of the overall structure of SIV Env. It was therefore of some concern that both attempts produced somewhat different results. It has been suggested [26] that the use, by the two groups, of a structure with imposed rotationally symmetry as an early reference for aligning the raw spike subvolumes and the subsequent application of 3-fold symmetry to the final image average may have generated or overly enhanced symmetry in the previously published SIV spike reconstructions [18],[19] It has also been suggested that our choice to ignore the consequences of a missing wedge of information resulting from the inherent inability to capture images of the virions over a full 180° tilt range may have led to incomplete data collection and artifactual distortion of the spike model. The latter point is probably not as great an issue as implied since data collected separately from the spikes projecting from the sides of the virions and from spikes projecting from the tops and bottoms of the virions gave similar results; any missing wedge effects would have been expected to differentially distort the images along different axes in the two sets [18].
To address these concerns, we have now used a missing wedge compensated multi-reference alignment and averaging scheme [Winkler et al., J. Struct. Bio., in press] to analyze wtHIV-1 Env spikes which addresses the problems of reference bias and incomplete data. In addition, we have investigated the potential advantages of applying sorting and classification schemes not only to the entire spike volume but also, independently, to the key head and leg subregions of the spike volumes. This latter approach would be expected to generate an improved average if molecular motions of the heads and the legs are uncoupled to a significant degree.
Our initial selection of 2,874 Env spike volumes, derived from 181 wt HIV-1 virions was subsequently reduced to 2,070 through programmatic elimination of lower quality selected volumes during the automated phase of the classification process. Subsequently, the spikes were sorted into eight classes based on multivariate data analysis and the class members in each were aligned and averaged. The surface distribution of the selected spikes in the entire set as well as in each of the eight classes suggests that the sampling was random and that the classification scheme did not result in the biased clustering of positionally discrete subpopulations (e.g., top/bottom- or side-arrayed) as might be expected with inadequate missing wedge compensation (Figure 1). Figure 3A shows cross-sections through the broadest portion of the head (H), the leg region (L) just above the membrane (see boxed insert for illustration), and a side view (S) of a section parallel to the axial plane with the membrane at the bottom of each of the eight classes. Although heteromorphic, each class average displayed overall dimensions similar to each other (∼12 nm high, ∼11 nm wide) and to our previously published SIV spike model (13.7 nm high, 10.5 nm wide) [18], thus increasing our confidence that the selected viral surface molecules represent bona fide Env spikes. Four of the classes showed a tendency toward 3-fold symmetry (judged visually), at least in the head region (Figure 3A, H5–8) while the other four showed asymmetry, though none were circular or shapeless. A comparative analysis of a cross-section through the base of each spike just above the membrane showed heterogeneity in structure incompatible with the single gp41 stalk model [19]. Most classes show two (Figure 3A, L3, 6, 8) or three densities (Figure 3A, L1, 2, 5, 7), with some showing additional diffuse signals. None, however, show three truly distinct leg densities.
Subvolume classification schemes can give a sense of the range of structure variation in the subvolumes selected for comparison. The real strength of the approach is that it facilitates a determination as to whether subsets (classes) of objects comprise the larger population. Relatively rigid, structurally distinct subpopulations, if present and significantly different from one another, should sort into distinct classes when subjected to multivariate data analysis. Even an inherently flexible structure displaying more random movement over significant distances would be expected to sort into separate classes although the result would tend to be blurred compared to the result expected in the case of structurally discrete states. Conformational variability within individual Env subunits (i.e., gp120 protomers) would not be detectable at the current resolution limits of this technique. Evidence of inter-subunit conformational variability should be most easily discerned by examining the spikes down the polar axis, i.e., screening for symmetry by viewing from above or below.
Even greater modes of flexibility might be manifested as twisting and/or rocking motion of gp120 and gp41 with respect to one another (Figure 2A). Because the alignment algorithms assess voxel density (signal intensity), the entire mass of the spike contributes to the placement of any given spike density within the average. Because the gp120 head is of greater mass than the solvent exposed regions of gp41, i.e., the MPER, the head would be expected to have a proportionally greater influence on the average. As a consequence, if the gp41 legs are in a tripod-like configuration but some were twisted, for example, with respect to the gp120 heads, the gp41 signals would be blurred to yield a cone of weak density rather than discrete, dense tripod-like legs in the averaged models. Similarly, the gp41 and membrane density might also impede alignment of the gp120 heads should the head and legs be twisted or bent with respect to one another. This is, in fact, what we observed for the MPER when the various classes, representing all of the classified Env spikes, were averaged together (data not shown).
To test this hypothesis, we reasoned that if the gp41 leg/membrane regions and gp120 head regions were independently aligned and/or classified, the weighting effect of one region upon the other would be eliminated and structural subregion classes with more discrete structural characteristics might emerge. Figure 2B–2D depicts the various head and leg subvolume alignment and classification schemes used to address this issue. For simplicity, the membrane is not depicted in the diagram. Alignments and, independently, classifications were based on the density data within either the whole spike, head, or leg/membrane regions.
Overall, the data show that subunit alignment and classification enhanced the symmetry of the targeted subregion and conversely, had the predictable effect of blurring the detail of the subregion excluded from the classification. For example, when the gp120 head was excluded, allowing the leg densities to drive the classification (Figure 3C), the gp41 leg region showed 3 of 8 classes (Figure 3C, L4, 6, 8) (representing 38% of total spikes) with three more-or-less distinct densities (judged subjectively) with a tendency toward 3-fold symmetry, a pattern that was less obvious following alignment and classification of the whole spike where only 1 of 8 classes (11% of total spikes) showed this pattern (Figure 3A, L1). Surprisingly, when both alignment and classification were driven by the leg densities, one class showed clear 3-fold leg symmetry (Figure 3E, L6, 12%) though all the others showed multi-leg asymmetry. In some classes where two legs are obvious, one of the legs appears extra thick (Figure 3C, L1, 2, 3; 3E, L 4, 7) possibly indicating a (transient?) association of two of the three legs.
When the gp41 leg region was excluded from classification, 5 of 8 classes (representing 63% of total spikes) had gp120 head regions that displayed a tendency toward 3-fold symmetry (Figure 3B, H1, 3, 5, 6, 8) compared to 4 of 8 classes (51% of total spikes) for whole spike alignment and classification (Figure 3A, H5–8). This trend was more pronounced when both alignment and classification were performed using just the head volumes (Figure 3D) wherein 6 of 8 classes (75%) were trimer-like (Figure 3D 3–8). Thus, even though there was clearly a tendency toward symmetry within the spikes, it was less evident following whole spike averaging, even when whole spike classification (into 8 classes) was applied. Interestingly, subvolume classification appeared about as effective in enhancing applicable head or tail images irrespective of whether the alignment was based on the whole spike or the targeted subvolume of the spike (compare Figure 3B to 3D and 3C to 3E). Stated another way, the use of subvolume alignment appeared less important to the final outcome than the application of subvolume classification. The reason for the less dominant effect in enhancing substructures could be attributed to the fact that selective alignment based on head or leg regions was carried out as a refinement of the already aligned whole spikes. Thus, the applied incremental changes do not appear as significant as the structural differences obtained by classification.
A significant percentage of the classified spikes and spike components deviated considerably from 3-fold symmetry. The reasons for this are unclear but include bona fide segmental flexibility/heteromorphology, and “noisy” data, a general characteristic of cryoEM data where contrast is inherently low. It is worth noting that in neither SIV nor HIV-1 did we observe any evidence of conserved structural features immediately below the membrane as would have been expected if the CT of Env were rigid or associated with a geometrically arrayed submembrane matrix layer.
In order to construct a single volume rendering of an “idealized” HIV spike, we selected the spike subregion classes showing the most symmetric features and averaged them together as single classes. These averaged classes were then displayed as surface renderings as illustrated in Figure 4. For example, the head-aligned/head-classified classes represented by Figure 3D (3–8) were averaged as a single class (Figure 4C) and, in the final step, 3-fold symmetrized (Figure 4D). The best leg-aligned/leg-classified classes (from Figure 3E, 3, 6, 7) were similarly combined, aligned, and averaged (Figure 4E and F). The optimized heads from the first set of models (Figure 4C and 4D) were then grafted onto the legs of the second set of models (Figure 4E and 4F) to yield the composite HIV-1 spikes shown in Figure 4G and 4H. To determine the correct rotational orientation of the legs with respect to the head, we averaged the most symmetric classes from the whole unmasked classification scheme and measured the rotational orientation of the legs with respect to the head (data not shown). Figure 4I and 4J represent transverse digital sections through the unsymmetrized and symmetrized chimeric models in Figure 4G and 4H, respectively. The ‘a’, ‘b’ and ‘c’ designations represent the head, midsection, and membrane-proximal leg sections, respectively.
A comparison of the HIV-1 composite model (Figure 4G and 4H) to our previously published SIV model [18] shows protein masses comparable to the main and lateral lobes, a less well defined peak, but no mass corresponding to the proximal lobe (Figure 4C). Unlike the Zanetti et al. model, we find no discernable cavity at the head-leg interface [19]. Our HIV-1 model appears to have three splayed legs though they are less well defined compared to our SIV model [18]. The radii of the legs in those HIV-1 classes where three discrete legs were visible were comparable to that previously reported for SIV (∼4.8 nm) [18]. No legs were seen in the Zanetti et al. model [19].
Evidence for Env spike structural heterogeneity also comes from a reanalysis of the Zhu et al. SIV data [Winkler et al., J. Struct. Bio., in press]. In contrast to what was originally reported [18], heteromorphology in the spike appearance and apparent flexibility are also seen in that data when subjected to the same general alignment and classification scheme reported here. Although, not subjected to independent targeted head and leg classification and reassembly, trimeric structures in the SIV head and the splayed leg conformations were also seen. Thus, the methods used here have generated similar results on two independent data sets. We now feel that both the Zanetti et al. [19] and Zhu et al. 2006 [18] models were unduly influenced by the reference that was selected in the earliest cycle and that this reference accentuated certain features at the expense of others. Consequently, the details of the respective density maps and fitting of the atomic core structures in those reports should be considered as provisional.
There is a general view that one of the main reasons no suitable Env-based vaccine for the induction of effective humoral protection has been developed relates to the difficulty in engineering soluble versions that faithfully mimic the viral spike surface configuration (reviewed in [27],[28]). Early monomeric constructs largely failed due to the exposure of immunodominant epitopes on the non-neutralizing face, a region believed buried in the gp120 subunit interface in the oligomer. Consequently, numerous attempts have been made at generating trimeric soluble constructs. However, many such constructs have proven inherently unstable with unacceptably high levels of subunit dissociation and/or aggregation. Strategies to circumvent this obstacle include mutational disruption of the protease cleavage sites between gp120 and gp41, inter-subunit disulfide bonding and other stabilization enhancing point mutations, and the addition of trimerization motifs (reviewed in [27],[28]). These approaches have had varying degrees of success at stabilizing the trimer and occluding the non-neutralizing face but have yet to faithfully mimic the antigenic profile of authentic membrane-associated trimers.
Our initial observation that the MPER of Env gp41 appeared to be in an open tripod-like conformation rather than in the traditionally-depicted compact stalk configuration provided a plausible explanation for the failure of at least some engineered version of Env trimers to adopt the native configuration [18]. Specifically, the trimerization motifs used to date, which bunched the C termini of the MPER tightly together, might force the MPERs into an unnatural configuration, perhaps altering and/or weakening the already inherently unstable interactions between the rest of the tripartite subunits. The resultant constructs may thus behave more like three tethered monomers rather than true trimers. On the other hand, if, as our data indicates, the Env spike is, to a degree, polymorphic and/or displays considerable component flexibility, a fully rigidified recombinant Env spike may not mimic the structure of virion-associated Env either. However, it may well turn out that rigid constructs, even if they don't fully mimic natural virus-associated Env might nevertheless serve as more effective vaccines. The production of a crystal structure of an Env trimer in its (near) native configuration would significantly advance our understanding of these and other issues relating to Env, however the prospects of success are diminished if the variations in form we observe are the result of segmental flexibility.
Several reasons for moving cautiously in fully embracing the tripod-legged paradigm have been put forth [12],[26]. First, the original modeled cryoET Env spike was derived from 3D tomograms of SIV rather than HIV-1. Although the structure of the Env spikes on the two AIDS viruses have been assumed to be structurally similar and the atomic structures of the gp120 of the unliganded HIV-1 and liganded SIV core proteins have been extensively compared [10],[29],[30], true similarity at the atomic level has yet to be formally demonstrated. Indeed, our previous negative stain EM tomogram studies have found HIV-1 spikes to be less uniformly configured than those on SIV ([6] and unreported data). Other data indicate that the degree of compactness and subunit accessibility to ligand binding varies significantly between SIV and HIV-1 and even between different strains of each [11], [28], [31]–[33]. Second, the cryoET-modeled SIV Env spike derives from a mutated version of SIV displaying a truncated CT. While this feature enhanced the expression of Env spikes on virions, thus facilitating data collection, it could be argued that the loss of a considerable segment of CT might well influence Env spike structure, especially in the most closely associated MPER [34]–[36]. This concern is somewhat ameliorated by data demonstrating that the Env spikes on these mutants are sufficiently functional so as to support efficient viral fusion and host cell infection [37]. Third, as described above, it has been argued that our previous data collection and processing schemes might have skewed the data and thus the model. Fourth and finally, Zanetti et al. [19], analyzed a short-tailed SIV virion nearly identical to those used by us yet they generated a rather different Env spike average model. Differences were observed both in the head (more compact in Zanetti et al.) and the presumptive gp41 solvent exposed region (compact vertical stalk in Zanetti et al.). Some of the potential reasons and technical issues relating to these differences have been discussed elsewhere [12],[26].
The data reported here support one of the key findings of our previous spike model in that we again find evidence of tripod morphology in the MPER. More importantly, this feature is now extended to include Env spikes from non-mutated wtHIV-1. To allay concerns about artificially enhanced symmetry, no symmetric references were utilized nor was enforcement of three-fold symmetry applied in the alignment or classification schemes used to generate the eight classes. Yet evidence pointing to a tendency toward 3-fold symmetry both in the gp120 head region and in the gp41 MPER emerged, at least in some of the class averages. Even in those classes without three leg masses, typically two masses are present as is diffuse additional density, a pattern more consistent with a three flexible leg model than a compact stalk model. Only in the final surface rendered models of averaged selected classes was symmetry enforced (Figure 4B, 4D, 4F, 4H).
The accumulating evidence regarding the biophysical features of both the MPER and the neutralizing MAbs that target this region are consistent with it having extensive membrane association [38]–[44] (see [45] and [46] for a comprehensive reviews of the MPER). This region may also be fairly flexible. For example, the segment encompassed by the 4F10, Z13e1 and 2F5 epitopes may transition between alpha-helical and alternative motifs to allow exposure of key residues that would otherwise be on opposite sides of the presumed alpha helical structure of this region. Such a transition would be required for effective binding of these MAbs in a membrane-associated environment [43],[45]. Recent high resolution NMR evidence suggested that the HIV-1 4E10 targeted epitope of the MPER is initially largely buried in the lipid bilayer and may be partially extracted upon 4E10 binding [42]. The conformational change associated with this interaction is facilitated by a flexible hinge-like region within the epitope. Such inherent flexibility may well contribute to our observed heteromorphology in the leg region and is consistent with the spread tripod-like leg orientation in a significant fraction of the Env spikes.
During the manuscript review process, Liu et al. 2008 [47] published a cryoET model of the HIV-1 spike with features that differed from the both the Zhu et al. [18] and Zanetti et al. [19] SIV spike as well as those reported here in several respects. Liu et al. report a compact stalk for gp41 and a Z-axis-elongated structure for gp120 in which the monomeric subunits make minimal contact with each other. Their unliganded gp120 structure could not readily accommodate the unliganded core structure of Chen et al. [10] but was fitted instead with the CD4 liganded core structure [8],[9]. After evidence of symmetry became apparent in the early rounds of alignment, symmetry was imposed on subsequent rounds and spike densities not fitting this pattern were discarded. We suspect that this model, like those of Zhu et al [18] and Zanetti et al. [19], may be unduly influenced by reference bias and imposed symmetry.
|
10.1371/journal.pgen.1002034 | A Tradeoff Drives the Evolution of Reduced Metal Resistance in
Natural Populations of Yeast | Various types of genetic modification and selective forces have been implicated
in the process of adaptation to novel or adverse environments. However, the
underlying molecular mechanisms are not well understood in most natural
populations. Here we report that a set of yeast strains collected from Evolution
Canyon (EC), Israel, exhibit an extremely high tolerance to the heavy metal
cadmium. We found that cadmium resistance is primarily caused by an enhanced
function of a metal efflux pump, PCA1. Molecular analyses
demonstrate that this enhancement can be largely attributed to mutations in the
promoter sequence, while mutations in the coding region have a minor effect.
Reconstruction experiments show that three single nucleotide substitutions in
the PCA1 promoter quantitatively increase its activity and thus
enhance the cells' cadmium resistance. Comparison among different yeast
species shows that the critical nucleotides found in EC strains are conserved
and functionally important for cadmium resistance in other species, suggesting
that they represent an ancestral type. However, these nucleotides had diverged
in most Saccharomyces cerevisiae populations, which gave cells
growth advantages under conditions where cadmium is low or absent. Our results
provide a rare example of a selective sweep in yeast populations driven by a
tradeoff in metal resistance.
| Understanding the genetic and molecular bases of adaptive mutations allows us to
gain insight into how new biological functions evolve. In natural populations,
examples in which adaptive mutations are characterized at the molecular level
are still rare. We studied wild yeast strains isolated from Evolution Canyon
(EC), Israel, that exhibit an extremely high tolerance to the heavy metal
cadmium. We found that high cadmium resistance was mainly caused by DNA sequence
changes in the promoter of a metal transport gene, PCA1.These
mutations increase PCA1 gene expression, thus leading to a more
efficient cadmium pump-out. Comparison among different yeast species shows that
the critical nucleotides found in EC strains are conserved and functionally
important for cadmium resistance in other species, suggesting that they
represent an ancestral type. When the PCA1 sequence and the
cadmium resistance in different S. cerevisiae populations
collected globally were compared, we found that most populations carried weak
PCA1 alleles and had a low cadmium tolerance. Since cells
carrying the strong PCA1 allele grow slowly under low-cadmium
conditions, it is likely that the tradeoff between cadmium resistance and growth
rate drives the evolution of reduced cadmium tolerance in most S.
cerevisiae populations.
| Unicellular microorganisms are often challenged by fluctuating environmental
conditions. Especially for those organisms having limited mobility, adaptation to
such environmental stresses is critical for survival of their populations. However,
mutations beneficial for survival in one environment may impose a cost under other
conditions [1],
[2]. Cells
need to fine-tune the evolved gene function or regulation in order to maintain an
optimal physiology under a range of conditions. It is important to understand how
cells adapt to novel or adverse environments. Such information may not only allow us
to dissect the factors affecting evolution of organisms, but may also provide us
some insights into pathway or functional network flexibility (or evolvability) of
the cell. To address this issue, identifying the mutations responsible for the
adaptive phenotypes is the most direct approach, and yet it remains challenging even
in simple organisms such as E. coli. Moreover, even if a mutation
is identified, detailed population and phylogeny data are required in order to
deduce the evolutionary trajectory of adaptive traits.
Experimental evolution represents a simplified approach since it allows scientists to
follow the evolutionary history of populations exposed to known selective pressures.
Several adaptive mutations in microorganisms have been discovered and characterized
at the molecular level from laboratory experimental evolution, adding greatly to our
understanding of adaptive evolution [1], [3]–[9]. On the other hand, studies
related to natural adaptation are more complicated. Although the mechanistic basis
or phylogeny of adaptive traits have been revealed in several previous studies [10]–[15], systematic
approaches dealing with both aspects are still rare [16]–[18].
Metal ions such as copper, iron, zinc, potassium and sodium are essential nutrients
involved in a broad range of biological processes [19], [20]. These essential metals
function as catalysts for biochemical reactions, stabilizers of protein structures
or cell walls, or regulators of intracellular osmotic balance. Despite their
importance, unbalanced metal concentrations can cause deleterious effects, sometimes
leading to programmed cell death [21], and thus represent a double-edged sword. It is important
for cells to tightly regulate homeostasis of these metal ions.
In natural environments, cells often encounter other nonessential metal ions. Some of
them such as cadmium, lead and arsenic are highly toxic to cells. Toxicity often
occurs through the displacement of essential metals from their native binding sites
or through ligand interactions, resulting in altered structural conformations or
interference with biochemical reactions [22]. These metal ions can induce
the generation of reactive oxygen species and cause damages to various cellular
components [23]–[25]. Organisms have evolved several different mechanisms to
cope with metal induced stresses, including specific metal transporters, metal
sequestration proteins or compartments, and different detoxification enzymes [12], [22], [26], [27]. These various
systems often cooperate with each other to quickly respond to variations in
environmental metal concentrations, indicating the importance of metal ion balance
to cells.
In this study, we observed that a subset of diploid yeast strains collected from
different locations of the EC could tolerate a heavy metal, cadmium, to a level
unseen in most known yeast strains. We found that the cadmium-resistant phenotype is
primarily caused by regulatory changes in the PCA1 gene, which
encodes a P-type ATPase required for cadmium efflux [28], [29]. By performing functional assays
and phylogenetic analyses, we show that PCA1 has experienced
several rounds of selective adaptation during yeast evolution. More strikingly, we
observe that a weak PCA1 allele spread to most S.
cerevisiae populations, probably due to a tradeoff between metal
resistance and fitness under low cadmium conditions.
Evolution Canyon is an east-west-oriented canyon at Lower Nahal Oren, Israel. It
originated 3–5 million years ago and is believed to have experienced
minimal human disturbance [30], [31]. In contrast to other wild yeast, the strains
collected from EC are often polyploid and most of them are heterothallic [32],[33]. Previous
studies have revealed high allelic diversity among EC yeast strains [32], [33]. To
assess whether these strains also carry specific adaptive phenotypes, we
performed a panel of phenotypic assays including cell growth under several
stress conditions. Only diploid strains were included in this study since
triploid and tetraploid strains are less amenable to further genetic analyses.
The results showed that a subset of EC strains (EC9, 10, 35, 36, 39, 57, and 58)
was resistant to a very high concentration of cadmium (0.8 mM CdCl2),
while all other strains analyzed were unable to grow on plates containing 0.2 mM
CdCl2 (Figure S1A).
Because chromosomal rearrangement has been suggested to be involved in adaptive
evolution [17], [34],[35], we first examined the karyotype of 14 diploid EC
strains. Pulsed-field gel electrophoresis (PFGE) analysis revealed that these EC
strains comprised three major karyotypes, EC-C1, EC-C2 and EC-C3 (with some
minor deviations)(Figure S1B). Interestingly, all
cadmium-resistant strains belong to EC-C1, suggesting that the metal-resistant
phenotypes have already evolved before the EC-C1 populations split. Therefore,
we chose to use mainly one strain (EC9) from EC-C1 for subsequent genetic
analyses.
We performed a genetic analysis to assess how many genetic loci are involved in
the cadmium-resistant phenotype. A haploid Cd-resistant clone (EC9-8) was
crossed with two Cd-sensitive strains (EC13 from EC-C2 and lab strain S288C).
Both hybrid diploids were Cd resistant, indicating that the Cd-resistant
phenotype was dominant. The hybrid diploids were then induced to sporulate, and
their haploid segregants were examined for their cadmium tolerance. These
segregrants showed a 2∶2 (resistant to sensitive) segregation pattern,
indicating that the Cd-resistant phenotype was primarily controlled by a single
genetic locus.
To screen for the gene responsible for cadmium resistance, a genomic DNA library
constructed from EC9 genomic DNA was transformed into Cd-sensitive cells. All
Cd-resistant colonies carried plasmids containing PCA1.
Sequencing the PCA1 allele (PCA1-C1) of the
EC-C1 strains revealed many mutations in both promoter and protein-coding
regions as compared with the PCA1 sequences from other strains
(Table
S1). Next, we directly tested whether PCA1-C1 alone
is able to improve the cadmium tolerance of cells. Plasmids carrying
PCA1-C1 were transformed into three Cd-sensitive strains
(S288C, SK1, and YJM789) in which the PCA1 gene had been
deleted. As shown in Figure
1A, the different strains carrying PCA1-C1 all
exhibited a level of cadmium resistance close to the level of EC-C1. By
contrast, when the PCA1 alleles from Cd-sensitive strains were
transformed into an EC9 pca1Δ mutant, the transformants
remained cadmium sensitive (Figure
1B). Finally, we sequenced the PCA1 alleles of the
segregants obtained from the previous genetic analysis and confirmed that all
Cd-resistant segregants carry the PCA1-C1 allele.
A previous study by Adle and co-workers has shown that the PCA1-dependent cadmium
resistance is mainly a consequence of active cadmium export (efflux) [29]. To determine
whether cadmium efflux is higher in cells containing PCA1-C1, cells carrying
PCA1-C1 or PCA1-SK1 were pretreated with
cadmium, washed to remove extracellular cadmium, resuspended in fresh media, and
collected at different time points to measure the cellular cadmium content using
inductively coupled plasma-atomic emission spectroscopy (ICP-AES). The
PCA1-SK1 allele from the SK1 strain was chosen for
comparison because this allele is phylogenetically more related to
PCA1-C1 based on an analysis of the corresponding ORF
sequences (Figure S2D) and because it is a cadmium-sensitive allele lacking the
G970R mutation (which abolishes the activity of Pca1) present in other
laboratory strains. Indeed, cells carrying PCA1-C1 could reduce
the intracellular cadmium concentration more quickly than cells carrying
PCA1-SK1. These data indicate that cells containing
PCA1-C1 have a very efficient cadmium efflux (Figure 1C).
In order to understand how PCA1-C1 has evolved a high cadmium
resistance, chimeric proteins with regions from PCA1-C1 and
PCA1-SK1 were constructed and assayed for their ability to
complement cadmium sensitivity of the pca1Δ mutant. We
found that swapping the promoters drastically affected cadmium resistance (Figure 2A, compare C1 with H6
and SK1 with H3), whereas swapping the region between amino acids 207 and 1216
had a mild effect (C1 vs. H1 and H2 vs. H3). Four nonsynonymous mutations are
present in this coding region: N223, T358, T363, and G365. Previous studies have
identified a few domains important for the stability or function of Pca1 [29], [36]. However,
none of these four mutations are located within these functional domains.
To assess whether the different levels of cadmium resistance resulted from
differences in promoter strength, we fused the PCA1-C1 or
PCA1-SK1 promoters to a luciferase reporter and assayed the
luciferase activity of these constructs. The expression driven by the
PCA1-C1 promoter was about four-fold higher than that
driven by the PCA1-SK1 promoter in the absence and presence of
cadmium treatments, suggesting that mutations in the PCA1-C1
promoter increased the degree of cadmium resistance by increasing the
PCA1 gene expression without destroying its regulation
(Figure 2B). We also
performed quantitative PCR to determine the level of PCA1 mRNA
in EC9 and SK1 strains. The data were consistent with the results from the
luciferase reporter gene assay.
By comparing the promoter sequences (including 600 bp upstream of the initiation
codon) of PCA1-C1 and PCA1-SK1, we observed 18
single nucleotide polymorphisms (17 single nucleotide substitutions and one 1-bp
deletion) and one 10-bp insertion (Table S1A). Because altered
PCA1-C1 expression plays a key role in enhancing cadmium
resistance, we sought to understand how this gene evolved and the degree to
which changes in its promoter contribute to expression differences. To address
the latter issue, we fused chimeric promoters with regions from
PCA1-C1 and PCA1-SK1 promoters to a
luciferase reporter and assayed their expression levels. We found that only the
region immediately upstream of the initiation codon (−213 to −1)
contributed significantly to the enhanced gene expression (Figure 2B, p5-2). Only 6 of the 18 single
nucleotide polymorphisms are present in this region. To determine which
mutations led to the enhanced promoter activity, we introduced the
PCA1-C1 version of each of these sites into the
PCA1-SK1 promoter and assayed the reporter gene expression.
Only mutations in three nucleotides (PCA1-SK1 to
PCA1-C1: −97T > C, −148T > G, and
−159G > T) had obvious effects (Figure 2C; Table
S1A).
To confirm that the changed expression is important for the cadmium resistance,
we introduced the same mutations into the PCA1-SK1 allele and
then measured the cadmium resistance of cells carrying these mutant alleles. The
expression level of the PCA1-SK1 mutants was indeed correlated
with the cadmium resistance (Figure
2D). When all three mutations were combined together into a single
mutant clone (PCA1-SK1+2/5/6), cells carrying this clone
were as resistant to cadmium as the cells carrying PCA1-H3, in
which the PCA1-C1 promoter is fused to the
PCA1-SK1 ORF. This result demonstrated that the enhanced
cadmium resistance of EC-C1 strains is mainly caused by three single nucleotide
substitutions in the promoter of PCA1-C1. Currently, it is
still unclear how PCA1 transcription is regulated. Although we
could not identify any transcription factor binding motif in the sequences where
the critical mutations (−97C, −148G and −159T) are located, it
is quite possible that these regions contain some of the regulatory elements of
PCA1.
To determine whether the Cd-resistant phenotype is specific to EC-C1 strains, we
examined the cadmium sensitivity of two closely related species, S.
paradoxus and S. mikatae. Two S.
mikatae strains and 28 S. paradoxus strains
isolated from different niches were tested [37]. Although both species could
not tolerate, unlike EC-C1, a high level of cadmium (2.0 mM CdCl2),
they exhibited much higher Cd resistance (0.4–0.8 mM CdCl2)
than the other S. cerevisiae strains (Figure S3).
To confirm that the medium level of cadmium resistance in S.
paradoxus was also mediated through PCA1, we
cloned the S. paradoxus PCA1 gene (Sp-PCA1)
into a plasmid, transformed the resulting vector into S. cerevisiae
pca1Δ mutant cells, and then assayed the transformants for
cadmium sensitivity. The result showed that Sp-PCA1 was able to
generate a medium level of cadmium resistance in S. cerevisiae
pca1Δ mutants (Figure S3B). Consistent with these results,
deletions of PCA1 in S. paradoxus strains
resulted in a Cd-sensitive phenotype (Figure S3C).
Interestingly, comparison between S. cerevisiae-, S.
paradoxus-, and S. mikatae-PCA1 promoter sequences
revealed that those critical residues (−97C, −148G and −159T)
identified in the previous experiment are conserved between EC-C1 strains,
S. paradoxus (28 strains) and S. mikatae
(2 strains)(Figure 3). When
we mutated these nucleotides of Sp-PCA1 (−100C,
−149G and −162T) to non-EC-C1 Sc-PCA1 sequences,
the mutant allele became cadmium sensitive (Figure
S3D), indicating that these nucleotides were also critical for the
function of Sp-PCA1. Since both S. paradoxus
and S. mikatae can tolerate a medium level of cadmium, it is
likely that this phenotype represents an original phenotype of the common
ancestor of S. cerevisiae, S. paradoxus, and
S. mikatae, which has been lost in most S.
cerevisiae populations (Figure S4).
If the common ancestor of S. cerevisiae and S.
paradoxus was cadmium resistant, why did most S.
cerevisiae populations become cadmium sensitive? By comparing the
promoter sequences of PCA1 from EC-C1, EC-C2, EC-C3, and 38
other S. cerevisiae strains (collected from various habitats on
different continents; see [37]), a striking pattern was revealed: most S.
cerevisiae strains carry a weak PCA1 promoter
similar to the one in the SK1 strain (Figure 4A). Hence, reduced promoter strength
accounts for the cadmium-sensitive phenotype observed in these strains (Figure S4).
The only exceptions are EC-C1, UWOPS87_2421, and UWOPS83_787_3. Interestingly,
the PCA1 promoters of UWOPS87_2421 and UWOPS83_787_3 also
contain the mutations critical for cadmium resistance (−97C and
−159T in UWOPS87_2421 and −97C in UWOPS83_787_3), and both strains
show cadmium-resistant phenotypes (Table S1).
A previous study showed that cells overexpressing PCA1 in a
medium without cadmium suffered reduced fitness [38]. To determine whether
expression of PCA1-C1 also imposes a high fitness cost on
cells, we conducted a competition assay to measure the fitness of cells
containing PCA1-C1, PCA1-SK1 or
PCA1-S288C. Plasmids carrying either
PCA1-C1, PCA1-SK1 or
PCA1-S288C were transformed into S288C
pca1Δ mutants. The resulting transformants were then
mixed with a reference strain carrying a green fluorescent protein-tagged Pgk1
protein and grown in a medium without cadmium. The results showed that cells
containing PCA1-C1 had a lower fitness than cells containing
PCA1-SK1 or PCA1-S288C
(p<0.001, two-tailed t-test), suggesting a
tradeoff between high Cd resistance and the fitness of cells under Cd-free
conditions (Figure 4B). It
is possible that Sc-PCA1 was selected to reduce the fitness
cost, thus resulting in lower Cd resistance if most S.
cerevisiae cells were constantly living in environments containing
low levels of cadmium. Alternatively, Sc-PCA1 might have been
selected, at a cost of reduced Cd resistance, to enhance other activities.
Previous studies have suggested that Pca1 is also involved in copper resistance
[38].
However, we found that cells carrying PCA1-C1 or
PCA1-SK1 showed a similar level of copper resistance,
indicating that the mutations in PCA1-C1 are specific to
cadmium resistance (Figure S5).
It is unclear why cells carrying PCA1-C1 have a lower fitness
under Cd-free conditions. Since Pca1 is not a highly abundant protein, the
fitness reduction is unlikely due to the energy cost for producing extra amounts
of the Pca1 protein. PCA1 belongs to a P-type ATPase family
whose members have been shown to transport metal ions such as cadmium, copper,
zinc, cobalt, and lead [39], [40]. Hence, the fitness cost of PCA1-C1
under non-cadmium conditions may result from a depletion of essential metal ions
caused by enhanced PCA1 expression. In S.
cerevisiae, it has been shown that Pca1 exports cadmium, not copper
[28].
However, a study by Adle et al. has also demonstrated that Pca1 affects copper
balance by chelating copper ions in a manner analogous to metallothionine [29]. Thus, it is
possible that the high expression of PCA1-C1 depletes copper or
other unidentified vital metal ions by metal sequestration or by metal
exportation.
In nature, S. paradoxus and S. cerevisiae were
found to occasionally coexist in the same ecological niches [41]. We
have shown that most S. cerevisiae strains have lost the
ancestral cadmium-resistant phenotype probably due to its fitness cost. Why do
S. paradoxus populations still maintain this phenotype?
Using the aforementioned competitive fitness assay, we found that S.
paradoxus cells carrying either wild-type or low-expression alleles
of Sp-PCA1 exhibit similar fitness under Cd-free conditions
(Figure
S6). These data suggest that S. paradoxus has
evolved other mechanisms to offset the fitness cost of high
PCA1 expression.
In EC-C1 cells, the increase in the expression of PCA1-C1 was
caused by three nucleotide substitutions in the promoter that were also shared
by the S. paradoxus and S. mikatae PCA1 genes.
Horizontal gene transfer between different species of yeast has been observed in
previous studies [42], [43]. One possible explanation for the high Cd resistance
of EC-C1 strains is that the ancestor of EC-C1 strains acquired a S.
paradoxus PCA1 allele through a horizontal gene transfer event, and
that the transferred Sp-PCA1 function was reinforced later on
by natural selection in EC-C1 strains. If that was the case, we would expect to
see that the sequence of PCA1-C1 is more similar to that of
Sp-PCA1 than to sequences of PCA1 alleles
from other S. cerevisiae strains. Phylogenetic analyses using
the PCA1 coding or promoter sequences, however, showed that the
distance between Sp-PCA1 and PCA1-C1 is
farther than that between Sp-PCA1 and other S.
cerevisiae PCA1 alleles, suggesting that PCA1-C1
was not derived from Sp-PCA1 (Figure 3 and Figure S7).
Moreover, we can rule out the possibility that gene conversion of a small region
between Sc-PCA1 and Sp-PCA1 has occurred in
the ancestor of EC-C1 strains. At alignment of PCA1 promoter
sequences, we found that even in the region containing the critical nucleotides
several nucleotides (−99, −112, −115, −119, −124,
and −137) were shared by all S. cerevisiae strains, but
did not exist in S. paradoxus strains (Figure 3); it should be noted, however, that
these sequences were also conserved in all 28 S. paradoxus
strains.
We have shown that, unlike S. paradoxus and S.
mikatae, most S. cerevisiae strains except for
EC-C1 and two other strains (UWOPS87_2421 and UWOPS83_787_3) are highly
sensitive to cadmium. Intriguingly, we found that sequences of the
PCA1 promoter in Cd-sensitive strains showed high identity
(Table
S1A). When population data were analyzed, a dramatic decrease in the
frequency of DNA polymorphisms was observed in this region, inconsistent with
the phylogenetic relationship observed in both upstream and downstream regions
(Figure 4A and Figure S2;
Tajima's D = −1.94, p<0.05).
This result suggests that the cadmium-sensitive phenotype did not evolve
independently in different strains. Instead, it was caused by a selective sweep
of a weak PCA1 promoter in S. cerevisiae
populations. A selective sweep of nonfunctional aquaporin alleles in S.
cerevisiae populations has been reported recently [44]. However,
unlike the aquaporin case, the selective sweep of S. cerevisiae
PCA1 is mainly caused by a single allele and covers a wider range
of populations. In addition, the PCA1 allele involved in the
sweep is still functional. In S. cerevisiae populations, only
the PCA1 alleles from S288C and W303 have lost the function
completely due to a mutation (R970G) in the catalytic domain [28], [29]. Expression
of PCA1-SK1 is upregulated when cells sense environmental
cadmium and deletion of PCA1 in other Cd-sensitive S.
cerevisiae strains makes mutant cells at least 20-fold more
sensitive to cadmium.
The reduced cadmium resistance in most S. cerevisiae strains is
probably a result of regulatory fine-tuning that allows cells to maintain a
certain level of cadmium efflux activity, without compromising their fitness,
under normal conditions. Such an ‘optimized’ PCA1
might explain why this weak allele spread so efficiently to different S.
cerevisae populations. On the other hand, the EC-C1 strains
maintained and even reinforced the ancestral Pca1 activity, probably due to
constant selection in their living environments. We measured the soil cadmium
concentrations at the collection sites of Evolution Canyon using inductively
coupled plasma-atomic emission spectroscopy (see Materials and Methods) and found that they ranged from 2.5 to 4.2
ppm, which is about 17–28-fold higher the median soil cadmium
concentration in Europe [45]. It is intriguing that some cadmium-sensitive strains
(EC-C2 and EC-C3) were isolated from the same areas as the EC-C1 strains. We
found that the PCA1 sequences of EC-C2 and EC-C3 strains are
more closely related to those in the European isolates than to those in EC-C1
strains (Figure 4A and Figure S2).
One possible explanation is that these cadmium-sensitive strains arrived in
Evolution Canyon more recently and have not yet adapted to the environment. An
in-depth study combining genomics and population distribution of the EC strains
will help us address this issue.
Phenotypic studies in budding yeast have suggested that resistance to various
metal ions in different yeast strains is quite diverse [16], [37], [46]. A recent genomic analysis
of both promoter and coding regions of three S. cerevisiae
strains also indicates that metal ion transporter genes are significantly
enriched in the gene group showing signatures of positive selection [47]. Our data with
PCA1 provide a clear example how a metal transporter gene
evolves after experiencing various types of selection that occurred at both
inter- and intra-species levels.
In the present study, we found that mutations in the regulatory and coding
regions both contribute to the adaptive phenotype. However, the mutations in the
regulatory region have a more profound effect as compared to those in the coding
region. Recent studies in a variety of organisms have suggested that regulatory
changes are critical for adaptive evolution [46], [48]–[53]. It is possible that the
promoter is more flexible to accommodate functional changes since it has less
structural constraints than the coding region [54]. Our results showed that,
by fine-tuning the PCA1 gene expression, Cd tolerance and cell
growth could be dramatically affected in natural yeast isolates, further
emphasizing the importance of regulatory changes in evolution.
EC diploid strains consist of S. cerevisiae collected from an
east-west-oriented canyon (Evolution Canyon) at Lower Nahal Oren, Israel [32], [55]. Our
strain numbers are the same as the numbers shown in Figure 2 of reference 26. In brief, EC3, 5,
23, 33, 34, 35 and 36 were isolated from the south-facing slope (SFS), EC7, 9,
10, 39, 40, 45 and 48 were isolated from the valley bottom (VB), and EC13, 14,
57, 58, 59, 60 and 63 were isolated from the north-facing slope (NFS). S
paradoxus, S. mikatae and other S.
cerevisiae strains were obtained from the collections of Dr. Duncan
Greig (University College London, UK) and Dr. Edward Louis (University of
Nottingham) [37]. Substitutive and integrative transformations were
carried out by the lithium acetate procedure [56]. Media, microbial, and genetic
techniques were as described [57].
A total amount of ∼2×108 yeast cells was used for plug
preparation. Cells were washed with 1 ml EDTA/Tris (50 mM EDTA, 10 mM Tris-HCl,
pH 7.5) and transferred into EDTA/Tris containing 0.13 mg/ml zymolyase
(Seikagaku America Inc., St. Petersburg, FL). The cell mixtures were incubated
for 30 s at 42°C and then embedded in low melting point agarose
(Sigma-Aldrich, St. Louis, MO). The agarose plugs were then incubated at
37°C overnight for zymolyase digestion. After digestion, the agarose plugs
were placed in LET solution (0.5 M EDTA, 10 mM Tris-HCl (pH 7.5), 2 mg/ml
protease K, and 1% N-lauroylsarcosine) at 50°C overnight. This step
was repeated three times. The plugs were transferred to EDTA/Tris solution and
dialyzed four times for 1 h at 37°C. Yeast chromosomes were separated in
0.7% agarose gels by pulsed-field gel electrophoresis (PFGE) using a
Rotaphor Type V apparatus (Biometra, Göttingen, Germany). Electrophoresis
was performed for 48 h at 13°C in 0.5× TBE buffer at a fixed voltage
of 120 V and an angle of 115°, with pulse time intervals of 30 s.
To construct an EC9 genomic DNA library, yeast genomic DNA was extracted using
the Qiagen Genomic-Tip 100/G kit (Qiagen, Valencia, CA), digested with
restriction enzymes, and ligated into a yeast vector pRS416 as described [57]. To screen
for Cd-resistant genes, a Cd-sensitive lab strain (S288C) was transformed with
the EC9 genomic DNA library and plated on cadmium-containing plates (0.4 mM
CdCl2). Plasmids isolated from the Cd-resistant colonies were
sequenced to identify the insert-containing genes.
To measure the intracellular cadmium concentration, log-phase cells carrying
different alleles of PCA1 were pretreated with 0.4 mM
CdCl2 for 1 h, washed with PBS containing 10 mM EDTA, resuspended
in rich medium without cadmium, and then collected at indicated time points.
Collected cells were immediately washed with PBS containing 10 mM EDTA. Total
intracellular cadmium levels were measured by inductively coupled plasma-atomic
emission spectroscopy (ICP-AES).
After cadmium treatment (0.1 mM CdCl2) for 2 h at 28°C, total RNA
was isolated from cells using the Qiagen RNeasy Midi Kit (Qiagen, Valencia, CA).
First-strand cDNA was synthesized for 2 h at 37°C using the High Capacity
cDNA Reverse Transcriptase Kit (Applied Biosystems, Foster City, CA). A 20-fold
dilution of the reaction products was then subjected to real-time quantitative
PCR using gene-specific primers, the SYBR Green PCR master mix, and an ABI-7000
sequence detection system (Applied Biosystems). Data were analyzed using the
built-in analysis program.
To construct the luciferase reporter plasmids, different promoter regions (up to
668 bp upstream of the start codon) of PCA1-C1 and
PCA1-SK1 were amplified by PCR. The luciferase coding
region (from Renilla reniforms) was also amplified by PCR. The
PCR fragments were co-transformed with pRS416 digested with XhoI and SacI into
the lab strain S288C. The genomic DNA of Ura+ colonies was isolated and
transformed into component E. coli cells. The plasmids from
ampicillin-resistant clones were isolated and sequenced. The constructed
luciferase reporter plasmids were transformed into an EC9
pca1Δ mutant.
Yeast cells carrying different luciferase reporter plasmids were treated with 0.1
mM of CdCl2 for 2 h. After the treatment, 0.5×107
cells were harvested for detection of the luciferase activity on a luminometer
(PE Victor3 luminometer plus autojector, Perkin Elmer, Waltham, MA). To the test
samples, 100 µl of 1 µM substrate (coelenterazine) was added.
Following a 5 second equilibration time, luminescence was measured with a 10
second integration time.
We measured the fitness of the testing strains by competing them against a
reference strain expressing PGK1::GFP in CSM-URA media at 28°C. The testing
cells and reference cells were inoculated in the CSM-URA medium individually and
acclimated for 24 h. The cells were subsequently diluted and refreshed in new
media for another 4 h. The reference and testing cells were then mixed at a
1∶1 ratio, diluted into fresh medium at a final cell concentration of
5×103 cells/ml, and allowed to compete for 21 h, which
represents about 12 generations of growth. The ratio of the two competitors was
quantified at the initial and final time points using a fluorescence activated
cell sorter (FACSCalibur, Becton Dickinson, Franklin Lakes, NJ). Five
independent replicates for each fitness measurement were performed.
The evolutionary history of the PCA1 ORF (3651 bps),
SUL1 ORF (2580 bps), PCA1 promoter (213
bps) and PCA1-SUL1 intergenic region (820 bps) was inferred
using the Neighbor-Joining method [58]. Sequences were obtained
from previously released data [37]. The percentage of replicate trees in which the
associated taxa were clustered together in the bootstrap test (500 replicates)
is shown next to the branches, analogous to a previous study [59]. The
tree was drawn to scale, with branch lengths in the same units as those of the
evolutionary distances used to infer the phylogenetic tree. The evolutionary
distances were computed using the Maximum Composite Likelihood method [60] and are
expressed as number of base substitutions per site. All positions containing
gaps and missing data were eliminated from the dataset (Complete deletion
option). Phylogenetic analyses were conducted in MEGA4 [61]. Tajima's D of the
PCA1 promoter (213 bp) was calculated by DnaSP V5 [62].
Soil samples were collected at 7 locations of Evolution Canyon corresponding to
the collection sites of the EC yeast strains (3 at the SFS, one at the VB, and 3
at the NFS). Soil cadmium levels were measured by inductively coupled
plasma-atomic emission spectroscopy (ICP-AES) using at least 200 g of individual
samples.
|
10.1371/journal.pbio.1001657 | A Cullin1-Based SCF E3 Ubiquitin Ligase Targets the InR/PI3K/TOR Pathway to Regulate Neuronal Pruning | Pruning that selectively eliminates unnecessary axons/dendrites is crucial for sculpting the nervous system during development. During Drosophila metamorphosis, dendrite arborization neurons, ddaCs, selectively prune their larval dendrites in response to the steroid hormone ecdysone, whereas mushroom body γ neurons specifically eliminate their axon branches within dorsal and medial lobes. However, it is unknown which E3 ligase directs these two modes of pruning. Here, we identified a conserved SCF E3 ubiquitin ligase that plays a critical role in pruning of both ddaC dendrites and mushroom body γ axons. The SCF E3 ligase consists of four core components Cullin1/Roc1a/SkpA/Slimb and promotes ddaC dendrite pruning downstream of EcR-B1 and Sox14, but independently of Mical. Moreover, we demonstrate that the Cullin1-based E3 ligase facilitates ddaC dendrite pruning primarily through inactivation of the InR/PI3K/TOR pathway. We show that the F-box protein Slimb forms a complex with Akt, an activator of the InR/PI3K/TOR pathway, and promotes Akt ubiquitination. Activation of the InR/PI3K/TOR pathway is sufficient to inhibit ddaC dendrite pruning. Thus, our findings provide a novel link between the E3 ligase and the InR/PI3K/TOR pathway during dendrite pruning.
| Neurons have the ability to engage in selective pruning that eliminates unnecessary axons/dendrites. This process is crucial for sculpting the nervous system during development. During Drosophila development, dendrite arborization sensory neurons (ddaCs) selectively prune their larval dendrites in response to the molting steroid hormone ecdysone, whereas mushroom body γ neurons eliminate their axon branches. However, the underlying molecular mechanisms for both of these modes of pruning were not well understood. Here, we conduct a genome-wide screen and identify a conserved E3 ubiquitin ligase that is critical for pruning both ddaC dendrites and mushroom body γ axons. This ligase complex has four core components—Cullin1, Roc1a, SkpA, and Slimb—that promote ddaC dendrite pruning in response to ecdysone. We show that this ligase facilitates ddaC dendrite pruning through regulation of the InR/PI3K/TOR pathway. The substrate-recognition protein Slimb promotes ubiquitination of Akt, an activator of the InR/PI3K/TOR pathway. Akt ubiquitination leads to its degradation and inactivation of the InR/PI3K/TOR pathway, which is required for dendritic pruning. Consistent with this, ddaC dendrite pruning is inhibited when the InR/PI3K/TOR pathway is activated. Thus, we identify a link between the Cullin1-based E3 ligase and the InR/PI3K/TOR pathway in regulating dendrite pruning. This work represents the first link between neuronal pruning and the insulin signaling pathway, raising interesting questions about how metabolic states may influence the control of such developmental processes.
| The selective removal of unnecessary or exuberant neuronal processes without loss of neurons, referred to as pruning, is a central theme in the maturation of the nervous system during animal development [1]. Pruning occurs widely in a variety of neurons of invertebrates [2],[3] and vertebrates [4]. In vertebrates, neurons normally extend exuberant branches to multiple targets, such as muscles or partner neurons, and prune away inappropriate or redundant branches to develop mature and functional connectivity [1],[5]. One well-characterized example is pervasive synaptic branch removal in the mammalian neuromuscular system at birth [6]. In invertebrates, such as holometabolous insects Manduca and Drosophila, the nervous systems are extensively remodeled via pruning and apoptosis during metamorphosis, a transition stage between larval and adult forms [2],[3]. In the Drosophila central nervous system (CNS), mushroom body (MB) γ neurons, serotonergic neurons, and thoracic ventral neurons prune their larval dendrites and/or axons to form the adult neuronal circuits [7]–[10]. In contrast, in the peripheral nervous system (PNS), a subset of dorsal dendrite arborization (dda) sensory neurons, such as Class I (ddaD/E) and class IV (ddaC) neurons, selectively remove their larval dendrite arbors with their axons intact and subsequently regrow their adult-specific dendrites [11],[12], whereas class II (ddaB) and class III (ddaA/F) neurons are eliminated via apoptosis during early metamorphosis [11]. Despite the wide occurrence and key roles of pruning in the maturing nervous systems, the molecular and cellular mechanisms underlying pruning remain poorly understood in both invertebrates and vertebrates.
Drosophila MB γ and ddaC neurons have been emerging as appealing systems for unraveling mechanisms of axon pruning and dendrite pruning, respectively. These pruning processes occur in a stereotyped but context-specific manner: ddaC neurons sever their major larval dendrites at the proximal regions, followed by rapid fragmentation and hemocyte-dependent debris clearance (Figure 1A) [11],[12], whereas MB γ neurons selectively prune their axon branches within the dorsal and medial lobes via local degeneration and glia-mediated engulfment (Figure 2A) [7],[13]–[15]. Pruning of the Drosophila nervous system is regulated by the steroid molting hormone 20-hydroxyecdysone (ecdysone) and its nuclear receptor heterodimer (EcR-B1/Usp) [16]. Ecdysone binds to a nuclear receptor heterodimer consisting of EcR-B1 and its co-receptor Ultraspiracle (Usp) to trigger the onset of pruning by activating the downstream pruning programs, rather than the conventional ecdysone response genes Broad complex, E74, and E75 [7]. The ecdysone-induced pruning programs are activated via a multilayered regulatory process. First, in response to a late larval pulse of ecdysone, EcR-B1 expression is upregulated, depending on TGF-β signaling [17],[18], the cohesin complex [19],[20], and the Ftz-F1/Hr39 nuclear receptors [21]. Second, EcR-B1, together with two epigenetic factors, Brahma and CREB-binding protein, activates the expression of their common target gene sox14 [22]. Third, Sox14, a key transcription factor, in turn induces the expression of the cytoskeletal regulator Mical [23]. Interestingly, despite its elevated expression in ddaC and MB γ neurons, Mical is only required for ddaC dendrite pruning but dispensable for MB axon pruning [23]. Aside from Mical, other molecules, such as caspases [24],[25], Ik2 and Kat60L [26], are also specifically required for ddaC dendrite pruning. Therefore, it remains poorly understood what common pruning machinery directs two distinct modes of pruning in ddaC and MB γ neurons.
The ubiquitin-proteasome system (UPS) is a potent cellular pathway regulating protein stability and homeostasis in eukaryotic cells [27]. Growing evidence indicates that the UPS pathway plays critical roles in neuronal development [28] and neuro-disorders [29]. Pharmacological inhibition of the UPS activity leads to apparent retardation of injury-induced axon degeneration [30]. In the UPS pathway, the polypeptide ubiquitin is activated by a ubiquitin-activating enzyme E1; the activated ubiquitin is transferred to a ubiquitin-conjugating enzyme E2 and then, via a ubiquitin ligase E3, to protein substrates, destined for 26S proteasome-mediated degradation. Substrate specificity is conferred by different classes of E3 ligases, one of which contains RING (really interesting new gene) domain. An important RING E3 ligase is the SCF (Skp1-Cullin-F-box) E3 ligase composed of the adaptor Skp1, the scaffold protein Cullin, the RING-domain protein Roc1-Rbx1-Hrt1, as well as the F-box protein for substrate recognition [31]. The conserved SCF E3 ligase complex containing the RPM-1/Highwire RING protein negatively regulates the MAP kinase signaling or the receptor tyrosine kinase ALK to control synapse formation and homeostasis in C. elegans and Drosophila [32]–[34]. Another worm SEL-10 SCF E3 ligase, whose activity is spatially restricted, directs selective synapse pruning, presumably by locally disrupting as-yet-unidentified substrates or pathways [35]. In Drosophila, inactivation of the UPS pathway by removing the proteasome subunits, the E1 activating enzyme Uba1, or expressing the yeast ubiquitin protease UBP2 blocks both MB γ axon pruning [13] and ddaC dendrite pruning [12]. Autoubiquitination and degradation of the caspase-antagonizing RING E3 ligase, Drosophila inhibitor of apoptosis protein 1 (DIAP1), allows for local activation of the Dronc caspase and thereby dendrite pruning of ddaC neurons [24]. However, DIAP1 (Wong and Yu, unpublished data) and Dronc [24] are not involved in axon pruning of MB γ neurons, raising the intriguing questions of whether and which E3 ligase complex directs two distinct modes of pruning in ddaC and MB γ neurons. Moreover, it is also of great interest to understand which downstream pathway is inactivated by the E3 ligase in order to facilitate the pruning process in ddaC neurons during early metamorphosis.
Here, we identified an E3 ubiquitin ligase complex that plays a crucial role in directing two distinct modes of pruning in ddaC and MB γ neurons during early metamorphosis. In a genome-wide RNA interference (RNAi) screen, we isolated Cullin1 (Cul1), a core scaffold protein of the SCF E3 ligase, which governs pruning of ddaC dendrites and MB axons. Additional components of the Cul1-based SCF E3 ligase include the RING domain protein Roc1a, the adaptor protein SkpA, and the F-box protein Slimb, all of which are essential for ddaC dendrite pruning and MB γ axon pruning. Cul1 acts together with Roc1a, SkpA, and Slimb to promote ddaC dendrite pruning downstream of EcR-B1 and Sox14, but independently of Mical. Furthermore, the Cul1-based SCF E3 ligase promotes ddaC dendrite pruning through negative regulation of the InR/PI3K/TOR signaling pathway but not its conventional target pathways, such as the Hedgehog (Hh) and Wingless (Wg) pathways [36]. We show that attenuation of the InR/PI3K/TOR pathway leads to strong suppression of dendrite pruning defects in mutant ddaC neurons deficient in the Cul1 SCF ligase, whereas activation of the InR/PI3K/TOR pathway alone is sufficient to inhibit ddaC dendrite pruning. Moreover, the F-box protein Slimb forms a complex with Akt, an activator of the InR/PI3K/TOR pathway, and promotes Akt ubiquitination. We demonstrate that the Cul1-based SCF E3 ligase complex facilitates ddaC dendrite pruning primarily through inactivation of the InR/PI3K/TOR pathway. Finally, the SCF ligase and the InR/PI3K/TOR pathway regulate dendrite pruning in ddaC neurons at least in part by promoting local caspase activation in the dendrites. Thus, this study provides the first example, to our knowledge, of a critical E3 ligase complex in directing two distinct modes of pruning and also provides mechanistic insights into how this evolutionarily conserved SCF E3 ligase attenuates a key signaling pathway to promote pruning during the maturation of the nervous system.
We previously reported high efficacy and specificity of inducible RNAi knockdown in ddaC neurons during the larval-pupal transition [23]. It prompted us to carry out an unbiased genome-wide RNAi screen searching for novel players of dendrite pruning (Wong and Yu, unpublished data). In this large-scale screen, we isolated two independent RNAi transgenes, v108558 (#1) and v42445 (#2), corresponding to cullin-1(cul1, also known as lin19) [37]. RNAi knockdown of Cul1, via one copy of the ppk-Gal4 driver that can target Gal4 expression in the class IV da neurons, led to dendrite pruning defects in ddaC neurons at 16 h after puparium formation (APF) (60%, n = 25 and 100%, n = 17, respectively; Figure S1A). The expression of cul1 RNAi transgene (#2) with two copies of ppk-Gal4 drivers caused severe defects with approximately 10.2 primary and secondary dendrites attached at 16 h APF (100%, n = 21; Figure 1C′ and 1G). Larval dendrites of cul1 RNAi ddaC neurons were eventually removed by 48 h APF (n = 13, unpublished data), probably due to extensive migration and death of the abdominal epidermis on which the dendrites arborize [11]. Likewise, knockdown with other cul1 RNAi lines, v33406 and v33407, also resulted in similar pruning defects at 16 h APF (unpublished data). In contrast, all dendrites were pruned completely in the wild-type ddaC neurons (n = 15; Figure 1B′ and 1G). The Cullin proteins serve as scaffold proteins of the SCF E3 ligase and interact with Roc1-Rbx1-Hrt1. We next generated transgenic flies expressing the dominant-negative form of Cul1 (Cul1DN) lacking its C-terminal putative Roc1-binding domain and neddylation site [38],[39]. The expression of Cul1DN also resulted in consistent dendrite pruning defects at 16 h APF with an average of 4.5 primary and secondary dendrites (n = 20, 100%; Figure 1D′ and 1G). Development of major dendrite branches appears not to be affected in cul1 RNAi and Cul1DN-expressing ddaC neurons, as judged by the number of their respective primary and secondary dendrites at the white prepupal (WP) stage (Figure 1C–D and 1G). Importantly, cul1 RNAi (#2) ddaC neurons under one copy of ppk-Gal4 exhibited normal larval dendrite morphology, as shown by the number of dendrite termini at the wandering 3rd instar (wL3) stage (Figure S2A), while these mutant neurons showed notable dendrite pruning defects at 16 h APF (Figure S1A). Further, we made use of the RU486-inducible Gene-Switch system [40] to drive the expression of Cul1DN from the early 3rd instar (eL3) stage when larval dendrite morphogenesis was largely completed. RU486 treatment did not affect WP dendrite morphology in Cul1DN-expressing ddaC neurons (Figure S2B) or dendrite pruning in the control ddaC neurons (n = 35; Figure S2B). However, we consistently observed dendrite pruning defects in 67% of Cul1DN-expressing ddaC neurons at 16 h APF (n = 30; Figure S2B), contrasting with the noninduced controls (0%, n = 12, Figure S2B). The Cul1DN pruning defect in the Gene-Switch system was not as strong as those induced by ppk-Gal4 driver (three copies) (Figure 1D′), due to weaker expression of the elav-GS-Gal4 driver (one copy). Thus, cul1 plays a specific role in dendrite pruning in ddaC neurons.
To further verify the role of cul1 in ddaC dendrite pruning, we generated homozygous MARCM clones [41] using the previously reported null allele, cul1EX [37]. All cul1EX ddaC clones exhibited severe dendrite pruning defects with 15.8 of primary and secondary dendrites attached by 16 h APF (n = 6, 100%; Figure 1E′; WP, Figure 1E and 1G), which were fully rescued by the expression of the Cul1 protein (n = 8; Figure 1F′ and 1G). Likewise, cul1EX MARCM ddaD/E neurons failed to prune their larval dendrites at 18 h APF (n = 9, 100%; Figure S1B), compared to the wild-type neurons (n = 5; Figure S1B). cul1EX ddaF clones survived by 16 h APF (n = 4), whereas the wild-type ddaF neurons were apoptotic (n = 5; Figure S1C). Neddylation of Cul1, in which the ubiquitin-like polypeptide Nedd8 covalently conjugates onto the conserved lysine residue (Lys718), is essential for its activity and function [38]. Consistently, all ddaC clones from the null allele nedd8AN015 exhibited dendrite pruning defects (n = 18; Figure S1D), further supporting the requirement of the Cul1-based E3 ligase activity for pruning of larval ddaC dendrites. Loss of nedd8 function also disrupted ddaD/E pruning (n = 3; Figure S1B) and ddaF apoptosis (n = 4, Figure S1C). Thus, Cul1 is required for pruning and apoptosis of sensory neurons during early metamorphosis.
Taken together, Cul1 is identified as the core component of the SCF E3 ligase that regulates pruning and apoptosis of sensory neurons during early metamorphosis.
We next attempted to identify other components of the Cul1-based SCF E3 ligase complex that are required for ddaC dendrite pruning. The Cullin proteins, via their C-terminal regions, bind to the RING-domain protein Roc1-Rbx1-Hrt1, which recruits an ubiquitin conjugating enzyme E2 [31]. In Drosophila, two closely related Roc1 proteins, Roc1a and Roc1b, were reported to play differential roles during development [42],[43]. We observed a strong dendrite pruning defect in roc1a mutants, but not in roc1b mutants, suggesting a specific requirement of Roc1a for pruning. MARCM analysis of roc1aG1, a null allele [42], revealed a severe cul1-like pruning defect in all ddaC neurons: about 10.3 primary and secondary dendrites were attached at 16 h APF (n = 9; Figure 3B′ and 3F). Furthermore, RNAi knockdown of Roc1a, via independent v106315 (#1) and v32398 (#2) lines, also resulted in pruning defects (n = 22 and n = 24, respectively; Figure S3). Loss of roc1a function also led to severe defects in ddaD/E pruning (n = 2; Figure S4A) and ddaF apoptosis (n = 6; Figure S4B). In contrast, roc1b null mutants showed no defects in dendrite pruning of ddaC (n = 9; Figure S5) and ddaD/E neurons (n = 8; unpublished data) or ddaF apoptosis (n = 8; unpublished data). Consistent with the preferential binding between Roc1b and Cul3 [43], MARCM ddaC clones for cul3gft2, a null cul3 allele, did not exhibit apparent pruning defects (n = 14; Figure S5). These results underscore the specific roles of Cul1 and Roc1a in governing dendrite pruning.
Thus, Roc1a, like Cul1, plays an essential role in dendrite pruning of ddaC sensory neurons.
The N-terminal region of the Cullin proteins interacts with the adaptor subunit Skp1, and via it with an F-box protein to recruit protein substrates in proximity to the E2 enzyme, thereby promoting substrate ubiquitination [31]. We first examined the potential involvement of SkpA, a Drosophila homologue of Skp1 [44], in ddaC dendrite pruning. Knockdown of SkpA with three independent RNAi lines, BL28974 (#1), v32790 (#2), and v107815 (#3), resulted in consistent dendrite pruning defects in most of ddaC neurons (Figure 3C′ and 3F, and unpublished data). The expression of v32790 (#2) with two copies of ppk-Gal4 driver resulted in moderate ddaC dendrite pruning defects with full penetrance: an average of 6.9 primary and secondary dendrites remained attached (n = 21, 100%; Figure 3D′ and 3F). Attenuation of skpA function also inhibited ddaD/E pruning (n = 13; Figure S4A) and ddaF apoptosis (n = 15; Figure S4B). No MARCM clones for skpA1, a skpA null allele, could be recovered in ddaC neurons, presumably due to its essential functions during cell division [44]. Hence, SkpA, like Cul1 and Roc1a, appears to govern ddaC dendrite pruning.
Given that the substrate specificity of the Cul1-based SCF E3 ligase complex is conferred by an F-box protein, we further performed a RNAi screen to examine the potential roles of 31 putative F-box proteins [45] in dendrite pruning. Among them, Supernumerary limbs (Slimb), which when knocked down via RNAi, exhibited dendrite pruning defects (Figure S3). Slimb, a Drosophila homologue of the mammalian β-TrCP proteins, acts to ubiquitinate Cubitus interruptus (Ci) and Armadillo, two key effectors of Hh and Wg signaling pathways, respectively [36],[45]. Knockdown of Slimb with two independent RNAi lines, v107825 (#1) and v34273 (#2), caused mild pruning defects (37.5%, n = 32 and 66.6%, n = 12, respectively; Figure S3). These phenotypes have been confirmed by generating MARCM ddaC clones of the null allele slimb8, as stronger pruning defects were observed; 4.4 primary and secondary dendrites were attached by 16 h APF (n = 16, 50%, Figure 3E′ and 3F). Loss of slimb function also inhibited ddaD/E pruning (n = 8, 75%; Figure S4A) and ddaF apoptosis (n = 6, 50%; Figure S4B). Compared to those of cul1, roc1a, and skpA, the phenotypes of slimb mutants are less severe, presumably due to the perdurance of the wild-type protein. Alternatively, we cannot exclude the possible existence of other F-box proteins involved in ddaC dendrite pruning and ddaF apoptosis.
Thus, SkpA and Slimb appear to play important roles in regulating ddaC/D/E dendrite pruning and ddaF apoptosis.
We next examined the potential requirements of Cul1, Roc1a, SkpA, and Slimb for MB axon pruning. In wild type, MB γ neurons selectively eliminated their larval axon branches by 24 h APF (n = 5; Figure 2A and 2B–B′) and regenerated the medial branches in the adulthood (n = 6; Figure 2G). Importantly, cul1EX mutant MB γ neurons retained many larval axons, which were labeled by 201Y-Gal4-driven mCD8GFP expression (n = 19, 100%; Figure 2C′). These unpruned larval axons co-labeled by FasII were located outside the major FasII-positive α/β lobes at 24 h APF (arrowheads in the right panels of Figure 2C′) and persisted in the adult brains (n = 5, 100%; Figure 2H). In addition to the pruning defect, we also observed a neuroblast-proliferation defect, as adult cul1 MB clones lacked late-born α/β neurons (Figure 2H). Moreover, the cul1EX axon pruning defects were fully rescued by reintroducing the Cul1 protein (Figure S6A). Thus, Cul1, a core component of the SCF E3 ligase, also plays a critical role in regulating axon pruning of MB γ neurons. The MB axon pruning and proliferation defects in cul1EX mutant resemble those in mutants of uba1, a single E1 gene in fly, as reported previously [13]. Interestingly, roc1aG1 MARCM analyses also revealed a notable axon pruning defect in MB γ neurons. Unpruned axon branches, positively labeled by 201Y-Gal4-driven mCD8GFP expression, remained at 24 h APF (n = 13, 62%; Figure 2D′). The axon pruning defects were rescued by expression of the full-length Roc1a protein (n = 11; Figure S6A). In contrast, roc1b exhibited normal pruning of MB γ axons at 24 h APF (Figure S6B). We were able to generate skpA1 MARCM clones in MB γ neurons, all of which exhibited strong axon pruning defects at 24 h APF (n = 8, 100%; Figure 2E′). Likewise, MB axon pruning was also inhibited at 24 h APF in slimb2 (78%, n = 14; Figure 2F′) and slimb8 (57%, n = 14; unpublished data) MARCM clones. Finally, dendrite pruning of MB γ neurons failed to occur in cul1EX (n = 9, 100%) or skpA1 (n = 4, 100%) MB neuroblast clones at 24 h APF (Figure S6C). Taken together, Cul1, Roc1a, SkpA, and Slimb govern both ddaC dendrite pruning and MB axon pruning, presumably as a Cul1-based SCF E3 ligase complex.
We next demonstrated the physical association among Cul1, Roc1a, SkpA, and Slimb. To this end, we first performed co-immunoprecipitation (co-IP) experiments in nontreated and ecdysone-treated S2 cells transfected with Myc-tagged Slimb. Endogenous unmodified Cul1 and Neddylated-Cul1 (Nedd8-Cul1 [37]), together with endogenous SkpA, were detected specifically in the immune complex when Myc-Slimb was immunoprecipitated from nontreated or ecdysone-treated protein extracts using an anti-Myc antibody (Figure 4A), suggesting that Cul1, SkpA, and Slimb form a protein complex, independently of ecdysone. Moreover, Roc1a was specifically co-immunoprecipitated using the protein extracts of S2 cells co-transfected with Flag-tagged Roc1a and Myc-Slimb (Figure 4B), also supported by a previous report that Roc1a preferentially associates with Cul1 in embryos [43]. We next performed coIP experiments using larval brain extracts expressing Myc-Slimb. Slimb was able to specifically co-immunoprecipitate with endogenous Cul1 and SkpA (Figure 4C), suggesting the in vivo association in postmitotic neurons.
To investigate whether the Cul1 SCF E3 ligase expression is upregulated during the larval-pupal transition, we utilized Laser Capture Microdissection (LCM) technique to microdissect MB γ neurons from wild-type, EcRDN, and sox14 brains, subject to total RNA extraction and quantitative real-time PCR experiments (Q-PCR). Importantly, mRNA levels of cul1, like mical, were significantly upregulated from eL3, WP to 6 h APF (Figure 4D). Upregulation of cul1 transcription at 6 h APF, like mical, was strongly inhibited by EcRDN expression or loss of sox14 function (50.4% and 48.9% reduction, respectively; Figure 4E), suggesting that cul1 upregulation is dependent on EcR-B1 and Sox14. Therefore, ecdysone signaling appears to regulate the abundance, but not assembly, of the Cul1 SCF E3 ligase complex. mRNA levels of the E1 gene uba1 were also increased upon pupal formation (Figure 4D), in agreement with a previous microarray analysis showing that the UPS genes including cul1 and uba1 were upregulated in MB γ neurons during the larval-pupal transition [46]. Moreover, knockdown of cul1 did not enhance the dendrite pruning defects in sox14 null mutant ddaC neurons (n = 41; Figure S8). We also made MARCM ddaC clones for sox14Δ13 and cul1EX double mutant and observed an average of 15.2 primary and secondary dendrites attached to the double mutant ddaC neurons (n = 5; Figure 5G′ and 5J), similar to either cul1EX (15.8; Figure 1E′ and 1G) or sox14Δ13 (14.5, n = 6; Figure 5F′ and 5J) null MARCM neurons, supporting a linear relationship between sox14 and cul1. Thus, cul1 appears to act downstream of sox14 during dendrite pruning.
Taken together, Cul1, Roc1a, SkpA, and Slimb function as the components of the SCF E3 ligase complex, among which the expression of cul1 is dependent on EcR-B1 and sox14 during dendrite pruning.
Given that the EcR-B1/Sox14/Mical pathway governs ddaC dendrite pruning [23], we next investigated how the SCF E3 ligase complex may integrate into this pathway. We first examined the protein levels of Mical, Sox14, and EcR-B1 in various SCF mutant ddaC neurons. Mical protein levels remained largely unchanged in the cul1 RNAi ddaC neurons (n = 12; Figure 5B and 5E) or cul1EX MARCM (n = 8; unpublished data) at the WP stage, compared to those in the wild-type neurons (n = 14; Figure 5A and 5E). Likewise, the Mical levels were not affected in roc1aG1 MARCM (n = 7; Figure 5C and 5E), skpA RNAi (n = 12; Figure 5D and 5E), as well as slimb8 MARCM (n = 13; unpublished data) ddaC neurons. Using a mical-lacZ reporter that drives upregulation of the LacZ expression under a mical enhancer (Y. Gu and F. Yu, unpublished data), we detected similar LacZ expression in wild-type, cul1 RNAi, and skpA RNAi WP ddaC neurons (Figure S7A). These data indicate that the Cul1 SCF E3 ligase is dispensable for regulation of Mical transcription/expression in ddaC neurons. Moreover, the expression levels of EcR-B1 and Sox14 were unchanged in WP cul1EX mutant ddaC neurons, compared with the wild-type controls (Figure S7B). Likewise, the rest of the SCF components are not important for the expression of EcR-B1 and Sox14, as their protein levels were not affected in roc1aG1, skpA RNAi, slimb8, or nedd8AN015 mutant ddaC neurons (Figure S7B). Thus, these data are consistent with the conclusion that the SCF complex functions downstream of EcR-B1 or sox14 during dendrite pruning.
Given that Mical expression is unaffected by the mutants of the SCF components, we hypothesized that the Cul1 SCF E3 complex might act in parallel with Mical during dendrite pruning. If this is true, we would expect enhancement of dendrite defects of mical null mutants by compromising the SCF components. Indeed, RNAi knockdown of Cul1 in mical mutant ddaC neurons resulted in a drastic pruning defect with the persistence of 10.1 primary and secondary dendrites at 16 h APF (n = 22; Figure S8), whereas the mical null ddaC neurons retained 5.3 major dendrites (n = 26; Figure S8). Likewise, knockdown of Roc1a (n = 36; Figure S8), SkpA, or Slimb (n = 22 and 30, respectively; unpublished data) significantly enhanced the mical null mutant phenotypes. Knockdown of cul1 or roc1a in mical mutants exhibited normal elaboration of major dendrites at WP stage (Figure S9). Moreover, MARCM clones of mical15256 and slimb8 double null mutant also exhibited a significant enhancement of dendrite pruning defects (Figure 5K). An average of 9.7 primary and secondary dendrites were attached to the double mutant ddaC neurons (n = 10; Figure 5I′ and 5K), compared to either mical15256 (3.3, n = 14; Figure 5H′ and 5K) or slimb8 (4.4, n = 16; Figure 3E′ and 3F) null MARCM neurons. Thus, these data suggest that the Cul1 SCF E3 ligase facilitates ddaC dendrite pruning in parallel to Mical.
Taken together, our data suggest that the Cul1-based SCF E3 ligase complex acts downstream of EcR-B1/Sox14 and governs dendrite pruning in parallel with Mical.
During tissue growth and pattern formation, Cul1, Roc1a, and Slimb negatively regulate Hh or Wg pathways by specifically degrading their respective effectors, Ci or Armadillo [45]. We hypothesized that if the Cul1-based SCF complex also attenuates the Hh and/or Wg pathways during ddaC pruning, the dendrite pruning defects associated with loss of the SCF ligase complex would be attributable to hyperactivation of either or both pathways. However, three lines of evidence indicate that the Cul1-based SCF complex acts independently of Hh and Wg pathways during dendrite pruning. First, inhibition of either pathway did not suppress cul1 RNAi dendrite pruning defects. Expression of the Hh repressors, CiCell (n = 29; Figure 6B and 6I) or the Patched receptor (unpublished data), did not suppress the pruning defects associated with cul1 RNAi, compared to the nonfunctional MicalN-ter control (n = 30; Figure 6A and 6I) that was unable to rescue the dendrite pruning defect in mical mutant ddaC neurons and its expression alone did not interfere with ddaC dendrite pruning [23]. Likewise, expression of the Wg inhibitors, SggS9A (n = 19; Figure 6C and 6I) or the truncated form of Dishevelled (unpublished data), did not suppress the cul1 RNAi effects. Second, the expression of the nondegradable Hh activator CiU or the Wg activator ArmS10 did not affect normal ddaC dendrite pruning (Figure S10). Finally, forced expression of the Hh activators, CiU (n = 16; Figure S11) or the truncated form of Smoothened (Δ661–818; unpublished data), did not enhance the Cul1DN pruning defects, compared to the MicalN-ter control (n = 25; Figure S11). Similarly, the expression of the Wg activator ArmS10 did not affect the pruning defects in the Cul1DN-expressing ddaC neurons (n = 22; Figure S11).
To further investigate which signaling pathway might be attenuated by the Cul1-based SCF complex during ddaC pruning, we performed a candidate-based screen to systematically examine other important signaling pathways including Notch, Insulin, JNK, JAK/STAT, Hippo, EGFR, PVR, and Dpp. We expressed the dominant-negative repressors of these pathways, such as NotchDN, InRDN, BskDN, DomeΔCYT, YorkieS168A, EGFRDN, PVRDN, and TkvDN, in cul1 RNAi ddaC neurons. From this screen, the Insulin pathway was identified as a potential target pathway that is negatively regulated by the Cul1 SCF E3 ligase complex. Notably, the expression of a dominant negative form of the Insulin Receptor (InRDN), via one copy of ppk-Gal4 driver, dramatically suppressed the dendrite pruning defects in the cul1 RNAi-expressing ddaC neurons (n = 43; Figure 6D). On average, only 1.2 primary and secondary dendrites remained attached to these InRDN, cul1 RNAi double ddaC neurons, contrasting with 7.8 major dendrites observed in the micalN-ter, cul1 RNAi control (n = 30; Figure 6A and Figure 6I). In contrast, the expression of NotchDN (n = 24; Figure 6E and 6I) or other repressors did not influence the cul1 RNAi effects on ddaC dendrite pruning (Figure S12A). Expression of either of these dominant-negative transgenes did not affect normal ddaC pruning (Figure S10, unpublished data). The numbers of primary and secondary dendrites were primarily unchanged at the WP stage in various double-mutant combinations, except an apparent reduction in the SggS9A, cul1 RNAi combination (Figure S12B).
To verify the specific effect of InR, we further conducted the genetic enhancement experiments in which InRCA or NotchCA, known to activate Insulin or Notch pathways, respectively, was expressed in cul1DN ddaC neurons. The expression of InRCA (n = 28; Figure S11) but not NotchCA (n = 18; Figure S11) resulted in a significant enhancement of the cul1DN-associated pruning defects with full penetrance. Moreover, the InRDN expression significantly mitigated either cul1DN (n = 22, Figure S13) or skpA RNAi (n = 15; Figure 6G and 6J) effects on dendrite pruning. Conversely, the expression of InRCA significantly enhanced the ddaC dendrite pruning defects caused by skpA RNAi knockdown (n = 19; Figure 6H and 6J). The numbers of primary and secondary dendrites remained similar at the WP stage in these double-mutant combinations, compared with single mutants (Figure S12B).
Collectively, these genetic suppression/enhancement results indicate that the Cul1-based SCF complex promotes ddaC dendrite pruning primarily through attenuation of the Insulin pathway.
InR functions through the PI3K/TOR signaling pathway to mediate protein translation, metabolism, and ribosome biogenesis [47]. To examine whether the PI3K/TOR pathway could also be inhibited by the Cul1-based SCF complex during pruning, we performed genetic suppression assays by inactivating the PI3K/TOR pathway in cul1 RNAi or cul1DN ddaC neurons. Importantly, the expression of the dominant-negative form of PI3K (PI3KDN) [48] or the Phosphatase and tensin homologue (PTEN) [49], both known to inactivate the PI3K pathway, like InRDN, drastically mitigated the dendrite pruning phenotypes in cul1 RNAi ddaC neurons (n = 32 and 39, respectively; Figure 7B, 7C, and 7K). Compared to 8.1 primary and secondary dendrites in the MicalN-ter control, PI3KDN or PTEN expression resulted in an average of 1.1 or 2.3 primary/secondary dendrites connected to the cul1 RNAi ddaC neurons (Figure 7K), respectively. Moreover, the expression of PI3KDN (n = 22), PTEN (n = 21), or InRDN (n = 22) largely rescued the cul1DN-mediated pruning defects (Figure S13). Thus, these data suggest that Cul1 also attenuates the PI3K pathway during ddaC pruning.
The PI3K pathway is interconnected with the Target of Rapamycin (TOR) pathway via TSC1 and TSC2, which act as negative regulators of the TOR pathway [47]. We next ascertained whether the TOR signaling pathway is also attenuated by the Cul1 SCF complex during ddaC dendrite pruning. Co-expression of TSC1 and TSC2, like PTEN, significantly suppressed the pruning defects in cul1 RNAi (n = 25; Figure 7E and 7K) or cul1DN (n = 30; Figure S13) ddaC neurons. The truncated protein TORTED lacking its toxic effector domain behaves as the dominant-negative form to disrupt the TOR signaling [50]. The TORTED expression also strongly suppressed the dendrite pruning defects in cul1 RNAi (n = 47; Figure 7F and 7K) or cul1DN (n = 24; Figure S13) ddaC neurons, respectively. The protein kinase TOR regulates protein synthesis via phosphorylation of the p70 ribosomal protein S6 kinase (S6K) [47] and the eukaryotic translation initiation factor 4E binding protein (4E-BP) [51]. S6KKQ, a catalytically inactive version, and 4E-BP(AA), a nonphosphorylated version, are able to repress protein translation and the TOR pathway [51],[52]. The expression of 4E-BP(AA) rescued the dendrite pruning defects in almost all cul1 RNAi (n = 55; Figure 7G and 7K) or Cul1DN-expressing (n = 32; Figure S13) ddaC neurons. Likewise, the expression of S6KKQ also dramatically mitigated the dendrite pruning defects caused by cul1 RNAi knockdown (n = 42, Figure 7H and 7K) or cul1DN expression (n = 18; Figure S13). Moreover, the expression of InRDN (n = 8; Figure 7I′) or 4E-BP(AA) (n = 8; Figure 7J′) also strongly suppressed the dendrite pruning defects in cul1EX MARCM ddaC neurons, as approximately 4.9 and 4.3 primary and secondary dendrites remained attached at 16 h APF, respectively, compared to 15.8 major dendrites remaining in cul1EX mutant neurons (Figure 7L). To further confirm these genetic suppressions, we attenuated the InR/PI3K/TOR pathway by feeding the 3rd instar larvae with Rapamycin, a pharmacological inhibitor of TOR. Rapamycin treatment did not affect the onset of puparium formation/adult eclosion, larval dendrite development (Figure S14A) or wild-type ddaC dendrite pruning (Figure S14B). However, Rapamycin treatment significantly suppressed the ddaC dendrite pruning defects in cul1 RNAi ddaC neurons (n = 58), but not in mical RNAi neurons (n = 48, Figure S14B). Thus, these data further support the conclusion that the InR/PI3K/TOR pathway is inactivated by the Cul1-based SCF complex during ddaC dendrite pruning.
The following lines of evidence indicate the specificity of these suppression effects. First, attenuation of the InR/PI3K/TOR pathway did not affect normal elaboration of major dendrites in cul1 RNAi or cul1DN mutant ddaC neurons, as the numbers of their primary and secondary WP dendrites were essentially unchanged despite the simplified terminal branches (Figure S15A and S15B). Inactivation of the InR/PI3K/TOR pathway alone did not affect normal ddaC dendrite pruning (Figure S10 and unpublished data). More importantly, the severing of major dendrites, a hallmark feature of dendrite pruning, occurred, similar to the wild type, in these suppression experiments (empty arrowheads, Figure S16A), suggesting that inactivation of the InR/PI3K/TOR pathway restores the severing of proximal dendrites from the cul1 RNAi ddaC neurons. Furthermore, the expression of InRDN, PI3KDN, PTEN, TORTED, S6KKQ, or 4E-BP(AA) was not able to suppress the dendrite pruning defects associated with mical RNAi (unpublished data) or mical15256 mutant ddaC neurons (Figure S16B), supporting their specific genetic interactions with the SCF components.
Taken together, the InR/PI3K/TOR pathway is specifically inhibited by the Cul1-based SCF E3 complex in order to promote ddaC dendrite pruning during early metamorphosis.
Since inactivation of the InR/PI3K/TOR pathway suppresses the dendrite pruning defects in ddaC neurons lacking the Cul1 E3 ligase activity, we next assessed whether compromised E3 ligase function causes hyperactivation of the InR/PI3K/TOR pathway. To this end, we examined the expression and activity of Akt, a positive regulator of the InR/PI3K/TOR pathway, in ddaC neurons and 6 h APF brain lysates. Using an anti-Akt antibody, a weak localization of endogenous Akt was detected in ddaC somas (Figure 8A), but not in dendrites and axons (unpublished data). Endogenous Akt was significantly upregulated in the somas of cul1 RNAi ddaC neurons at the WP stage (2.8 folds, n = 9; Figure 8B and 8D), compared to the wild-type somas (n = 8, Figure 8A and 8D). Akt signals were abolished in akt RNAi ddaC neurons (Figure 8C and 8D). Since overexpressed Akt could be observed weakly in axons and dendrites in addition to its robust localization in ddaC somas (Figure S17), we co-expressed Akt with cul1 RNAi or the control RNAi to examine the Akt protein levels throughout the neurons. Consistently, cul1 RNAi knockdown also caused a significant increase in Akt protein levels in ddaC somas (2 folds, n = 16), dendrites (2.2 folds, n = 12), and axons (1.9 folds, n = 11; Figure 8F–F″ and 8G), compared to the RNAi controls (Figure 8E–E″ and 8G). As a control, the expression levels of GFP were the same or similar throughout cul1 RNAi or the control RNAi ddaC neurons (unpublished data). These data suggest that SCF-dependent Akt degradation is not restricted to dendrites of ddaC neurons. Consistently, the Cul1 SCF E3 complex appears to be localized throughout ddaC neurons as indicated by uniform distribution of exogenously expressed SkpA-RFP in the dendrites, axons, and somas (n = 7; Figure S17). Moreover, the protein levels of Akt were significantly increased in the cul1 RNAi brain lysates (the upper panel, Figure 8H) where Cul1 proteins were knocked down via the cul1 RNAi line (#1) using a pan-neuronal driver elav-Gal4 (Figure S18A). Concomitantly, Akt activity was also substantially increased, as judged by an increase in active and phosphorylated Akt levels (the middle panel, Figure 8H). Thus, the Cul1-based SCF E3 ligase negatively regulates Akt expression and activity in ddaC neurons and prepupal brains. We then investigated whether attenuation of Akt suppresses the dendrite pruning defects in cul1 RNAi ddaC neurons. Interestingly, knockdown of Akt with two independent RNAi lines (#1: BL31701 or #2: BL33615) potently suppressed the dendrite pruning defects in cul1 RNAi ddaC neurons (n = 26, Figure 7D; unpublished data). Reduction of akt function resulted in an average of 0.2 primary/secondary dendrites connected to the cul1 RNAi ddaC neurons (Figure 7K), compared to 8.1 in the MicalN-ter control (Figure 7A and 7K). These biochemical and genetic data indicate that compromised SCF E3 ligase function causes hyperactivation of the InR/PI3K/TOR pathway.
To further examine a potential link between the Cul1 SCF E3 ligase and the InR/PI3K/TOR pathway, we assessed the physical interaction between Akt and the F-box protein Slimb. The C-terminal region of Slimb contains seven WD40 domains that are responsible for binding to its substrates and targeting them for ubiquitination. Interestingly, Akt was co-immunoprecipitated with Slimb using the protein extracts of S2 cells co-transfected with Myc-Akt and Flag-Slimb (Figure 8I). We confirmed this interaction in postmitotic neurons, as Slimb was specifically co-immunoprecipitated with Akt in the prepual brain extracts expressing Myc-Slimb and Akt (Figure 8J). Furthermore, Akt specifically interacted with Slimb, as Akt was not pulled down by either the truncated Slimb protein lacking its WD40 domains (SlimbΔWD40, Figure 8K) or another F-box protein Ago (Figure S18B). We then investigated whether Slimb can mediate ubiquitination of Akt. Notably, Slimb expression strongly increased the amount of polyubiquitinated Akt (lane 7, Figure 8L), compared to the control (lane 6, Figure 8L). In contrast, the expression of SlimbΔWD40 failed to facilitate Akt ubiquitination (lane 8, Figure 8L), suggesting that the WD40 domains are responsible for Slimb-mediated ubiquitination of Akt. Thus, Slimb associates with Akt and targets Akt for ubiquitination.
We have demonstrated that inactivation of the InR/PI3K/TOR pathway by the Cul1-based SCF E3 ligase complex is required to facilitate ddaC dendrite pruning during early metamorphosis. To further substantiate it, we examined whether constitutive activation of the InR/PI3K/TOR pathway alone is sufficient to inhibit normal progression of ddaC dendrite pruning. Notably, the expression of InRCA, or PI3KCA, both known to constitutively activate the InR/PI3K pathway [48],[53],[54], via two copies of ppk-Gal4 driver, caused consistent dendrite pruning defects in the vast majority of ddaC neurons (InRCA, n = 25, 92%; and PI3KCA, n = 23, 91%; Figure 9B′ and 9C′). On average, 5.2 (InRCA) and 4.3 (PI3KCA) primary/secondary dendrites retained the attachment to their respective ddaC neurons at 16 h APF (Figure 9I). The dendrite pruning defect caused by InRCA expression was fully suppressed by akt RNAi knockdown (n = 20; Figure S19D), suggesting that expression of InRCA likely activates residual Akt to inhibit dendrite pruning. Supportively, using two PTEN null/strong alleles, PTENC494 and PTEN1, we observed similar dendrite pruning defects in ddaC neurons (89%, n = 9 and 48%, n = 21, respectively; Figures 9D′ and S19B). Loss of PTEN function also inhibited ddaD/E dendrite pruning, but not ddaF apoptosis or MB axon pruning (Figure S19B and S19C, unpublished data). Moreover, the expression of the small GTPase Rheb (Ras homologue enriched in brain) or the constitutively active form of S6K (S6KSTDETE), activators of the TOR pathway [47],[52], led to dendrite pruning defects in the majority of ddaC neurons (77%, n = 25 and 63%, n = 24, respectively; Figure S19A). Activation of the InR/PI3K/TOR pathway appears not to affect the morphology and numbers of their major WP dendrites (Figures 9A–D and S19A). Further, we co-expressed InRCA, PI3KCA, Rheb, or S6KSTDETE with Cul1DN (Figure S20) or roc1a RNAi (unpublished data) in ddaC neurons, resulting in more severe dendrite pruning defects.
Similar to the Cul1 SCF E3 ligase, the InR/PI3K/TOR pathway regulates ddaC dendrite pruning in a Mical-independent manner. First, the protein levels of Mical, Sox14, or EcR-B1 were unaffected upon activation of the InR/PI3K/TOR pathway, via InRCA/PI3KCA expression or loss of PTEN function (Figure S21A–B). The upregulation of the Mical expression during the larval-pupal transition was also unaffected upon inactivation of the InR/PI3K/TOR pathway (unpublished data). Second, inactivation of the InR/PI3K/TOR pathway, via expression of InRDN, PI3KDN, PTEN, TORTED, S6KKQ, or 4E-BP(AA), did not suppress the dendrite pruning defects in mical null mutant ddaCs (Figure S16B). Finally, activation of the InR/PI3K/TOR pathway, via expression of InRCA (n = 27, Figure 9F), PI3K (n = 21, Figure 9G), or Rheb (n = 20, Figure 9H), significantly enhanced the mical null mutant pruning phenotypes (Figure 9J).
Local caspase activation in dendrites was shown to be required for elimination of dendrites in ddaC neurons [24],[25]. We therefore assessed whether SCF and InR/PI3K/TOR govern dendrite pruning through local caspase activation. Using the genetically encoded caspase reporter CD8::PARP::Venus [25], we observed no or negligible caspase activity in dendrites of cul1 RNAi (n = 9) or InRCA-expressing (n = 6) ddaC neurons at 6 h APF, in contrast to strong caspase activity in wild-type ddaC dendrites (n = 6; Figure S22). Therefore, the SCF ligase and the InR/PI3K/TOR pathway regulate dendrite-specific pruning in ddaC neurons at least in part by promoting local caspase activation in the dendrites.
In summary, our data indicate that activation of the InR/PI3K/TOR pathway alone is sufficient to inhibit ddaC dendrite pruning in a Mical-independent manner. Thus, we demonstrate that during early metamorphosis, the Cul1-based SCF E3 ligase complex facilitates ddaC dendrite pruning primarily through inactivation of the InR/PI3K/TOR pathway.
Previous studies showed that the UPS activity plays an intrinsic and essential role in governing both modes of pruning in ddaC [12] and MB γ neurons [13]. However, little is known about the E3 ubiquitin ligase that is able to direct two distinct modes of neuronal pruning in Drosophila. Moreover, it is also unknown which downstream pathway is inactivated by the E3 ligase in order for ddaC neurons to prune their dendrites. Here, we report the identification of the Cul1-based SCF E3 ligase complex that plays a critical role in both modes of pruning of ddaC and MB γ neurons. In a genome-wide RNAi screen, we first isolated Cul1, a core scaffold protein of the SCF E3 ligase, which is required for ddaC dendrite pruning and MB axon pruning during early metamorphosis. We further identified the other components of the E3 ligase complex including the RING domain protein Roc1a, the adaptor protein SkpA, as well as the F-box protein Slimb. These molecules, like Cul1, are all required for pruning of ddaC and MB γ neurons. We show that the Cul1-based SCF E3 ligase acts downstream of EcR-B1/Sox14 and promotes ddaC dendrite pruning in a Mical-independent manner. Moreover, via a candidate-based screen, we observed that during ddaC dendrite pruning, the Cul1 E3 ligase negatively regulates the InR/PI3K/TOR signaling pathway but not other major developmental pathways examined. We demonstrate that inactivation of the InR/PI3K/TOR pathway leads to strong suppression of dendrite pruning defects in ddaC neurons deficient in the Cul1 E3 ligase activity, whereas activation of the InR/PI3K/TOR pathway alone is sufficient to inhibit ddaC dendrite pruning. Thus, the Cul1-based SCF E3 ligase promotes dendrite pruning in ddaC neurons primarily through inactivation of the InR/PI3K/TOR pathway (Figure 9K).
Previous studies indicated that UPS activity is cell-autonomously required for both MB axon pruning and ddaC dendrite pruning in Drosophila [12],[13]. First, the expression of a yeast ubiquitin protease, which eliminates ubiquitin from substrates and inhibits UPS-mediated degradation [55], causes severe pruning defects in MB γ and ddaC neurons [12],[13]. Second, loss of the E1 enzyme Uba1 leads to strong pruning defects in both types of neurons. Third, removal of the proteasome subunits, Mov34 or Rpn6, also results in uba1-like pruning defects in MB and ddaC neurons. The selectivity of the UPS machinery for pruning is mainly conferred by a specific E3 ligase. Given the existence of a large number of E3 ligases in the Drosophila genome, it is challenging to identify the specific one that directs these two modes of neuronal pruning. It was reported that DIAP1, an RING E3 ligase, antagonizes the Dronc caspase activity and inhibits ddaC dendrite severing [24]. A model had been proposed that autoubiquitination and degradation of DIAP1 allows for the activation of the Dronc caspase in ddaC dendrites and thereby pruning of ddaC neurons. However, recent studies reported that all ddaC neurons sever their dendrites normally in DIAP1 gain-of-function mutants [26] and overexpression condition [25]. Nevertheless, Dronc caspase and DIAP1 appear not to be important for axon pruning of MB γ neurons, as analyses of Dronc MARCM clones [24], expression of the caspase inhibitor p35 [13], or gain of DIAP1 function (Wong and Yu, unpublished data) revealed no axon pruning defects. Thus, the question of which specific E3 ligase directs both modes of pruning remains open for a long time. In this study, we demonstrate that the Cul1-based SCF E3 ligase complex plays key roles in regulating both ddaC dendrite pruning and MB axon pruning during Drosophila metamorphosis.
Several lines of evidence indicate that the Cul1-based SCF E3 ligase complex is the specific one that governs both ddaC dendrite pruning and MB axon pruning. First, while Cul1 and its binding partner Roc1a are critical for pruning of ddaC and MB neurons, Cul3 and Roc1b that preferentially bind each other are not important. Second, the Cul4-based E3 ligase, which regulates TSC2 protein stability and TSC1/2 complex turnover in Drosophila [56], is dispensable for ddaC dendrite pruning. MARCM analyses of the null allele cul-411L exhibited no ddaC dendrite pruning defect (n = 5; Figure S5). Consistently, overexpression of TSC1/TSC2 complex did not inhibit ddaC pruning (unpublished data). Third, among 31 fly F-box proteins, Slimb was identified, which when knocked down via RNAi, resulted in apparent dendrite pruning defects in ddaC neurons. Furthermore, Slimb appears to be important for both ddaC dendrite pruning and MB γ axon pruning. Another F-box protein, Archipelago (Ago), is not required for ddaC pruning, in contrast to the essential role of its worm homologue SEL10 in synapse elimination [35]. MARCM analyses with ago3, a null allele [57], revealed no pruning defect in ddaC neurons (n = 3; Figure S5), supporting the selectivity of the E3 ligases for pruning. Finally, Nutcracker, the F-box protein of an SCF ubiquitin ligase (E3) required for caspase activation during sperm differentiation [58], appears not to be essential for ddaC pruning (n = 9, unpublished data).
We previously reported that a transcriptional hierarchy, consisting of EcR-B1, Sox14, and Mical, is commonly induced in both ddaC neurons and MB γ neurons, and essential for ddaC dendrite pruning [23]. How is the Cul1 E3 ligase integrated into this linear pathway? We propose that the Cul1-based SCF E3 ligase likely acts downstream of the transcriptional activators EcR-B1 and Sox14, however, in parallel to Mical, during ddaC dendrite pruning (see the model in Figure 9K). First, previous microarray analyses showed that the UPS genes, involved in all steps of the UPS pathway including cul1 and skpA, appear to be upregulated in remodeling MB γ neurons during the larval-pupal transition [46]. Upregulation of these genes is abolished by the expression of EcRDN [46]. Second, our Q-PCR data further verify that cul1 is a downstream effector of EcR-B1 and sox14, suggesting ecdysone signaling regulates the abundance, but not assembly, of the Cul1 SCF E3 ligase complex. Third, ddaC neurons devoid of the Cul1 E3 ligase retain strong expression of Sox14. Reduction of cul1 function does not enhance sox14 mutant phenotype, supporting that the Cul1 E3 ligase acts downstream of sox14 during ddaC pruning. Finally, the Cul1-based E3 ligase is dispensable for the transcription and translation of mical in ddaC neurons, suggesting that it likely exerts its roles in a Mical-independent pathway. This is further supported by the fact that reduction of the E3 ligase function in mical null mutant ddaC neurons results in additive effects on dendrite pruning. Taken together, our data support the model in which the Cul1-based SCF E3 ligase acts in parallel with Mical to govern ddaC dendrite pruning, downstream of EcR-B1 and Sox14 (Figure 9K).
During pattern formation, the Cul1-based SCF E3 ligase degrades Ci or Arm to negatively regulate Hh or Wg pathways, respectively [45]. Importantly, our data demonstrate that the InR/PI3K/TOR pathway, rather than Wg or Hh pathways, is inactivated by the Cul1-based E3 ligase in order for ddaC neurons to prune their dendrites. First, we show that inactivation of the InR/PI3K/TOR pathway suppresses the dendrite pruning defects in ddaC neurons lacking the Cul1 E3 ligase activity. Second, compromised cul1 function causes enhanced expression/activity of Akt, a positive regulator of the InR/PI3K/TOR pathway. These data suggest that compromised E3 ligase function causes activation of the InR/PI3K/TOR pathway. Consistently, a reversal of the UPS pathway by a deubiquitinating enzyme also leads to enhancement of the insulin signaling in mammals [59]. Finally, activation of the InR/PI3K/TOR pathway alone is sufficient to inhibit ddaC dendrite pruning. In contrast, activation of the InR/PI3K/TOR pathway by loss of PTEN function (n = 11) or InRCA expression (n = 7) is dispensable for MB axon pruning (Figure S19C). Given the distinct architecture and morphology of ddaC dendrites and MB γ axons, it is conceivable that the SCF E3 ligase regulates differential target pathways during these two types of pruning. Like the Cul1 E3 ligase, the InR/PI3K/TOR pathway acts downstream of EcR-B1/Sox14 and regulates dendrite pruning in parallel to Mical (see the model in Figure 9K).
It has been reported that ecdysone can inhibit the release of insulin-like peptides from insulin-producing-cells during the larval-pupal transition, systemically inactivate the InR/PI3K/TOR pathway, and thereby terminate larval growth [60]. We show that during dendrite pruning, ecdysone signaling increases abundance of the Cul1-based SCF complex, which in turn inactivates the InR/PI3K/TOR pathway (Figure 9K). These regulations with different mechanisms together would ultimately ensure lower levels of InR/PI3K/TOR activity/function in ddaC neurons, thereby resulting in dendrite pruning in these neurons during early metamorphosis. How does the Cul1-based SCF E3 ligase inactivate the InR/PI3K/TOR pathway during ddaC pruning? This negative regulation may occur through direct degradation of positive regulators of the pathway, such as IRS (fly Chico), Akt, and TOR. Several lines of evidence indicate that Akt is a good candidate as a substrate of the Cul1 SCF E3 ligase. First, the F-box protein Slimb, a component of the Cul1 SCF E3 ligase required for substrate recognition, specifically interacts with Akt. Second, Slimb promotes ubiquitination of Akt in a WD40-dependent manner. Third, Akt levels are significantly increased in cul1 RNAi mutant ddaC neurons and brains. Finally, knockdown of Akt strongly suppresses the cul1 RNAi dendrite pruning defects. Since the InR/PI3K/TOR pathway and Akt are important for ddaC dendrite pruning but not for MB axon pruning, it is conceivable that the targets of the Cul1 SCF E3 ligase are divergent in these two modes of neuronal pruning. Interestingly, the SCF-Fbxo40 ligase targets the mammalian IRS for degradation and limits the insulin/InR signaling in skeletal muscle [61], whereas the RING E3 ligase ZNRF1 ubiquitinates Akt to promote Wallerian degeneration of injured dorsal root ganglia neurons, a pruning-like process [62]. Finally, our results also show that the Cul1-based SCF E3 ligase and the InR/PI3K/TOR pathway regulate local caspase activation in ddaC dendrites during pruning. Attenuation of the insulin pathway can downregulate IAP, an inhibitor of caspases, during the differentiation of embryonic chicken lens epithelial cells [63]. It will be of great interest to determine whether global attenuation of InR/PI3K/TOR pathway can similarly downregulate DIAP1, a Drosophila homologue of IAP, to locally activate caspases during dendrite pruning.
In summary, we demonstrate that the Cul1-based SCF E3 ligase plays crucial roles in directing two distinct modes of pruning in ddaC and MB γ neurons in Drosophila. We further show that the Cul1 SCF E3 ligase inactivates the InR/PI3K/TOR pathway, a key signaling pathway, in order for ddaC neurons to prune their unnecessary larval dendrites during metamorphosis. This study provides a novel link between the SCF E3 ligase and the InR/PI3K/TOR pathway in regulating neuronal pruning. Thus, we open up new avenues for further studies of the E3 ligase in the remodeling and maturation of the developing nervous system, as well as their implications in the pathogenesis of many neurodegenerative diseases.
The following fly stocks were used in this study: cul1EX (C.T. Chien) [37], UAS-Flag-Cul1 (C.T. Chien) [37], UAS-cul1DN(generated in this study) [37], nedd8AN015 (C.T. Chien) [37], roc1aG1 [42], roc1bdc3 (R.J. Duronio) [43], slimb8 (B. Limbourg-Bouchon) [64], slimb2, tub-Myc-slimb (J. Jiang) [36], Df(3R)swp2MICAL, UAS-micalN-ter (A. Kolodkin) [65], mical15256 (the Yu lab), ppk-Gal4 (on Chr II or Chr III; Y. Jan) [66], UAS-CiCell (K. Basler), UAS-CiU (K. Basler) [67], UAS-DshDIX (N. Perrimon) [68], UAS-SmoΔ661-818 (J. Jiang) [69], UAS-NotchDN (d.n.N), UAS-NotchCA (act.N) (S. Artavanis-Tsakonas) [70], UAS-DomeΔCYT (J. Castelli-Gair Hombria) [71], UAS-PVRDN (P. Rorth) [72], UAS-Tkv1ΔGSK (M. O'Connor) [73], UAS-PTEN (T. Xu) [49], UAS-TSC1,UAS-TSC2 (T. Xu) [74], UAS-4E-BP(AA) (S. Cohen) [51], ago3 (I. Hariharan [57]), sox14Δ13(the Yu lab), mical-lacZ (the Yu lab), PTENc494 (T, Xu) [49], PTEN1 (C. Wilson) [75], UAS-CD8::PARP::Venus [25], elav-GeneSwitch-Gal4 [40], and UAS-skpA-RFP (generated in this study).
The following stocks were obtained from Bloomington stock centre (BSC): skpA1, skpA RNAi #1 (BL29874), FRT40A Cul3gft2,UAS-ptc, FRTG13 Cul411L, UAS-sggS9A, UAS-BskDN, UAS-InR, UAS-InRDN (InRK1409A), UAS-InRCA (InRA1325D), UAS-ArmS10, UAS-YkiS168A, UAS-EGFRDN, UAS-PI3K, UAS-PI3KDN (PI3KD954A), UAS-PI3KCA (PI3KCAAX), UAS-TORTED, UAS-S6KKQ, UAS-S6KCA (S6KSTDETE), UAS-akt, UAS-Rheb, akt RNAi (BL31701 and BL33615), and Gal42–21, 201Y-Gal4, OK107-Gal4, Gal4109(2)80, elav-Gal4, ppk-CD4-tdTomato.
The following Stocks were obtained from Vienna Drosophila RNAi Centre (VDRC): cul1 RNAi #1 (v108558), cul1 RNAi #2(v42445), cul1 RNAi lines (v33406 and v33407), roc1a RNAi #1 (v106315), roc1a RNAi #2 (v32398), skpA RNAi #2 (v32790), skpA RNAi #3 (v107815), slimb RNAi #1 (v107825), and slimb RNAi #2 (v34273).
The GATEWAY pTW vector containing a fragment of the cul1 cDNA (encoding aa 1–532; Cul1DN) or GATEWAY pTRW vector containing a fragment of the skpA cDNA were constructed and several transgenic lines were established by the Bestgene Inc.
The cDNA fragment corresponding to the last 359 aa of Mical was amplified by PCR and verified by DNA sequencing. The product was expressed using the GST expression vector (pGEX 4T-1, Pharmacia) and the purified protein was used to immunize guinea pigs to generate antibodies against Mical. The specificity of the guinea pig anti-Mical antibody was verified using mical mutants.
We carried out MARCM analysis, dendrite imaging, and quantification as previously described [23].
Larval and pupal fillet samples for each set of experiments were processed simultaneously, stained in the same tube, and imaged with the same parameters using Leica SPE confocal microscope. The following antibodies were used for immunohistochemistry at the indicated dilution: guinea pig Anti-Mical (1∶500), mouse polyclonal anti-Sox14 (1∶200), mouse anti-EcR-B1 (1∶50, DDA2.7, DSHB), rabbit anti-GFP (1∶1000, A11122, Invitrogen), mouse anti-FasII (1∶100; 1D4, DSHB), rabbit anti-Akt (1∶500, #4691L, Cell Signaling), and rabbit anti-cleaved PARP (1∶500, 2317-50, Abcam). Cy3- and Cy5-conjugated secondary antibodies (Jackson Laboratories) were used at 1∶400 dilution.
We carried out S2 cell culture, ecdysone treatment, and Western blotting as described [23]. For brain extracts, mutant brains were dissected in cold PBS and lysed in 2×SDS protein loading dye and boiled for 5 min, before Western blot analyses. We used mouse anti-Myc (1∶2,000, ab32, Abcam), rabbit anti-Flag (1∶1,000, F-3165, Sigma), rabbit anti-Cullin-1 (C.T. Chien), rabbit anti-SkpA (T. Murphy), rabbit anti-Akt (1∶1,000, #4691L, Cell Signaling), rabbit anti-Akt P-Ser505 (1∶1,000, #4054S, Cell Signaling), and rat anti-HA (1∶1,000, 11867423001, Roche). Flag-Slimb, Flag-SlimbΔWD40, Myc-Slimb, Myc-SlimbΔWD40, Myc-Akt, Flag-Akt, and Flag-Roc1a expression vectors were generated by Gateway cloning and were transfected into S2 cells using Effectene Transfection Reagent (Qiagen). The specificity of all the antibodies was examined in the individual RNAi S2 cells.
Transfected S2 cells or prepupae brains were homogenized with lysis buffer (25 mM Tris pH8/27.5 mM NaCl/20 mM KCl/25 mM sucrose/10 mM EDTA/10 mM EGTA/1 mM DTT/10% (v/v) glycerol/0.5% Nonidet P40) with protease inhibitors (Complete, Boehringer; PMSF 10 µg/ml, Sodium orthovanadate 10 µg/ml) in presence or absence of ecdysone (20E). The supernatants were used for immunoprecipitation with anti-Myc, anti-Flag, or anti-Akt overnight at 4°C, followed by incubation with protein A/G beads (Pierce Chemical Co.) for 2 h. Protein A/G beads were washed four times with cold PBS. Bound proteins were separated by SDS-PAGE and analysed by Western blotting with anti-Myc, anti-Flag, anti-Cullin-1, anti-SkpA, and anti-Akt.
S2 cells were transfected with Flag-Slimb, Flag-SlimbΔWD40, Myc-Akt, and pHsp70-hemagglutinin (HA)-ubiquitin (A.Sehgal). At 48 h postransfection, cells were homogenized with the lysis buffer (25 mM Tris pH8/27.5 mM NaCl/20 mM KCl/25 mM sucrose/10 mM EDTA/10 mM EGTA/1 mM DTT/10% (v/v) glycerol/0.5% Nonidet P40) with protease inhibitors (Complete, Boehringer; PMSF 10 µg/ml, Sodium orthovanadate 10 µg/ml). The supernatants were used for immunoprecipitation with anti-Myc, overnight at 4°C, followed by incubation with protein A/G beads (Pierce Chemical Co.) for 2 h. Protein A/G beads were washed four times with cold PBS. Bound proteins were separated by SDS-PAGE and analysed by Western blotting with anti-Myc, anti-HA, and anti-Flag.
Live confocal images of ddaC neurons expressing UAS-mCD8-GFP driven by ppk-GAL4 were shown at w3L, WP, and 16 h APF. The average number of primary and secondary dendrites attached to soma or total dendritic termini was counted from wild-type and mutant ddaC neurons. The number of samples (n) in each group is shown on the bars. Error represents S.E.M. Dorsal is up in all images. The strength of the pruning phenotypes was divided into three levels: mild (less than 5 primary and secondary dendrites), moderate (5–10 primary and secondary dendrites), and strong/severe (>10 primary and secondary dendrites).
Larval, pupal, and adult brains were dissected in PBS and fixed in 4% formaldehyde for 15 min. Brains were washed in PBS+1%Triton X for 3 times for 10 min each. For the clonal analysis study (MARCM), embryos were collected at 6 h interval. The clones were induced in the first instar lavae by applying a 1 h heat shock at 38°C. The antibody to FASII (1D4) was obtained from the Developmental Studies Hybridoma Bank and used at 1∶50. Rabbit anti-GFP antibody was obtained from Invitrogen and used at 1∶1,000. 201Y-Gal4 labels postmitotic γ neurons and a small subset of late-born α/β neurons. The Anti-FasII (1D4) antibody labels α/β neurons strongly and γ neurons weakly. The severity of the axon pruning phenotypes was divided into three levels: weak, strong, and complete, according to a previous study [21].
To quantify the immunolabeling intensities, cell nuclei (EcR-B1/Sox14 immunostaining) or whole cell body (Mical immunostaining) were drawn on the appropriate fluorescent channel based on the GFP channel relative cellular localization in ImageJ software. After subtracting the background (Rolling Ball Radius = 30) on the entire image of that channel, we measured the mean grey value in the marked areas in ddaC and ddaE on the same images and calculated their ratios. The ratios were normalized to corresponding average control values and subjected to statistical t test for comparison between different conditions (*p<0.05, **p<0.01, ***p<0.001, n.s., not significant). Graphs display the average values of ddaC/ddaE ratios and the standard error of means (S.E.M). n is shown on the bars. Insets show the ddaC neurons labeled by ppk-GAL4 driven UAS-mCD8-GFP expression. Dorsal is up in all images.
To quantify the immunolabeling intensities of Akt, soma/dendrite/axon regions were drawn on the appropriate fluorescent channel based on the GFP channel relative cellular localization in ImageJ software. After subtracting the background, we measured the mean grey value of Akt in the marked areas. The values were normalized to corresponding average control values and subjected to statistical t test for comparison between different conditions. Graphs display the average values of normalized Akt expression and the standard error of means (S.E.M). n is shown on the bars. Insets show the soma/dendrite/axon labeled by ppk-GAL4 driven UAS-mCD8-GFP expression (*p<0.05, **p<0.01, ***p<0.001, n.s., not significant).
To avoid any developmental delay, wild-type or mutant embryos were collected at 6 h intervals and were reared on standard food to the 3rd instar stage (96 h after egg laying, AEL) before being transferred to the standard culture medium containing 2 µM of Rapamycin dissolved in ethanol (Sigma Aldrich R0395). Larvae were fed in Rapamycin food for approximately 8 h before cessation of feeding. The onset of puparium formation and adult eclosion was not affected by Rapamycin treatment. wL3 larvae were used to quantify the total dendrite termini.
To avoid any developmental delay, wild-type or mutant embryos were collected at 6 h intervals and were reared on standard food to the e3L stage before being transferred to the standard culture medium containing 240 µg/ml mifepristone dissolved in ethanol (Sigma Aldrich M8046). The onset of puparium formation and adult eclosion was not affected by RU486 treatment.
Isolation of RNA from MB γ neurons was accomplished using Laser Capture Microdissection (LCM). Ten-micrometer frozen sections were cut from larval or pupal brains. MB γ neurons labeled by 201Y-GAL4>mCD8-GFP were microdissected using the Zeiss PALM microbeam microdissection system. Each capture consisted of ∼80 cell bodies, and 30 captures were pooled to obtain each replicate. Total RNA was extracted using the PicoPure RNA isolation kit from Arcturus and subject to reverse transcription using Oligo dT and the SuperScript III First-Stand Synthesis SuperMix (Invitrogen). The genomic DNA was digested by RNase-free DNase (Qiagen).
Independent experiments were conducted in triplicates using Maxima SYBR Green/ROX qPCR Master Mix (Fermentas) and 7900HT Fast Real-time PCR system (Applied Biosystems) according to the manufacturer's recommendations. rp49 (CG7939) was used as an internal control gene. Results were normalized to the controls indicated. Error bars represent standard error of mean (SEM) for four experimental repeats (n = 4).
Primers listed in 5′-to-3′ Sequence:
EcR-B1
CTGCTCATAGCCATCCTGGT
GCGGCCAAGACTTTGTTAAG
sox14
GAAAGATCTCCGAGCCACAG
ATCTGGCTCCAAACCATGAA
mical
TTGGTGGGCTTCCTTAGATG
GTTCAAACCGAGTCCGAGAG
cul1
CCACATGCGAAGAGGTTCTTAT
CAAGGATGGACTTGAGATCTGTC
uba1
GATATCCTTCTGTCGGGACTTG
GATATCGGCTTCCGTGAGATAG
rp49
GCTTCAAGGGACAGTATCTGATG
GACAATCTCCTTGCGCTTCTT
|
10.1371/journal.pgen.1004994 | Discovery of Transcription Factors and Regulatory Regions Driving In Vivo Tumor Development by ATAC-seq and FAIRE-seq Open Chromatin Profiling | Genomic enhancers regulate spatio-temporal gene expression by recruiting specific combinations of transcription factors (TFs). When TFs are bound to active regulatory regions, they displace canonical nucleosomes, making these regions biochemically detectable as nucleosome-depleted regions or accessible/open chromatin. Here we ask whether open chromatin profiling can be used to identify the entire repertoire of active promoters and enhancers underlying tissue-specific gene expression during normal development and oncogenesis in vivo. To this end, we first compare two different approaches to detect open chromatin in vivo using the Drosophila eye primordium as a model system: FAIRE-seq, based on physical separation of open versus closed chromatin; and ATAC-seq, based on preferential integration of a transposon into open chromatin. We find that both methods reproducibly capture the tissue-specific chromatin activity of regulatory regions, including promoters, enhancers, and insulators. Using both techniques, we screened for regulatory regions that become ectopically active during Ras-dependent oncogenesis, and identified 3778 regions that become (over-)activated during tumor development. Next, we applied motif discovery to search for candidate transcription factors that could bind these regions and identified AP-1 and Stat92E as key regulators. We validated the importance of Stat92E in the development of the tumors by introducing a loss of function Stat92E mutant, which was sufficient to rescue the tumor phenotype. Additionally we tested if the predicted Stat92E responsive regulatory regions are genuine, using ectopic induction of JAK/STAT signaling in developing eye discs, and observed that similar chromatin changes indeed occurred. Finally, we determine that these are functionally significant regulatory changes, as nearby target genes are up- or down-regulated. In conclusion, we show that FAIRE-seq and ATAC-seq based open chromatin profiling, combined with motif discovery, is a straightforward approach to identify functional genomic regulatory regions, master regulators, and gene regulatory networks controlling complex in vivo processes.
| The functional expression of all genes is regulated by proteins, namely transcription factors that bind to specific areas of DNA known as regulatory regions. Whereas most DNA in our genome is normally bound by other proteins (histones) and packaged into units called nucleosomes, a specific subset of tissue-specific regulatory regions is responsible for tissue-specific gene expression; these active regions are nucleosome-depleted and bound by transcription factors. We use two techniques to identify these open chromatin regions, in a normal tissue and a RasV12 induced cancer tissue. We discovered a remarkable change in the accessible regulatory landscape between these two tissues, with several thousand regions becoming more accessible in the cancer tissue. We identified two transcription factors known to be involved in cancer (AP-1 and Stat92E) controlling these newly accessible regulatory regions. Finally, we introduced a mutation resulting in Stat92E becoming non-functional in the cancer tissue, which decreased the severity of the tumor. Our study shows that open chromatin profiling can be used to identify complex in vivo processes, and we shed new light on Ras dependent cancer development.
| Gene expression in higher eukaryotes is tightly controlled by complex cis-regulatory systems consisting of multiple enhancers modulating the transcription levels of a target gene. Mapping all active promoters and enhancers in a cell type provides an entry point to reverse engineer the functional gene regulatory network and to decipher the genomic cis-regulatory logic underlying a cell type’s functional transcriptome. Understanding changes in regulatory landscapes between different cell types, for example during cellular differentiation, or between normal and diseased cellular states, can furthermore provide a bridge between the genomic sequence and emerging transcriptome changes. Recent advances in chromatin studies have uncovered several characteristic features of active and repressed chromatin within and around regulatory regions. A typical active promoter or enhancer in higher eukaryotes is depleted for nucleosomes over relatively large regions of up to hundreds of base pairs, often spanning the entire enhancer length [1]. In addition, nucleosomes flanking active regulatory regions usually carry histone modifications, such as H3K27Ac and H3K4me1 [2], and in human and other vertebrates, active promoters and enhancers may have dynamic hyper- or hypomethylated CpG dinucleotides, denoting inactive and active enhancers respectively, whereby changes in methylation usually accompany changes in activity [3]. All of these features, to some extent, correlate significantly with the expression of the target gene(s) that they control [1,4]. Furthermore, these features can be exploited to identify regulatory regions at a genome-wide scale, for example, integrative methods have been developed based on Hidden Markov Models that accurately predict various types of regulatory regions using particular combinations of histone modifications [5]. Another, more recently discovered feature of active enhancers is that RNA, known as eRNA, is transcribed from them in a bidirectional way by RNApolII [4]. This property was used in the Fantom project and lead to the identification of ∼44,000 tissue and cell type specific enhancers in the human genome. However, the technique that has been primarily used to map regulatory landscapes across hundreds of human cell lines, notably in the ENCODE and Epigenomics Roadmap projects, is the detection of open chromatin using DNaseI hypersensitivity coupled with high throughput sequencing (DNaseI-seq) [1,6,7]. DNaseI-seq identifies regions by fragmenting chromatin with the DNase I enzyme, an endonuclease that randomly cleaves regions of accessible DNA, and for which cleavage is hindered by the presence of nucleosomes. This cleavage results in an increased number of cut sites in nucleosome-depleted regions, generating fragments with smaller sizes, allowing their enrichment based on size-selection, before high-throughput sequencing. As such, DNaseI-seq has mapped genome-wide regulatory element across many human and mouse cell types, as well as across developmental stages during Drosophila embryogenesis [8]. DNaseI-seq however, has some limitations as it requires large amounts of input material and has a complicated procedure. Consequently, it has been mainly applied to in vitro cell cultures and cancer cell lines and has seen limited applications to in vivo post-embryonic biological systems. For example, it is generally not possible with this approach to assay to what extent the open chromatin profile of cells within a tumor, or during tumor development, is altered. It is indeed conceivable that the joint interpretation of cancer genomes and cancer transcriptomes, both of which can be sequenced directly from tumor biopsies down to the single cell level, will require the intermediate layer of the “functional genome” to permit us to understand how the epigenomic changes drive changes in gene expression. Therefore, to profile this chromatin layer, alternative approaches are required to progress towards smaller cell populations.
Two alternative methods for open chromatin profiling were recently developed which overcome the limitation of sample size, namely FAIRE-seq (Formaldehyde Assisted Isolation of Regulatory Elements) [9], and ATAC-seq (Assay for Transposase Accessible Chromatin) [10]. We wanted to know whether these methods are both suitable to identify the functional regulatory elements operating in a wild type tissue or in a tumor. FAIRE-seq relies on the separation of open versus closed chromatin by phenol-chloroform extraction, whereby fragments with high nucleosome content are captured in the organic phase, and sequencing libraries are prepared from the aqueous phase [9]. FAIRE-seq derived genome-wide enhancer maps, although noisier than DNaseI-seq have been shown to be highly correlated with DNaseI-seq [11]. FAIRE-seq has recently been applied successfully to identify differential enhancer usage between multiple Drosophila tissues and developmental time points, finding thousands of enhancers that change activity during development [12]. The most recent method, ATAC-seq, uses a bacterial (Tn5) transposase, an enzyme that inserts a short fragment of DNA (a transposon) into another molecule of DNA, or in this case, inserts two short fragments separate from each other [10]. As the transposase is unable to access DNA that is either bound by nucleosomes or strongly bound transcription factors, it integrates its transposons preferentially into accessible or open chromatin. In the case of ATAC-seq the transposase inserts two fragments of DNA which act as tags, and a mechanism to fragment the DNA (as they are inserted ∼9 base pairs apart), a process known as tagmentation [13]. Buenrostro et al. recently showed that ATAC-seq was able to accurately identify the nucleosome free regions of a lymphoblastoid cell line (GM12878), and the authors were able to obtain profiles comparable to DNaseI-seq in signal to noise ratio and specificity from much lower quantities of cells (100–10000 fold less) than generally used for DNaseI-seq [10]. This technique is thus promising to apply to dissected tissues, for example by micro-dissection of tumor samples, sorting of low populations of cells or other low-input samples.
In this study we compare FAIRE-seq and ATAC-seq to discover active regulatory regions in a wild type tissue and during tumor development, and compare both approaches in terms of signal-to-noise ratio, accuracy of enhancer identification, resolution to recover TF footprints, and ability to identify changes during cellular state transitions, such as during oncogenesis. As a model system, we have used the developing Drosophila eye on the one hand, and a genetically induced tumor model in the developing Drosophila eye on the other hand. The tumor model is based on the over-expression of oncogenic RasV12 combined with the loss of the cell polarity gene scribble (scrib), and is a well-established model to study Ras-dependent oncogenesis in vivo [14,15]. The combination of RasV12 and scrib-/- mutations cause differentiation defects, coupled with over-proliferation and invasion of the surrounding tissue [16]. Whether such a severe phenotypic change is driven by transcription factors operating within a stable, unchanged chromatin state, or whether changing gene expression profiles are accompanied by widespread changes in the regulatory landscape, has not yet been investigated. Given the similarity between this invasive cancer model and the mesenchymal state transitions occurring in human epithelial cancers, such as epithelial-to-mesenchymal transition, further insight into chromatin modulation in this model system may also be relevant to understand regulatory changes during human oncogenic processes.
Our results suggest that open chromatin profiling can provide valuable and previously unobtainable information crucial for understanding how regulatory information is encoded in the genome and how regulatory programs determine phenotype and behavior in vivo.
The larval eye-antennal disc of Drosophila melanogaster is a widely used model system for the study of spatio-temporal gene regulation and cellular differentiation. This is because in one tissue both dimensions of "time and space" are present, with cells in many different states, from pluripotent cells and specified neuronal precursors, to the lineage committed sensory neurons and accessory cells [17–20]. To map the plethora of regulatory regions operating in all of these cell types, and to simultaneously assess the performance of different biochemical approaches to identify accessible regions, we applied both ATAC-seq and FAIRE-seq to several different genotypes of wild type eye-antennal discs (see Materials and Methods) (Fig. 1). When looking at well-known master regulators of eye development, such as Optix and sine oculis (so), we found both methods to yield highly reproducible results whereby the open chromatin peaks mark previously known enhancers within these genes (Fig. 1). Peaks called by both methods highly overlap (Fig. 2A), and the normalized peak heights between the methods are strongly correlated (Fig. 2B-D). Both methods show a similar distribution of peaks within promoters, intronic, and intergenic regions, and a very strong depletion in coding exons, although ATAC-seq peaks are slightly more biased towards distal regions relative to promoter peaks than FAIRE-seq (S1 Fig.).
To investigate how accurately these methods can identify promoters, enhancers, and insulators, we looked in turn to each of these three classes of regulatory elements. Firstly, the accessibility of promoters and transcription start sites is overall highly enriched for both methods, although ATAC-seq shows slightly lower background levels, indicating a higher signal-to-noise ratio for ATAC-seq (Fig. 2E). Secondly, to evaluate whether distal open chromatin peaks also identify functional enhancers, we made use of two collections of Drosophila enhancers, namely the REDFly database and the FlyLight [21] database. Out of 56 REDFly enhancers that are active in the eye, ATAC-seq and FAIRE-seq recover 28 and 24 respectively whilst combined they recover 34. The recovery rates for eye enhancers are the highest of all tested enhancer categories from REDFly, which illustrates the specificity of these approaches at similar thresholds in the ranked list of all genomic regions (S2 Fig.). Likewise, eye enhancers from the FlyLight database show increased chromatin activity (Fig. 2F); and the highest overlap between our open chromatin peaks and all FlyLight enhancers is found for eye enhancers (402 out of 576 eye enhancers have an open chromatin peak), much more than for adult brain enhancers and embryo enhancers [22] (Fig. 2G). Although both approaches have a good performance for enhancer detection, ATAC-seq shows a higher recall of true enhancers than FAIRE-seq, detecting ∼18.75% more enhancers (blue versus red bars in Fig. 2G), even at similar levels of specificity (S2 Fig.). Thirdly, we also asked whether the open chromatin peaks overlap with one more regulatory genomic feature, namely functional insulators. To identify insulators, we performed ChIP-seq against CTCF in the wild type eye-antennal disc, under the same conditions as the ATAC-seq and FAIRE-seq (see Materials and Methods). We identified 3682 CTCF peaks, which are significantly enriched for the CTCF motif (PeakMotif adj. p-value = 2.63*10–15) (Fig. 3A). Based on this motif, we selected 805 high-confidence CTCF binding sites, having both a significant peak and a significant motif. Both ATAC-seq and FAIRE-seq signals are strongly enriched in the regions around the CTCF binding sites (Fig. 3A-C). Generally, CTCF peaks were identified as accessible regions, with 2244 of all the 3682 CTCF peaks overlapping (minimum 40%) with ATAC or FAIRE peaks. Of the 805 high-confidence CTCF sites, 279 and 278 are effectively (minimum 40% overlap) called as peaks by ATAC-seq and FAIRE-seq respectively, indicating a highly similar detection rate of both techniques. Thus, both approaches allow for efficient genome-wide detection of promoters and enhancers whilst also detecting insulator sites, starting from low input tissue samples.
To further analyze the resolution of both open chromatin profiling methods we assessed whether ATAC-seq and FAIRE-seq can be used for transcription factor footprinting [23]. In the original publication of ATAC-seq [10], it was shown that CTCF binding regions, determined by ChIP-seq, show increased ATAC-seq signals, while certain nucleotides within the actual CTCF binding site are protected from tagmentation, similar to protection from DNaseI cleavage in DNaseI-seq. To test whether this is also the case in Drosophila eye development, we centered the 805 high-confidence CTCF peaks on the best scoring occurrence of the CTCF motif (we used the strongest enriched PWM) to investigate open chromatin signals around CTCF binding sites (Fig. 3A) with background signal removed. Similarly to DNaseI-seq, for ATAC-seq the actual CTCF sites show a clear drop in cut sites. This indicates that, although the regions are generally open and therefore accessible by the transposase, the CTCF protein protects its binding site from being cut at specific nucleotides (Fig. 3B-C). This further demonstrates the high degree of resolution obtained with ATAC-seq while identifying active genomic regulatory elements.
A second structural feature of ATAC-seq, as shown in the original publication for human chromatin in vitro, is its ability to correlate distances between cut sites with nucleosomal positioning [24]. To assess whether such information could also be obtained from small in vivo samples in Drosophila, we sequenced two samples with paired-end sequencing at low coverage, and found a distribution of insert sizes that almost perfectly resembles that of human chromatin (Fig. 3D-E). Interestingly, this analysis also shows that even by shallow sequencing, with as few as 0.5 million mapped reads, regulatory regions can be accurately identified across the entire genome due to the high signal-to-noise ratio of ATAC-seq (S3 Fig.), and that both single-end and paired-end sequencing provide near identical results (S4 Fig.). Taken together, these two experiments illustrate that ATAC-seq can identify nucleosome-depleted regions, and protected nucleotides at high resolution, even from low-input in vivo material from a heterogeneous tissue.
Encouraged by the power of ATAC-seq and FAIRE-seq to identify active regulatory regions in heterogeneous wild type tissues, we combined both methods to map all the differentially activated regulatory regions during tumor development. To this end, we used a well-established cancer model in which the eye disc is transformed by over-expression of oncogenic Ras protein (RasV12) in combination with a homozygous scrib-/- mutation (Fig. 4A-E). The combination of these two perturbations in clones of cells in the eye disc has been shown to generate invasive tumors and to serve as a bona fide cancer model [14,25–29], but there is only limited molecular and pathway characterization of these tumors.
To identify which parts of the genome are differentially open in the tumors we performed ATAC-seq and FAIRE-seq both on early tumors (RSE, Fig. 4C,D) and late tumors (RSL, Fig. 4E) (see Materials and Methods) and compared their open chromatin landscapes to that of the wild type tissue. We found that the regulatory landscape in tumors is drastically different from the wild type, having both thousands of significantly increased peaks (4851 for ATAC-seq) and thousands of significantly decreased peaks (4984 for ATAC-seq). Notably, the dynamic range of ATAC-seq seems to be greater than that of FAIRE-seq, as ATAC can detect both smaller and greater significant differences between normal and tumor states than FAIRE (Fig. 4F). Interestingly, when we apply a statistical model allowing for the analysis of both ATAC-seq and FAIRE-seq signals together, the total amount of significantly opening and closing peaks increases to 11516 (Fig. 4G and Materials and Methods); these differential peaks have slightly different genomic distributions, although at present it is unclear which mechanism could be causing this (S5 Fig.).
Since a database with regulatory regions specific for these tumors does not exist, we needed an alternative approach to test whether our candidate regulatory regions are indeed functional regions affecting gene expression of their candidate target genes. To investigate this, we ranked all our regions according to their fold-change between wild type and tumor tissue, and linked them to their candidate target genes based on their location in the genome (see Materials and Methods). Using publicly available gene expression data obtained under the same conditions as ours (GEO accession: GSE42938) [30] we examined the enrichment of the differentially expressed genes in the RasV12; scrib-/- tumors relative to this ranked region-gene list. The enrichment plots in Fig. 4H (by Gene Set Enrichment Analysis, see Materials and Methods) show that the up-regulated genes strongly correlate (p-value < 10–7, and Normalized Enrichment Score (NES) = 2.4) with the differentially opening regions, while genes down-regulated in the tumor strongly correlate (p-value < 10–7, NES = -2.4) with the genomic regions showing a decrease in open chromatin (differential gene-region pairs are available in S1–S2 Tables). The correlation still holds when we stratify our putative regulatory regions in two groups based on their distance from a TSS (proximal versus distal) or when we assign an alternative way to assign peaks to genes (S6 Fig.). This indicates that the differentially opening/closing chromatin regions in the tumor tissue are overall functional and play a role in the perturbation of the transcriptome. An example of a tumor-specific regulatory region is a previously unknown putative enhancer within an intron of p53 for which the strongly increased chromatin opening likely points to the underlying activation of p53 in the tumor cells (Fig. 4I).
Next, we wanted to test whether the newly opened chromatin during tumor formation corresponds to functional regulatory regions that have an endogenous role in other tissues during development. To this end we took the 4111 differential peaks whose normalized number of reads was at least two fold increased in the tumor samples when compared to the wild type control, and clustered these peaks into 3778 unique candidate regulatory regions that gain activity in the tumor (see Materials and Methods). We compared this set of differentially active regions against the entire collection of REDFly, FlyLight, and VDRC enhancer resources, covering a total of more than 16000 enhancers with at least one tissue of activity. The most significant overlap was found with sets of enhancers known to be active in the “leading edge of invading tissue”, and with “epidermis” and “midgut primordium” (S3 Table). These relationships may indicate re-activation of endogenous invasive processes. Activated regulatory regions also overlap with genitalia enhancers, which may indicate a re-activation of germline expression typical for pluripotent stem cells, corresponding to previous studies [31]. An example of such a gene is fruitless (fru), for which our data pinpoints a previously described genitalia enhancer [21] that may underlie the observed over-expression of fru during the oncogenic process in the RasV12; scrib-/- cancer model, and in another cancer model in the Drosophila brain [31] (S7 Fig.).
In conclusion, profound changes in the open chromatin landscape can be identified between wild type tissue and RasV12; scrib-/- tumor tissue using ATAC-seq and FAIRE-seq. In the tumor, many endogenous enhancers and promoters are ectopically activated, and their activity strongly correlates with changes in gene expression.
Having identified the exact locations of activity gaining regulatory regions in the tumors downstream of oncogenic RasV12; scrib-/- provides us with a high-quality set of sequences that are likely regulated/bound by a shared set of transcription factors. To predict which transcription factors might be binding to these activity gaining tumor-induced regulatory regions, we used a recently developed motif discovery method, known as i-cisTarget [32]. From a total of 9713 candidate TF motifs (as position weight matrices (PWM)), i-cisTarget yielded the AP-1 and Stat92E motifs as the two most enriched, AP-1 is ranked 1st with a normalized enrichment score (NES) of 8.73 and Stat92E is ranked 2nd with a NES of 5.1. In the active, but unchanging, regulatory regions the Stat92E motif is not enriched and there is only a minor enrichment for the AP-1 motif (NES = 2.7), indicating that the enrichment is specifically in the activity gaining regulatory regions. (see Materials and Methods for an explanation of the NES score). Remarkably, of the tumor-induced regulatory regions, AP-1 is predicted to target the majority (3065 of the 3778, or 81%), pointing to its driving role in oncogenesis. The AP-1 complex is a homo- or heterodimer of bZIP proteins such as Jun and Fos, binds to highly similar DNA motifs, and is functionally activated downstream of phosphorylated Jun Kinases (JNK). Interestingly, multiple labs have shown the importance of JNK signaling, and of the AP-1 complex, specifically to the development of RasV12; scrib-/- tumors. Particularly, either suppressing JNK signaling or knocking-down the AP-1 complex, is sufficient to block tumor development [14,29,33,34]. The Stat92E motif is enriched in a smaller part of the regulatory regions and shows a significant overlap with the predicted AP-1 responsive regions (Fig. 5A, B). This is consistent with reports that show that these pathways can have synergistic effects on RasV12; scrib-/- tumorigenesis [16].
Besides AP-1 and Stat92E, a few additional motifs for other TFs are also significantly enriched, although with a lower representation (Fig. 5A). One of them is the Scalloped (Sd) motif, a transcription factor that acts together with its coactivator Yorkie (Yki) to promote tissue overgrowth, as effectors of the Hippo signaling pathway [35]. It has been shown that knockdown of either Scalloped or Yorkie can rescue scrib-/- mutant tissue overgrowth and reduces RasV12; scrib-/- tumor size [36]. Another group are the zinc-finger protein motifs of Zelda (Vfl) that are known for their activating roles in early Drosophila development [37], and for which a role during oncogenesis has not been described. Note that not all enriched motifs are necessarily involved in the regulatory oncogenic program, and some can be “bystander” motifs for the key regulators. An example of such a bystander motif could be the Zelda motif, which often co-occurs with Stat92E motifs in the same regulatory region because these two TFs cooperate during early embryonic development [38], but in the eye tumor the zelda gene is not mis-regulated, while all 3 the ligands of the JAK/STAT signaling pathway are (S1 Table). Finally, we recovered several nuclear receptor motifs (e.g., Ftz-f1 and Hr39) and motifs of the TFIIB-related factor (Brf), which increases RNA polymerase III-mediated transcription, and its overexpression has been linked to several human cancer types [39]. Interestingly, this set of candidate Brf-regulated tRNA genes could be discovered by open chromatin profiling, but not by microarrays or mRNA-seq, showing another advantage, and complementarity, of using chromatin profiling besides classical gene expression profiling.
Next we asked whether the activated regulatory regions in the early tumors become even more activated in the late tumors, or if different regulatory regions become activated during tumor progression. In the early tumor, about half of the eye disc still consists of normal tissue, while in the late phase the tumor tissue has overtaken the entire eye disc and invades into the optic lobes of the brain and the ventral nerve cord to a greater extent. By comparing wild type versus early tumors, and early tumors versus late tumors, we found that the majority of changes are gradual, showing an increase in the height of the open chromatin peak between wild type and early tumor, and a further increase between the early and the late tumor. Such a gradual increase of the peak height on a regulatory region most likely indicates that this region is becoming accessible in a higher fraction of cells in the dissected tumor tissue, which is likely the consequence of a lower percentage of normal cells in the late tumor tissue. On the other hand, we found a subset of regulatory regions that are more open in the early tumor versus wild type, but do not show an increasing signal in the late tumor (determined by Fisher’s omnibus, see Materials and Methods). Interestingly, the motif enrichment scores for AP-1 and Stat92E are very different between the gradually and stably opening regulatory regions. More specifically, the stably open regions are mainly enriched for Stat92E (ranked 1st, NES = 7.83), while the enrichment for AP-1 motifs is reduced (ranked 3rd, NES = 4.74). On the other hand, the gradually open regions are strongly enriched for AP-1 motifs (ranked 1st, NES = 12.4), while the Stat92E motif is no longer enriched in this set (NES < 2.5) (Fig. 5C). This finding may indicate that, either Stat92E targets are activated earlier than AP-1 targets, or that a relatively small proportion of cells in the invasive tumor retain Stat92E activity. It could also indicate that Stat92E is active in both tumor and non-tumor cells, and that the secretion of Unpaired ligands from the tumor cells can cause non-autonomous activation of the JAK/STAT pathway in surrounding non-tumor tissue [40].
In conclusion, motif inference on differentially open chromatin peaks during tumor development provides valuable hypotheses about the identity of the transcription factors that are driving the oncogenic regulatory program. Our data confirms previous observations of an important role for AP-1 and Stat92E downstream of RasV12; scrib-/- induced oncogenesis, and now suggests that these two regulators can explain a very large fraction of the changing chromatin landscape.
One of the predicted transcription factors involved in changing the open chromatin landscape in the tumors is Stat92E, the effector of the JAK/STAT signaling pathway [41]. All three the ligands of this pathway are present in our list of significantly up-regulated gene-peak pairs (S1 Table), strongly supporting the predicted role of JAK/STAT signaling in the tumor. Indeed, previous reports have shown that blocking the activity of the JAK/STAT pathway, for example by using a dominant negative form of the receptor Domeless, reduced the tumor phenotype [16]. However, the direct involvement of Stat92E in this process has not been explicitly tested. We therefore incorporated a null mutation of Stat92E in the RasV12; scrib-/- tumors (Stat92E[85c9], see Materials and Methods) to determine if Stat activity is required for the tumor phenotype. We observed that tumor growth was severely reduced (Fig. 6A-C). In addition, in the Stat92E loss-of-function, a fraction of larvae with reduced tumors now also reach pupation stages, which is not observed in RasV12; scrib-/- larvae.
Based on our motif predictions (see above), Stat92E may activate more than three hundred regulatory regions and thereby play a role in regulating the nearby target genes, downstream of RasV12; scrib-/- during oncogenesis. If these promoters and putative enhancers depend on Stat92E for their opening, the same regions may be activated and opened by Stat92E alone, independently of the RasV12; scrib-/- induced oncogenesis. To test this we hyper-activated the JAK/STAT pathway in the wild type eye disc, and thereby the downstream Stat92E activity, by overexpressing one of the ligands that is up-regulated in the tumor, namely Unpaired (Upd, encoded by os). ATAC-seq on these discs could demonstrate that the tumor-induced changes at Stat92E predicted target sites are recapitulated by JAK/STAT activation alone. The changes are significant and mainly in the same direction (Fig. 6D), but as expected the changes in open chromatin caused by the Upd overexpression are quantitatively more subtle compared to those in the tumors. For example, an intronic regulatory region of Imp shows activated chromatin by Stat92E (log2 fold change = 1.23) but this activation is stronger during RasV12; scrib-/- tumor formation (log2 fold change = 3.44) (Fig. 6E).
In the RasV12; scrib-/- tumors we had identified 356 candidate target regulatory regions of Stat92E, of these we found that 72% have a positive fold change in the Upd over-expressing discs (Fig. 6D), indicating that this group is Stat92E responsive (p-value 0.0097). However, this still means that 28% of the candidate regions did not respond to the activation of Stat92E by overexpressing Upd. We analyzed these responsive and non-responsive regions and found that the AP-1 motif is only enriched in the regions that are not responsive to Stat92E, indicating that for those regions AP-1 might be the main input and the Stat92E motif might be of less importance here.
Next, we asked whether these changes in cis-regulatory activity directly affect gene expression of target genes. To test this, we linked the 254 Stat92E responsive regulatory regions to nearby (<5kb upstream or intronic) candidate target genes and examined their expression in Upd over-expressing discs using publicly available gene expression data (GEO accession: GSE15868) [42]. After applying two filtering steps, requiring the regulatory region to go significantly open both in tumor and UPD-overexpressing tissue, combined with a significant differential expression of the assigned gene, we end up with a final list of 28 potential direct Stat92E target genes (Fig. 6F). Among these are at least seven well-known transcriptional targets of the JAK/STAT signaling pathway including: dorsal, pointed, lama, chinmo and trol, as well as two key negative regulators of this pathway: Socs36E and Ptp61F [42–49]. Interestingly, not all genes that have an opened Stat92E regulatory region are up-regulated. The JAK/STAT pathway can also repress transcription of target genes and is known to block the Wnt/Wingless signaling pathway in the eye imaginal disc [50]. In our list we recover two genes that are known to be involved in the Wingless signaling pathway, wingless itself and sulfateless, an essential enzyme for this pathway [51]. This may indicate that Stat92E can function directly in the repression of genes, through an as yet unidentified mechanism. An interesting future challenge will be to understand the cis-regulatory mechanisms and possible co-factors that determine the mechanisms by which Stat92E can act as activator or repressor.
In conclusion, we demonstrate that Stat92E is a significant transcriptional regulator and required for the growth of our tumor model. We identify known and newly predicted Stat92E targets in the tumor and are able to independently recover >70% of those targets using a Stat92E activation model in the normal eye. This second analysis also illustrates the feasibility of an integrated approach of ATAC-seq and motif discovery to capture and annotate modest, yet functional, changes in the chromatin landscape. We conclude that Stat92E induced chromatin-opening correlates with a change in the transcription of nearby genes, and that Stat92E may function as a key regulator of the tumor transcriptome.
Genome-wide characterization of all promoters and enhancers controlling a particular gene expression profile and/or phenotype is a key challenge for understanding the regulatory underpinnings of any biological process in vivo. Our study first compares, and subsequently combines, two recent methods for open chromatin profiling to obtain genome-wide regulatory landscapes in the developing Drosophila eye as model system. By applying FAIRE-seq and ATAC-seq to the normal eye-antennal imaginal disc, alongside anti-CTCF ChIP-seq, we found both methods to be highly robust in identifying accessible or open regions, with few differences between strains with different genetic backgrounds. The main advantages of ATAC-seq, for our application, are besides its undemanding experimental procedure, (1) its higher signal-to-noise ratio, with low background signal and sharper peaks; (2) its ability to identify TF footprints, as binding sites are protected from transposon insertion, similar to their protection from DNaseI cleavage [23]; and (3) its ability to determine nucleosome positioning when using paired-end sequencing.
Although these assets of ATAC-seq may be important for some studies, overall both FAIRE and ATAC allow identification of promoters and candidate enhancers, and here we use them as independent “replicate” measurements (taking batch effects into account) to examine the open chromatin status of a tumor model.
We observe that the cis-regulatory landscape of active promoters and enhancers changes dramatically in RasV12; scrib-/- eye tumors, while more moderate changes were observed in tissues with JAK/STAT induced hyperplasia. A possible explanation for the marked differences in the intensity of change may be found in the cellular composition of the samples. Open chromatin signals represent the average signal across all cells within a sample, and since each regulatory region, in each cell can yield only two open alleles, this observed activity mainly reflects the number of cells in which the region is active, rather than the quantitative activity of the region.
Once the repertoire of tumor-induced regulatory regions was identified by open chromatin profiling, we reasoned that if many of these functional regions are activated by a small set of master regulators, then the motifs of these TFs should be enriched within the sequences of these regions. While motif inference on gene sets is challenging due to the large intergenic and intronic regulatory space around genes, motif inference on enhancer-size regulatory sequences often gives highly accurate results [32,52–54]. We used the tool i-cisTarget, which is optimized for Drosophila genomes, and found not only AP-1 and Stat92E, but also Zelda, Scalloped (the Hippo pathway effector), Brf, and Ftz-f1 as candidate regulators of the oncogenic, Ras-dependent program. All of these TFs (except perhaps Zelda) can be directly linked to cancer-related processes that are conserved to human [35,39], and AP-1, Stat, and Scalloped have each been previously linked to the RasV12; scrib-/- program specifically [16,29,36,55,56]. We could confirm that the RasV12; scrib-/- tumor phenotype depends on Stat92E, and furthermore reveal that JAK/STAT signaling causes specific chromatin changes at Stat92E-responsive regulatory regions.
Finally, once the regulatory regions and their candidate regulators are identified, an important next step is to examine which target genes are now differently regulated, as a consequence of the activation of these promoters and enhancers. Previous work has found that open chromatin peaks obtained across cell lines are correlated with gene expression changes of “nearby” target genes [4]. We also found evidence that a high degree of chromatin changes are concordant with transcriptional changes of nearby target genes, both in the tumor and in the overgrown tissue with hyper-activated JAK/STAT, confirming that the chromatin state of these regions is not only altered, but that they are also functionally activated. Interestingly, not all putative target genes with an increased open chromatin peak are up-regulated, but a small subset is also down-regulated, which indicates that regulatory regions can be “activated” by TF binding and nucleosome depletion, but that the consequence of this activation can also be gene repression (e.g., if the bound TF act as a repressor).
Overall, our integrated approach reveals a large cistrome of changing activity at promoters and enhancers during tumor development, which is mainly operated by AP-1 and Stat92E; and illustrates how integrative open chromatin profiling, motif detection, and gene expression analyses have great potential to unravel tissue and cell type specific regulatory programs in vivo, in health and disease.
The following fly stocks were used during this investigation: y,w,eyFlp; act>y+>Gal4, UAS-GFP; FRT82 tub Gal80 and y,w; UAS-RasV12; FRT82 scrib2, e/ TM6B, yw; UAS-Rasv12; FRT82 Stat92E85c9, scrib2, e/ FRT82 tub-Gal80, Optix-GFP [57] and UAS-GFP:Upd (Courtesy of Fernando Casares), GMR-Gal4, isogenic wild type DGRP-208 (Bloomington stock 25174), y, w; FRT82 and CantonS wild types. Crosses were raised at 25°C on a yeast based medium.
Wild type, RasV12; scrib-/- early and Upd overexpression eye-antennal discs were dissected from wandering third instar larvae (day 6) in 1xPBS. RasV12; scrib-/- late discs were collected three days after larvae began wandering (day 9); this is possible because RasV12; scrib-/- do not pupate, but can persist more than one week in a prolonged larval stage. Immunohistochemistry was performed as previously described in [58]. Confocal images were taken on an Olympus FV1000 or FV1200 microscope. Images were processed using ImageJ and Adobe Photoshop software.
The previously described ATAC-seq protocol was adapted for working with Drosophila rather than human cells [10]. Ten eye antennal imaginal discs (or three RasV12; scrib-/- late total tumors) were immediately placed in 50 μl ice cold ATAC lysis buffer (10 mM Tris-HCl, pH 7.4, 10mM NaCl, 3mM MgCl2, 0.1% IGEPAL CA-630). Lysed discs were then centrifuged at 800 xg for 10 minutes at 4’C and the supernatant was discarded. The rest of the ATAC-seq protocol was performed as described previously [10] using the following primers: Fwd:- ‘AATGATACGGCGACCACCGAGATCTA CACTCGTCGGCAGCGTCAGATGTG’ and Rev:- ‘CAAGCAGAAGACGGCATACGAGATXXX XXXGTCTCGTGGGCTCGGAGATGT’ (where X indicates barcode nucleotides). The final library was purified using a Qiagen MinElute kit (Qiagen) and Ampure XP beads (Ampure) (1:1.2 ratio) were used to remove remaining adapters. The final library was first checked on an Agilent Bioanalyzer 2000 for the average fragment size. Resulting successful libraries were sequenced with 50bp, single end reads on the Illumina HiSeq 2000 platform. Single end sequencing was chosen for this study because we were not interested in the fragment contents (i.e., how many nucleosomes are placed between two insertion sites), rather just the profile of insertion sites, which also made comparisons with pre-existing FAIRE-seq data easier.
Methodology adapted from [12]. In short, 150 head complexes were dissected from wandering third instar larvae, these were fixed for 10 min with 4% formaldehyde,. The formaldehyde was then replaced with 750 μl quenching buffer (125 mM Glycine 0.01% Triton X-100 in PBS) was added and incubated at room temperature for 10 minutes. Quenching buffer was replaced with buffer A (10 mM HEPES-KOH pH8.0, 20 mM EDTA pH8.0, 1mM EGTA pH8.0, 0.25% Triton X-100, 1mM PMSF) and 200 eye-antennal discs were then dissected in buffer A and kept on ice, these were centrifuged at 6000rpm, 4’C to pellet the discs and lysed in lysis buffer 1 (50mM HEPES-KOH, pH 7.5, 140mM NaCl, 1mM EDTA, 10% glycerol, 0.5% NP-40, 0.25% Triton X-100) rocking at 4’C for 10 minutes, centrifuged and the supernatant removed. Next lysis buffer 2 (10mM Tris-HCl, pH 8.0, 200 mM NaCl, 1mM EDTA, 0.5 mM. EGTA) was added and the sample was rocked at room temperature for 10 minutes. Finally lysis buffer 3 (10 mM Tris-HCL, pH 8.0, 100mM NaCl, 1mM EDTA, 0.5mM EGTA, 0.1% Na-deoxycholate, 0.5% N-lauroylsarcosine) was added and samples were sonicated (Bioruptor UCD-200 (Diagenode) at 4’C, 8 Cycles set to pulse high 30 seconds, rest 30 seconds) immediately.
Double phenol/chloroform extraction was performed with a final chloroform extraction. DNA was precipitated using Sodium acetate (0.3 M, pH 5.2), 20mg Glycogen and 100% ethanol. DNA was washed with 500 μl ice cold 70% Ethanol. Supernatant was removed and the pellet air-dried. The dried pellet was re-suspended in 50 μl TE buffer (10 mM Tris pH 8.0, 1 mM EDTA in MilliQ water) and incubated at 65’C overnight to reverse crosslinks. Finally 1 μl 10mg/ml RNAseA was added and incubated at 37’C for 1 hour, samples were cleaned using the QiaQuick MinElute kit (Qiagen) and DNA was measure using a Qubit analyzer. Final libraries were prepared as per standard Illumina protocols.
Head complexes were dissected from wandering third instar larvae (500, in batches of 100) and fixed in 1ml crosslinking solution (1.8% formaldehyde, 50 mM HEPES pH 7.9, 1 mM EDTA, 0.5 mM EGTA, 100 mM NaCl in MilliQ water) for 25 minutes at room temperature while rotating. Crosslinking solution was replaced 5 times after 5 minutes each time. Crosslinking was stopped by adding 1ml stop solution (0.01% Triton X-100, 125 mM Glycine in PBS) and incubating at room temperature for 10 minutes while rotating, this was repeated for 3 more washes. Complexes were washed in 1 ml wash A (10 mM HEPES pH7.9, 10 mM EDTA, 0.5 mM EGTA, 0.25% Triton-X100 in MilliQ water) for 10 minutes at room temperature while rotating. Complexes were washed in 1 ml wash B (10 mM HEPES pH7.9, 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.01% Triton X-100 in MilliQ water) for 10 minutes at room temperature while rotating, this was repeated 3 more times. Eye-antennal discs (200) were dissected from head complexes in wash B and collected in a tube containing 100 μl sonication buffer (10 mM HEPES pH7.9, 1mM EDTA, 0.5 mM EGTA in MilliQ water), samples were sonicated (Bioruptor UCD-200 at 4’C, 8 Cycles set to pulse high 30 seconds, rest 30 seconds) immediately. Samples were centrifuged at 21000 xg for 10 minutes at 4’C. For each immunoprecipitation, the pellet from 100 μl extract was re-suspended in 1 ml RIPA buffer (140 mM NaCl, 10 mM Tris-HCl pH8.0, 1 mM EDTA, 1% Triton X-100, 0.1% SDS, 0.1% Na-deoxychoate, 1% PMSF in MilliQ water), an extra sample (10 μl) was kept aside as input.
Immunoprecipitation was performed by adding 20 μl protein A/G magnetic beads (Millipore) and incubating for 1 hour at 4’C; samples were centrifuged at 3000 rpm for 2 minutes and supernatant kept. Anti-CTCF antibody (10 μl crude rabbit serum—A kind gift from Dr. Ranier Renkawitz) was added to the supernatant and rotated at 4’C overnight. Immunocomplexes were recovered by adding 20 μl protein A/G magnetic beads to sample and incubating at 4’C for 3 hours while rotating. Magnetic beads were separated using a magnetic stand and supernatant discarded. Beads were re-suspended and washed for 5 minutes on a rotating platform, with 1 ml of the following buffers in order:- Low salt immune complex wash buffer (Millipore), High salt immune complex wash buffer(Millipore), LiCl immune complex wash buffer (Millipore), TE buffer. ChIP elution buffer (Millipore) was warmed to 38’C and 100 μl was added to the beads, 1μl RNaseA was also added, samples were incubated at 36’C for 30 minutes, shaking at 950rpm. Both immunoprecipitated and control samples had 1ul proteinase K added and were incubated at 62’C for 2 hours, shaking at 950rpm and then 95’C for 10 minutes. Samples were allowed to cool to room temperature and beads were separated with a magnetic stand. Final DNA was purified according to the manufacturers guidelines (MagnaChIP, Millipore) Final libraries were prepared as per standard Illumina protocols.
Raw reads were first cleaned for adapter sequences using fastq-mcf using default parameters and an adapter file containing common Illumina adapter sequences. Cleaned samples were mapped to the Drosophila melanogaster FlyBase release r5.53 genome [59] using bowtie2 [60] (default parameters) and reads with a mapping quality of less than 4 were discarded. In several samples we discovered reads mapping to the Wolbachia genome (60–80%), these were also discarded. Both ATAC-seq and DNaseI-seq reads (publicly available data) were adjusted to better represent the open chromatin by centering reads on the cut-sites and extending this by 5bp on either side, samples were finally sorted and indexed using Samtools [61]. Peaks were called on ATAC-seq, FAIRE-seq and DNaseI-seq samples using the MACS2 software suite [62] with the added parameters “-g dm –nomodel –shiftsize 50 –q 0.01”. For comparison between samples, all peaks from each sample were merged to provide one set of combined peaks. Peaks on ChIP-seq data were called also using the MACS2 software suite with the parameters “-g dm –q 0.01” also using input as a control. For quantification of peaks, bed files of combined peaks were converted into a GFF3 format and then the number of reads per peak, per sample were counted using htseq-count [63].
Finally bigWig files were created from bam files for each sample using genomeCoverageBed [64] (using the –scale option) and bedGraphToBigWig [65]. Scales were determined by the ‘sizeFactors()’ command from DESeq2 [66] on a matrix of all samples, counted on all combined peaks. Genomic locations of peaks was determined by the CEAS software package using default parameters and a prebuilt dm3 gene annotation file [67]. For the CTCF protection analysis motif scanning was performed with Cluster Buster to find occurrences of the JASPAR-MA0139.1 motif in CTCF peaks using the parameters ‘-m7 –c0’, each peak was then re-centered on the center of the best scoring motif present. The cut sites from each read was determined and plotted 500bp around the re-centered CTCF peaks. Signal from the same motif at random regions of DNA was subtracted to remove background noise.
To determine overlaps between peaks, the tool intersectBed was used, with all CTCF peaks or high confidence regions as file A and the ATAC-seq/FAIRE-seq peaks as file B, the option ‘-f 0.4’ was also used to enforce a 40% overlap of the CTCF peak by the ATAC-seq/FAIRE-seq peak.
To identify differential peaks between conditions we used DESeq2 [66] with a p-adjusted cutoff of 0.01. This cutoff was supported by the leading edge of a Gene Set Enrichment Analysis (GSEA [68]) analysis whereby all genes ranked by their most significant peak (tumor vs wild type) are compared to a Ras/Scrib gene signature. To link peaks to genes we assigned a peak within any intron or in the 5kb upstream region of the TSS to the gene. To rank genes according to peak heights (for GSEA analysis) we used the peak with the most significant adjusted p-value. When FAIRE-seq and ATAC-seq are used as replicates, we took batch effects into account in DESeq2.
Rankings for the recovery curves seen in S2 Fig. were generated by scoring peaks (ATAC-seq wild type and FAIRE-seq wild type, merged) based on the number of reads falling within the region for the appropriate samples. ATAC-seq and FAIRE-seq were ranked individually and their recoveries overlaid.
We acquired two Affymetrix Drosophila Genome 2.0 Array data sets from GEO [30], one comprised of 3 wild type and 3 RasV12; scrib-/- biological replicates (GEO accession: GSE42938) [69], and the second comprised of 5 wild type and 5 Upd overexpression biological replicates (GEO accession: GSE15868) [42]. We discarded three low-quality samples from our analyses (GSM398336, GSM398339, GSM398341). Differential expression analysis was carried out in R using the Bioconductor packages affy, limma, Biobase and GEOquery, applying a standard limma protocol [70]. After obtaining differential values we associated each probe to their respective gene; for genes with more than one associated probe, we decided to use the probe with the most significant adjusted p-value.
The differential peaks (wild type control vs RasV12; scrib-/- tumors) were assigned to genes (<5kb upstream or intronic) and for each gene the most significantly differential peak was kept. The genes were ordered based on the relative openness of their assigned peak (opening on top and closing on bottom), to obtain the ranked gene list (x-axis of GSEA plots).
Two groups of differentially expressed genes were determined based on the publically available micro array data (GEO accession: GSE42938). One group contained the significantly upregulated and the other the significantly downregulated genes in the tumors. We used GSEA (with 100000 perturbations) to determine if these groups of differentially regulated genes were significant enriched on either side of the ranked gene list.
The globally opening regions in the RasV12; scrib-/- tumors were determined by differential peak calling between all 5 wild type controls and all 4 RasV12; scrib-/- samples (combining early and late). Of the significantly differential peaks, only those with a logFC greater than 1 were selected, ending up with 4111 globally opening regions.
To determine the early versus late RasV12; scrib-/- regulatory regions, differential peaks were called between the 5 wild type controls and the 2 RasV12; scrib-/- early samples and between the 2 RasV12; scrib-/- early samples and the 2 RasV12; scrib-/- late samples. We selected a subset of regulatory regions that were opening from WT to RSE (with a logFC > 1) and that remained at similar levels between RSE and RSL (-0.2 < logFC < 0.2); we defined these regulatory regions as ‘stably opening’. For the regulatory regions defined as ‘gradually opening’, we selected the regions that are becoming more open between WT and RSE (logFC > 0) and that further open between RSE and RSL (logFC > 0.5). Using a Fisher's Omnibus test we combined the p-values for each regulatory region (one from WT vs RSE and one from RSE vs RSL) and obtain a new chi-squared p-value.
The regulatory regions opening in Upd-overexpression were determined by differential peak calling between the 3 ATAC-seq wild type controls and the 2 ATAC-seq Upd-overexpressing samples. We took the top 250 opening regions (ordered on singed P Value) to perform motif enrichment analysis.
For motif enrichment we used i-cisTarget [32], a tool developed in our lab to discover motifs significantly enriched in our four regulatory region groups (the stably, globally and gradually opening in RasV12; scrib-/- and opening in Upd-overexpression). We ran i-cisTarget via the command line with the rank threshold = 10000, enrichment score threshold = 2 and a collection of 9713 motifs. The enrichment of each motif in the input set is calculated as an area under the recovery curve (AUC), whereby recovery is observed over a genome-wide ranking of 136K a priori defined candidate regulatory regions [32]. The AUC score is normalized by subtracting the mean of all AUCs over all motifs, and dividing it by the standard deviation, to obtain a Normalized Enrichment Score (NES). We use a cutoff of NES > 2.5 to select significantly enriched motifs. Relationships between NES and False Discovery Rates can be found in [71]. For each factor of interest with multiple enriched motifs, we selected the motif with the highest NES score.
ATAC-seq, ChIP-seq and FAIRE-seq data for all conditions are available in GEO (http://www.ncbi.nlm.nih.gov/geo/), with accession number GSE59078.
Genome browser tracks for all data, and called peaks for wild type and cancer-related regulatory regions are all available within a UCSC Genome Browser hub from this URL: http://genome.ucsc.edu/cgi-bin/hgTracks?db=dm3&hubUrl=http://ucsctracks.aertslab.org/ATAC-paper/hub.txt
|
10.1371/journal.pcbi.1004135 | Reconstructible Phylogenetic Networks: Do Not Distinguish the Indistinguishable | Phylogenetic networks represent the evolution of organisms that have undergone reticulate events, such as recombination, hybrid speciation or lateral gene transfer. An important way to interpret a phylogenetic network is in terms of the trees it displays, which represent all the possible histories of the characters carried by the organisms in the network. Interestingly, however, different networks may display exactly the same set of trees, an observation that poses a problem for network reconstruction: from the perspective of many inference methods such networks are indistinguishable. This is true for all methods that evaluate a phylogenetic network solely on the basis of how well the displayed trees fit the available data, including all methods based on input data consisting of clades, triples, quartets, or trees with any number of taxa, and also sequence-based approaches such as popular formalisations of maximum parsimony and maximum likelihood for networks. This identifiability problem is partially solved by accounting for branch lengths, although this merely reduces the frequency of the problem. Here we propose that network inference methods should only attempt to reconstruct what they can uniquely identify. To this end, we introduce a novel definition of what constitutes a uniquely reconstructible network. For any given set of indistinguishable networks, we define a canonical network that, under mild assumptions, is unique and thus representative of the entire set. Given data that underwent reticulate evolution, only the canonical form of the underlying phylogenetic network can be uniquely reconstructed. While on the methodological side this will imply a drastic reduction of the solution space in network inference, for the study of reticulate evolution this is a fundamental limitation that will require an important change of perspective when interpreting phylogenetic networks.
| We consider here an elementary question for the inference of phylogenetic networks: what networks can be reconstructed. Indeed, whereas in theory it is always possible to reconstruct a phylogenetic tree, given sufficient data for this task, the same does not hold for phylogenetic networks: most notably, the relative order of consecutive reticulate events cannot be determined by standard network inference methods. This problem has been described before, but no solutions to deal with it have been put forward. Here we propose limiting the space of reconstructible phylogenetic networks to what we call “canonical networks”. We formally prove that each network has a (usually unique) canonical form—where a number of nodes and branches are merged—representing all that can be uniquely reconstructed about the original network. Once a canonical network N ̂ is inferred, it must be kept in mind that—even with perfect and unlimited data—the true phylogenetic network is just one of the potentially many networks having N ̂ as canonical form. This is an important difference to what biologists are used to for phylogenetic trees, where in principle it is always possible to resolve uncertainties, given enough data.
| Explicit [1] or evolutionary [2, 3] phylogenetic networks are used to represent the evolution of organisms or genes that may inherit genetic material from more than one source. This may be caused by events such as hybrid speciation (e.g. in plants and animals [4, 5]), horizontal gene transfer (e.g. in bacteria [6, 7]), viral reassortment [8], or recombination (e.g. in viruses [9, 10] or in the genomes of sexually reproducing species [11–13]). They are called “explicit” to distinguish them from “implicit” [14], “abstract” [1] or “data-display” [3] phylogenetic networks, which are used to display collections of alternative evolutionary hypotheses supported by conflicting signals in the data. In explicit networks, multiple-inheritance events are represented as reticulations, that is, nodes where two or more lineages converge to give rise to a new lineage, whose genetic material is a combination of that of its direct ancestors.
Explicit networks can be interpreted in terms of classic, tree-like evolution: if we focus on a single, indivisible and thus non-recombining inherited character (for example a single site in a DNA sequence), its history is still best described by a tree. This observation gives rise to the notion of trees displayed by a network, which are all the possible single-character histories implied by a phylogenetic network. (See, e.g., Fig. 1, where T1, T2 and T3 are the trees displayed by networks N1 and N2. Formal definitions are in the Results section.)
Several works in the last few years have focused on the methodology for phylogenetic network inference, and data-display networks in particular have begun to make a real impact on the everyday practice of biologists (e.g., [15–17]). There remains, however, a strong demand for automatic reconstruction of networks that not only display conflicting signals in the data, but also seek to explain these signals with explicit inferences of past reticulation events (see, e.g., [18–20]). This is evidenced, for example, by the abundance of manually reconstructed networks in the literature [8, 21–27]. As a result of this demand, the inference of explicit networks is now a rapidly growing field of research [1].
Some paradigms in the proposed methodology are beginning to emerge. Not surprisingly, the notion of trees displayed by a phylogenetic network plays a central role: the general idea is to evaluate the fit of a network N with the data indirectly—on the basis of how well the trees displayed by N explain the data. In the following, we describe how this applies to the two main approaches for explicit network reconstruction: consistency-based approaches (see [28] for a survey)—seeking a network consistent with a number of prior evolutionary inferences (typically trees or groupings of taxa)—and sequence-based approaches, such as standard formulations of maximum parsimony and maximum likelihood for networks [2, 29–33].
Although evaluating a network via the trees it displays is evolutionarily meaningful, it has a problematic consequence: from the perspective of these reconstruction methods, all networks displaying the same set of trees are “indistinguishable”, as the function that these methods seek to optimize will always assign the same score to all networks displaying the same set of trees, regardless of the input data. In other words, the central parameter of phylogenetic network inference, the network itself, is in some cases not identifiable.
As an example, consider again networks N1 and N2 in Fig. 1, which display the same trees 𝓣(N) = {T1, T2, T3}. (In the following, 𝓣(N) denotes the set of trees displayed by N.) By displaying the same trees, these networks display the same clades, the same triples, the same quartets (triples and quartets are rooted subtrees with 3 leaves and unrooted subtrees with 4 leaves, respectively) and in general the same subtrees with an arbitrary number of leaves. Therefore, any method that reconstructs a network based on its consistency with collections of such data will not be able to distinguish between networks N1 and N2. This includes all the methods whose data consists of clusters of taxa (e.g., [34]), triples (e.g., [35]), quartets (e.g., [36]), or any trees (e.g., [37]).
The same holds for many, sequence-based, maximum parsimony and maximum likelihood approaches proposed in recent papers. For maximum parsimony, a practical approach [2, 29–31] is to consider that the input is partitioned in a number of alignments A1, A2, …, Am, each from a different non-recombining genomic region (possibly consisting of just one site each), and then take, for each of these alignments, the best parsimony score Ps(T∣Ai) among all those of the trees displayed by a network N. The parsimony score of N is then the sum of all the parsimony scores thus obtained. Formally, we have
Ps ( N | A 1 , A 2 , … , A m ) = ∑ i = 1 m min T ∈ 𝓣 ( N ) Ps ( T | A i ) .
It is clear that if two networks display the same set of trees (as in Fig. 1), then their parsimony score with respect to any input alignments will be the same—because they take the minimum value over the same set 𝓣(N)—and thus they are indistinguishable to any method based on the maximum parsimony principle above.
As for maximum likelihood (ML), Nakhleh and collaborators [2, 32, 33, 38] have proposed an elegant framework whereby a phylogenetic network N is not only described by a network topology, but also edge lengths and inheritance probabilities associated to the reticulations of N. As a result, any tree T displayed by N has edge lengths—allowing the calculation of its likelihood Pr(A∣T) with respect to any alignment A—and an associated probability of being observed Pr(T∣N). The likelihood function with respect to a set of alignments A1, A2, …, Am, each from a different non-recombining genomic region, is then given by:
Pr ( A 1 , A 2 , … , A m | N ) = ∏ i = 1 m Pr ( A i | N ) = ∏ i = 1 m ∑ T ∈ 𝓣 ( N ) Pr ( A i | T ) Pr ( T | N ) .
Note that an important difference with the consistency-based and parsimony methods described above is that any tree T displayed by a network has now edge lengths and an associated probability Pr(T∣N).
Unfortunately, this ML framework is also subject to identifiability problems. For example, it does not allow us to distinguish between networks with topologies N1 and N2 in Fig. 1: for every assignment of edge lengths and inheritance probabilities to N1, there exist corresponding assignments to N2 that make the resulting networks indistinguishable, that is, displaying the same trees, with the same edge lengths and the same probabilities of being observed (see the last section in the Supporting Information, S1 Text). As a result, the likelihoods of these two networks will be identical, regardless of the data, and no method based on this definition of likelihood will be able to favour one of them over the other. We refer to S1 Text for a more detailed discussion about networks with inheritance probabilities and likelihood-based reconstruction.
In general, we believe that these identifiability problems affect all network inference methods which seek consistency with unordered collections of sequence alignments or pre-inferred attributes such as clusters, triples, quartets or trees.
In this paper, as in the ML framework above, we adopt networks and trees with edge lengths as the primary objects of our study. The primary motivation for this is that this choice makes our results directly relevant to the statistical approaches for network inference, all of which need edge lengths to measure the fit of a phylogeny with the available data. In addition to ML, these approaches include distance-based and Bayesian methods [39], which are also promising for future work.
However, there is another motivation for our choice: accounting for edge lengths solves some of the identifiability problems outlined above, as in some cases it allows to distinguish between networks with different topologies, which would be otherwise impossible to tell apart. For example, consider the three network topologies in Fig. 2 (top), where taxon o is an outgroup used to identify the root of the phylogeny for a, b and c. These networks show three very different evolutionary histories: in N1 taxon b is the only one issued of a reticulation event—in other words the genome of b is recombinant—whereas in N2 and N3, it is a and c, respectively, that are recombinant. However, N1, N2 and N3 display the same tree topologies—those of T1 and T2—and thus would be indistinguishable to any approach that does not model edge lengths.
If instead edge lengths are accounted for (e.g. in a ML context) and the data supports T1 and T2 with the edge lengths in Fig. 2, then the only network fitting perfectly the data is N2, with the edge lengths in Fig. 2 (bottom right). It is easy to check that N2 now displays T1 and T2 with the shown edge lengths, whereas no edge length assignment to N1 or N3 can make these networks display T1 and T2.
We note that, throughout this paper, as in classical likelihood approaches, edge lengths measure evolutionary divergence, for example in terms of expected number of substitutions per site. No molecular clock is assumed, meaning that we do not expect edge lengths to be proportional to time.
Unfortunately, accounting for edge lengths only solves some of the identifiability problems for phylogenetic networks. Consider networks N1 and N2 in Fig. 3: for any set of edge lengths for N1, there exist an infinity of edge length assignments for N2 that make these two networks display exactly the same set of trees with the same edge lengths. In the following, we say that networks such as N1 and N2 are indistinguishable.
In fact it is not difficult to construct other examples of indistinguishable networks: each time a network has a reticulation v giving birth to only one edge (i.e. with outdegree 1), then we can reduce by Δλ the length of this edge and correspondingly increase by Δλ the lengths of the edges ending in v, without altering the set of trees displayed by the network. Note that this operation, which we refer to as “unzipping” reticulation v, can result in v coinciding with a speciation node or a leaf when Δλ is taken to equal the length of the edge going out of v. For example in Fig. 3, one may fully unzip the two reticulation nodes in N1, thus obtaining the network N′ of Fig. 4. As expected, N1 and N′ display the same set of trees ({T1, T2, T3}) and are thus indistinguishable. What is most interesting in this example is that, if we fully unzip the two reticulations in N2 (the other network in Fig. 3, also displaying {T1, T2, T3}), then we eventually end up obtaining N′ again. As we shall see in the following, this is not a coincidence: the unzipping transformations described above lead to what we call the canonical form of a network; under mild assumptions, two networks are indistinguishable if and only if they have the same canonical form (e.g. N1, N2 in Fig. 3 have the same canonical form N′; formal definitions and statements in the Results section).
Here, we propose to deal with the identifiability issues for phylogenetic networks in the following way: since no data will ever enable any of the standard inference methods described above to prefer a network over all of its indistinguishable equivalents, we propose that these methods should only attempt to reconstruct what they can uniquely identify, that is, networks in canonical form. This is a radical shift, not only for the developers of phylogenetic inference methods, who will see a drastic reduction of the solution space of their algorithms, but also for evolutionary biologists, who should abandon their hopes of seeing a network such as N1 or N2 in Fig. 3 being reconstructed by these inference methods.
Limiting the scope of network reconstruction to topologically-constrained classes of networks has been a recurring theme and an important goal in the literature on phylogenetic networks. Examples of such classes include galled trees [40, 41], galled networks [42], level-k networks [43], tree-child networks [44], tree-sibling networks [45], networks with visible reticulations [1]. Although the ultimate goal should be to establish what can be inferred from biological data, most of the proposed definitions are computationally-motivated: in general the rationale behind these classes is the possibility of devising an efficient algorithm to solve some formalization of the reconstruction problem. None of these definitions claims to have biological significance.
Our goals are more basic: starting from the observation that not all phylogenetic networks are identifiable, since many of them are mutually indistinguishable with most inference approaches, we aim to define a class of networks that is (existence goal) large enough that every phylogenetic network has an equivalent (i.e. indistinguishable) network within this class and (distinguishability goal) small enough that no two networks within this class are indistinguishable. From our standpoint, the computationally-motivated definitions above are at the same time too broad and too restrictive. Too broad, because they determine a set of networks that includes many pairs of indistinguishable networks: for example the three indistinguishable networks in Fig. 2 are all galled trees—and thus belong to every single one of the classes mentioned above (which are all generalizations of galled trees). Too restrictive, because these classes of networks do not include simple networks that it should be possible to reconstruct from real data. For example, Fig. 5a shows a network N with edge lengths that is not tree-sibling, nor has the visible property, and thus is not galled, nor tree-child (for definitions, see [1]), but which in practice should be reconstructible: apart from the lengths of three edges (x, y, z), N is uniquely determined by the trees that it displays (a consequence of the formal results that we will show in the following), meaning that, given large amounts of data strongly supporting each of these (seven) trees with their correct edge lengths, any method for network inference properly accounting for edge lengths (e.g. based on ML) should be able to reconstruct N, or its canonical form N′.
To the best of our knowledge, only three classes of networks have claims of unique identifiability: reduced networks [46, 47], regular networks [48] and binary galled trees with no gall containing exactly 4 nodes [49]. These approaches bear some resemblances to ours, but do not include edge lengths in the definition of a network. Moreover, we argue that these classes of networks are still too narrow to be biologically relevant. We briefly describe and comment these previous works below.
Moret et al. [46] defined notions of reconstructible, indistinguishable and reduced networks that resemble concepts that we will introduce here. Although some of their results were flawed [47, 50], some of the arguments in this introduction are inspired by their paper. Particularly relevant to the current paper is a reduction algorithm to transform a network into its reduced version. (However, the exact definition of the reduced version is unclear: as one of the authors later pointed out [47], “the reduction procedure of Moret et al. [46] is, in fact, inaccurate” and “in this paper we do not attempt to fix the procedure”.) The concept of reduced version is analogous to that of canonical form here, as the authors claim that networks displaying the same tree topologies have the same reduced version (up to isomorphism; Theorem 2 in [46]). This is somehow a weaker analogue of one of our results (Corollary 1); weaker, because it does not claim that, conversely, networks with the same reduced version display the same tree topologies. To have an idea of the difference between our canonical form and the reduced version of Moret and colleagues, in Fig. 6 we compare the canonical form and the reduced version of the same network N1. (N1 and its reduced version are taken from Fig. 15 of [46] to avoid possible issues with the reduction algorithm.) As one can see, the canonical form retains more of the complexity of the original network.
Another reduction procedure on network topologies has been studied by Gambette and Huber [49], who prove that if two network topologies reduce to the same topology, then they must display the same tree topologies. Again, this is analogous to, but somehow weaker than our results, since it only provides a sufficient condition for networks to be indistinguishable (which in their context means to display the same tree topologies). This means that there can be irreducible networks that are indistinguishable (e.g. those in Fig. 2) thus failing to achieve the distinguishability goal. Moreover, Gambette and Huber [49] show that a particular class of network topologies (binary galled trees with no gall containing exactly 4 nodes) are uniquely identified by the tree topologies they display. It is clear that this class is too small to achieve the existence goal.
Finally, a regular network is a network topology N in which, among other requirements, no two distinct nodes have the same set of descendant leaves (see [48] for a formal definition and characterizations). This requirement implies, among other things, that N cannot contain any reticulation v with outdegree 1 (v and its direct descendant would have the same descendant leaves), which in turn implies that regular networks are special cases of our canonical networks (the latter however also specify edge lengths). In fact regular networks satisfy a property that is analogous to the one we prove here for canonical networks: a regular network N is uniquely determined by the tree topologies that it displays [51], meaning that there can be no other regular network N′ displaying exactly the same set of tree topologies. Willson [51] shows this constructively by providing an algorithm that, given the (exponentially large) set of tree topologies displayed by a regular network R, reconstructs R itself. However, unlike for our canonical forms, for a given network there may exist no regular network displaying the same set of trees (e.g. consider the topology of N′′ in Fig. 5b), thus failing to meet the existence goal. Regularity is in fact a very restrictive constraint for a network. For example, none of the networks in Fig. 5 and Fig. 7 is regular, despite the fact that their topologies are uniquely determined by the trees with edge lengths that they display (a consequence of our results further below). Finally, going back to Fig. 6, collapsing the edge above taxa c and d in R(N1) yields the regular network displaying the same tree topologies as N1 and N2. Again, this shows that the canonical form retains more of the complexity of the original network than its regular counterpart.
Our main result consists of formally proving that for every network N there exists a network N′ in canonical form, indistinguishable from N; moreover, if we restrict ourselves to networks satisfying a mild condition (the NELP property below), such canonical form N′ is unique (see Theorem 1). In other words, although in general a phylogenetic network N is not uniquely recoverable from the data it generates, there always exists a canonical version N′ of N that is indeed determined by the data. Informally, N′ is all that can be reconstructed about N.
In order to formally state this result, we here introduce a theoretical framework for explicit phylogenetic networks with branch lengths. A directed acyclic graph (DAG) is a simple directed graph that is free of directed cycles. A DAG is rooted if it contains precisely one node of indegree 0, called the root. All nodes of outdegree 0 in a DAG are called leaves. A weighted rooted phylogenetic network N = (V, E, φ, Λ) on 𝓧 (in this paper also called a network for simplicity) consists of a rooted DAG (V, E) whose leaves are bijectively labeled (via φ:𝓧 → V) with the elements of 𝓧 (called taxa). Moreover, each edge e ∈ E is associated to a set of positive weights, called lengths, Λ(e) ⊂ ℝ>0. Figs. 3, 4, 5 contain examples of networks. A reticulation of a network N is a node v ∈ V with indegree greater than 1. A weighted phylogenetic tree on 𝓧 (a tree for simplicity) is a network on 𝓧 with no reticulations and such that each edge e has a unique length (∣Λ(e)∣ = 1), which we denote by λ(e). Below, we discuss the biological justification of various aspects of the definitions above.
Let v be a node with indegree 1 and outdegree 1 in a tree. Node v is said to be suppressible. Suppressing v means removing the in-edge e = (u, v) and the out-edge f = (v, w) and then creating a new edge g = (u, w) with length λ(g) = λ(e) + λ(f). Let N = (V, E, φ, Λ) be a network on 𝓧. A tree contained in N is a tree T = (V′,E′,φ′,λ) on the same taxon set 𝓧 such that: (1) the roots of T and N coincide, (2) the nodes and edges of T are also nodes and edges of N, that is V′ ⊆ V and E′ ⊆ E, (3) taxon labels are unchanged, that is φ′ = φ, and (4) the edge lengths of T are also edge lengths of N, that is, for every edge e ∈ E′, λ(e) ∈ Λ(e). A tree displayed by N is a tree T′ that can be obtained (up to isomorphism) by suppressing all suppressible nodes from a tree contained in N. The set of trees displayed by N is denoted by 𝓣(N). In Fig. 7, 𝓣 ( N 2 ′ ) is the set of trees isomorphic to T 1 ′ and T 2 ′. Two networks N1 and N2 are said to be indistinguishable if they display the same set of trees, that is 𝓣(N1) = 𝓣(N2). For example, N1 and N2 in Fig. 3 are indistinguishable, as they display the same set of trees (T1, T2 and T3, up to isomorphism).
Definition 1. Given a network N, a funnel is a node with indegree greater than 0 and outdegree 1. A funnel-free network, or canonical network, is a network that does not contain funnels. A canonical form of a network N is a network that is funnel-free and indistinguishable from N.
In Fig. 3, N1 and N2 each contain two funnels, and thus are not funnel-free. The network N′ in Fig. 4 is a canonical form of N1 and N2 in Fig. 3, as N′ is funnel-free and indistinguishable from N1 and N2. Similarly, N 2 ′ in Fig. 6 is a canonical form of N2. Note that nodes with indegree 1 and outdegree 1 are funnels. This implies that for trees the funnel-free condition coincides with the exclusion of suppressible nodes, which is a standard requirement in the definition of phylogenetic trees. It is thus appropriate to view the funnel-free condition as a natural extension of this requisite to networks.
Definition 2. A weighted path in a network N = (V, E, φ, Λ) is a pair (π, λ), where π is a directed path in the graph (V, E) and λ is a function that associates each edge e in π with a length λ(e) ∈ Λ(e). The length of a weighted path is the sum of the lengths assigned to its edges. A network satisfies the NELP (no equally long paths) property if no pair of distinct weighted paths having the same endpoints have the same length.
As we explain below, the NELP property is a mild condition to satisfy, unless edge lengths are taken to represent time. The following result states that if we restrict ourselves to networks satisfying the NELP property, then every network has exactly one canonical form. An outline of its proof can be found in the Methods section, including an algorithm showing how to reduce a network to canonical form. The detailed proof is presented in S1 Text.
Theorem 1. (i) Every network N has a canonical form. Moreover, (ii) if N has the NELP property, then there exists a unique canonical form of N among networks satisfying the NELP property (up to isomorphism).
(The notion of isomorphism between networks is only used for mathematical rigor and is defined in S1 Text.) The following result provides a necessary and sufficient condition for two networks satisfying the NELP property to be indistinguishable.
Corollary 1. Let N1 and N2 be networks with the NELP property and let N 1 ′ and N 2 ′ be their unique canonical forms satisfying the NELP property. Then N1 and N2 are indistinguishable if and only if N1′ and N2′ are the same network (up to isomorphism).
The following result states that a canonical network with the NELP property is uniquely determined by the trees it displays:
Corollary 2. Let N be a canonical network satisfying the NELP property. Then N is the unique (up to isomorphism) canonical network satisfying the NELP property that displays (all and only) the trees in 𝓣(N).
We now discuss the biological significance of a number of technical aspects of our framework.
All the phylogenies considered here—trees or networks—are rooted. This is because we assume that the analysis uses an outgroup (possibly consisting of multiple taxa, and with no reticulations) for rooting. For simplicity, outgroup lineages are not included in our phylogenies (an exception to this is in Fig. 2). Note however that, because our phylogenies have edge lengths, and because omitting the outgroup is just a convention, the omitted lineages must have the same lengths for a network and all the trees it displays. For example, if we wish to omit the outgroup from N2 in Fig. 2 and from the trees that it displays (T1 and T2 in Fig. 2), then what we obtain are N 2 ′ , T 1 ′ and T 2 ′ in Fig. 7. This has a notable consequence: the trees displayed by a rooted network with edge lengths may have a root with outdegree 1 (e.g. T 1 ′ in Fig. 7). For flexibility, we also allow a network to have a root with outdegree 1.
Moreover, we allow multiple lengths for an edge in a network, but not in a tree. For example, in Fig. 6, network N 2 ′ has an edge with two lengths (λ7 + λ12 + λ14 and λ7 + λ11 + λ13 + λ14). The motivation behind multiple lengths lies in the observation that, whereas each edge in a phylogenetic tree describing the evolution of non-reticulating organisms trivially corresponds to a unique evolutionary path in the underlying real evolutionary history, when reticulate events have occurred this is not necessarily true: Fig. 8 and Fig. 9 show that some evolutionary scenarios can either be represented using multiedges (multiple edges with the same endpoints) or edges with multiple lengths. Although these two options are mathematically equivalent, graphically the second one leads to more compact representations, and this is why we choose to allow multiple lengths rather than multiedges. For our purposes we only need to consider the case where e has a finite set of lengths (Λ(e) = {λ1(e), …, λk(e)}).
Another unconventional aspect of our networks is the possibility of having nodes with in-degree and out-degree both greater than one. (See, e.g., the last common ancestor of c and d in N 2 ′ in Fig. 6.) Traditionally, the internal nodes in a phylogenetic network are constrained to belong to one of two different categories: reticulate nodes, with more than one incoming edge and just one outgoing edge, and speciation (or coalescence) nodes, with one incoming edge and multiple outgoing edges. Because reticulate and speciation events are clearly distinct, it is reasonable to constrain internal nodes to only fall in the two categories above. In our framework, this requirement is dropped, and some networks, notably those in canonical form, may have nodes that both represent reticulate and speciation events. In this case, it is important to understand that these nodes represent a potentially complex (and unrecoverable) reticulate scenario, followed by one or more speciation events. Compare, for example, network N and its canonical form N′ in Fig. 5, or N2 and N 2 ′ in Fig. 6. (In the latter, it is especially instructive to consider the reticulate history above the direct ancestor of taxon e.)
We use network N1 of Fig. 3 to illustrate the NELP property. In N1 there are three distinct weighted paths having as endpoints the root of N1 and the direct ancestor of b. The lengths of these paths are ℓ1 = λ1 + λ6, ℓ2 = λ2 + λ3 + λ5 + λ8 and ℓ3 = λ2 + λ10 + λ9 + λ8. Moreover, there is another pair of paths having the same endpoints: those of lengths ℓ4 = λ3 + λ5 and ℓ5 = λ10 + λ9. Thus N1 has the NELP property if and only if the three numbers ℓ1, ℓ2 and ℓ3 are all different (note that this implies that also ℓ4 and ℓ5 are different). If edge lengths are taken to represent evolutionary change, rather than time, this is a very mild requirement: when edge lengths are drawn at random from a continuous distribution, the probability that two paths get exactly the same length is zero.
On the other hand, the NELP property does not hold for phylogenetic networks where edge lengths are taken to represent time. For these networks, canonical forms may not be unique (see Fig. 10 for an example of this). Even in this case, we believe that inference methods should only consider phylogenetic networks in their canonical form, as this allows to reduce the solution space without any loss in “expressive power”: since every network N has (at least one) canonical form that displays exactly the same set of trees—and therefore has the same fit with the data as N—restricting the solution space to canonical forms always leaves at least one optimal network within this space. The real weakness of using canonical forms in a molecular clock context is that if a canonical form is not unique, then it cannot be considered representative of all the networks indistinguishable from it. As an example of this, consider the indistinguishable networks in Fig. 10: none of these is representative of all the others.
Our results are both negative and positive. The bad news is that any method that scores the fit between a network N and the available data—which may be sequences, distances, splits, trees (with or without edge lengths)—based on the set of trees displayed by N must face an important theoretical limitation: regardless of the amount of available data from the taxa under consideration, some parts of the network representing their evolutionary history may be impossible to recover—most notably the relative order of consecutive reticulate events (see, e.g., Fig. 3). The good news is that, when edge lengths are taken into account, we can set precise limits to what is recoverable: the canonical form of a network N is a simplified version of N that excludes all the unrecoverable aspects of N. In a canonical form, reticulate events are brought as forward in time as possible, causing the collapse of multiple consecutive nodes. (Compare again network N2 and its canonical form N 2 ′ in Fig. 6.) The importance of the canonical form N′ of a network N lies in the fact that, if we restrict our consideration to networks with the NELP property, N′ is the unique canonical network consistent with perfect and unlimited data from the taxa in N.
There is an interesting analogy between soft polytomies in classical phylogenetics and collapsed nodes in a canonical network. Both represent lack of knowledge about the order of evolutionary events: speciations or more generally lineage splits in the first case, and reticulate events in the second. However, there is also an important difference between them: whereas in principle polytomies can be resolved by collecting further data from the taxa in the tree (for example, by extensive sequencing of their genomes [52]), the standard network inference methods considered here cannot resolve collapsed nodes in a canonical network, irrespective of the amount of data from the taxa under consideration. This difference is mitigated by the observation that increased taxon sampling may indeed permit to resolve the collapsed nodes, when the new lineages break adjacencies between reticulate nodes. However, such lineages may not always exist or they may be difficult to sample.
The present work has several consequences that should be of interest both to the biologists concerned by the use of methods for phylogenetic network inference, and to the researchers interested in the development of these methods. We illustrate these consequences starting from a well-known problem of network inference methods, that of multiple optima. It has been noted before that many of the inference methods that have been recently proposed—especially those solely based on topological features—often return multiple optimal networks: Huson and Scornavacca show a striking example of this (Fig. 2 in [53]), where the problem of finding the simplest network displaying two given tree topologies admits at least 486 optimal solutions.
The existence of multiple optimal networks for a given data set is essentially due to two reasons: insufficient data and non-identifiability. For the example of 486 optimal solutions, this large number may be partly due to the fact that the goal was to achieve consistency with only two tree topologies. More data may enable to discriminate among the 486 returned networks. Non-identifiability, which occurs when none of the allowed data can discriminate between two or more networks, is a more serious problem than insufficient data, as it cannot be solved by simply increasing the size of the input sample. Another interesting example appears in a paper by Albrecht et al. [54], which we reproduce here in Fig. 11. Here, there are only three optimal networks, essentially differing for which of the three clades {A.bicornis, A.longissima, A.sharonensis}, {A.uniaristata, A.comosa} and {A.tauschii} is considered as a hybrid (in this example reticulations represent hybridizations). This pattern is entirely analogous to that of the three networks in Fig. 2 (with a, b and c replaced by the three clades above), meaning that these three networks are indistinguishable to methods not accounting for edge lengths. Therefore, in this example, the existence of multiple optimal solutions is entirely due to non-identifiability.
All this motivates three recommendations:
Correspondingly, we recommend that edge lengths should be accounted for in the analyses (point 1) and, for each of the indistinguishable classes resulting from this choice, we identify a canonical network that, for all practical purposes, can be considered to be unique. Most important to the end users, we propose that a canonical network N ̂ is what should be given as the result of the inference, with the caveat that N ̂ is a way to represent a class of networks that are all equally supported (point 2). In a canonical form N ̂, the aspects that are not common to all networks in this class are collapsed, as described above. This will help the evolutionary biologist to locate the uncertainties in the phylogeny, and possibly to choose further taxa to resolve them. Finally, we propose that inference methods only attempt to search among—or construct—phylogenetic networks in their canonical form (point 3).
We note that accounting for yet more characteristics of the data may reduce (or eliminate altogether) the identifiability issues for phylogenetic networks. In the case of sequence-based methods, one may take into account the natural order of sites within a sequence [11–13, 55, 56]. Similarly, for reconstruction methods based on collections of subtrees, one could observe and use the relative position of the different genomic regions supporting the input trees. However, these relative positions must be conserved across the genomes being analyzed, a condition which may hold for recombining organisms (e.g. individuals within a population or different viral strains), but which is not obvious when studying a group of taxa that have undergone reticulate events (e.g., hybridization) at some point in a distant past.
The main conclusion of the present study is the following: unless one abandons any optimization criterion that scores a network solely based on the trees it displays, the reconstruction should be carried out in a reduced space of networks: that of the canonical forms defined here. The motivation for this lies in the fact that canonical networks are guaranteed to be uniquely determined, if sufficient data are available. Once a canonical form N ̂ is inferred, it must be kept in mind that even assuming that the inference is free of statistical error, the true phylogenetic network is just one of the many networks having N ̂ as canonical form. Compared to what biologists are used to for phylogenetic trees—where in principle it is always possible to resolve uncertainties—it is clear that this requires an important change of perspective.
The following three subsections describe the proofs of Theorem 1 part (i), of Theorem 1 part (ii), and of their corollaries, respectively. In the case of Theorem 1 part (ii), only the gist of the proof is provided here. The proof in full detail is deferred to S1 Text.
In order to prove that any network N has a canonical form, we describe an algorithm to transform N into a canonical network indistinguishable from N. The algorithm simply consists of repeatedly applying to N = (V, E, φ, Λ) one of the following two reduction rules, until neither can be executed (see Fig. 12):
Funnel suppression (R1). Given a funnel v with k ≥ 1 in-edges (u1, v), (u2, v), …, (uk, v) and out-edge (v, w), remove v and all these edges from N and introduce k new edges (u1, w), (u2, w), …, (uk, w). For all i ∈ {1, 2, …, k} assign to (ui, w) the lengths Λ((ui, w)): = Λ((ui, v)) + Λ((v, w)), where the sum of two sets of numbers A and B is defined as A + B = {a + b: a ∈ A, b ∈ B}.
Multiedge merging (R2). Given a collection of multi-edges (u, w) with multiplicity k and lengths Λ 1 ′ , Λ 2 ′ , … , Λ k ′, replace these edges with a single edge with lengths ⋃ i = 1 k Λ i ′.
An example of the reduction of a network to its canonical form is shown in Fig. 13. Note that, even if the algorithm may temporarily produce multi-edges, the network produced in the end obviously does not have any multi-edge (otherwise we could still apply rule R2).
Proof of part (i) of Theorem 1. We must prove that any network N = (V, E, φ, Λ) has a canonical form. For this, we apply the reduction algorithm described above, thus obtaining a sequence N0 = N, N1, …, Nm, where each Ni+1 is obtained from Ni by applying either R1 or R2. Neither R1 nor R2 can be applied to Nm. We prove that Nm is a canonical form of N. Although, strictly speaking, Ni may not be a network (as it potentially contains multi-edges), the notion of trees displayed by Ni, and thus that of indistinguishability, trivially extends to these multigraphs.
First, note that the algorithm terminates after a finite number of iterations (m). This is true because at each iteration the size of E is reduced by at least one. Moreover, the resulting network Nm is funnel-free, since no reduction of type R1 can be applied to it.
What is left to prove is that Nm is indistinguishable from N = N0. To this end we prove that, at each iteration, Ni and Ni+1 are indistinguishable, i.e. 𝓣(Ni) = 𝓣(Ni+1). In other words any tree T is displayed by Ni if and only if T is displayed by Ni+1.
Let T be displayed by Ni. Then T can be obtained by suppressing all suppressible nodes from a tree Ti contained in Ni. We consider three cases. (1) If none of the edges in Ti is involved in the reduction transforming Ni into Ni+1, then clearly Ti is still contained in Ni+1 and thus T is still displayed by Ni+1. (2) If Ti is involved in a R1 reduction, then it contains a funnel v and it contains one of the in-edges of the funnel, say (uj, v), with length λj ∈ Λj = Λ((uj, v)), along with the out-edge (v, w), with length λ0 ∈ Λ0 = Λ((v, w)). Now, let Ti+1 be the tree obtained from Ti by suppressing the suppressible node v and thus creating a new edge (uj, w) with length λj + λ0. Because the R1 reduction creates a new edge (uj, w) with length set Λj + Λ0, containing the value λj + λ0, then Ti+1 is contained in Ni+1. Moreover, it easy to see that T can still be obtained by suppressing all suppressible nodes from Ti+1. Thus T is still displayed by Ni+1. (3) If Ti is involved in a R2 reduction, then it contains one of the edges of a multi-edge (u, w), with a length λ belonging to one of the length sets Λ 1 ′ , Λ 2 ′ , … , Λ k ′ associated to the k copies of (u, w). Thus we have that λ ∈ ⋃ i = 1 k Λ i ′, which implies that Ti is still contained in Ni+1 and thus T is still displayed by Ni+1. This concludes the proof of 𝓣(Ni) ⊆ 𝓣(Ni+1).
In order to prove that, conversely, 𝓣(Ni+1) ⊆ 𝓣(Ni), one can proceed in a similar way as above: if T is displayed by Ni+1, then T can be obtained by suppressing all suppressible nodes from a tree Ti+1 contained in Ni. By considering three cases analogous to the ones above regarding the involvement of Ti+1 in the reduction transforming Ni into Ni+1, we can prove that in all these cases T is already displayed by Ni. Thus Ni and Ni+1 are indistinguishable, which concludes our proof. □
We note informally that the order of application of the possible reductions in the algorithm above is irrelevant to the end result. To see this, it suffices to show that if two different reductions are applicable to a network, then the result of applying them is the same irrespective of the order of application. As we do not need this remark for the other results in this paper, we do not give a formal proof of it.
Lemma 1. Let N be a network and N′ a canonical form of N obtained by applying the reduction algorithm. If N satisfies the NELP property, then N′ satisfies the NELP property.
Proof: We prove that for each basic step of the reduction algorithm—transforming Ni into Ni+1 via a reduction rule R1/R2—if Ni satisfies the NELP property, then Ni+1 also satisfies it. Suppose the contrary; then, Ni+1 contains two distinct weighted paths ρ1, ρ2 with the same endpoints u and v and same lengths. Because R1/R2 cannot create new nodes, u and v are also nodes in Ni. Moreover, it is easy to see that each weighted path ρ in Ni from u to v gives rise to exactly one weighted path f(ρ) in Ni+1 from u to v, with exactly the same length as ρ. Now take two weighted paths in Ni, one in the preimage f−1(ρ1) and the other in the preimage f−1(ρ2). These two weighted paths in Ni are distinct (as ρ1 ≠ ρ2), have the same endpoints (u and v) and the same length. But then Ni violates the NELP property, leading to a contradiction. We thus have that if Ni satisfies the NELP property, then Ni+1 also satisfies it. By iterating the argument above for each step in the reduction algorithm, the lemma follows. □
The proof of Theorem 1, part (ii), is rather technical. In this section, we introduce a number of new concepts and state the main intermediate results that are necessary to obtain this result. We leave their detailed proofs to S1 Text, together with the obvious definitions of basic concepts such as that of isomorphic networks, sub-network and union of two networks.
Definition 3. (Root-leaf path, prefix, postfix, wishbone, crack.) Let N be a network on 𝓧 and (π, λ) be a weighted path in N from the root of N to a leaf labelled by x ∈ 𝓧. Now consider the sub-network P = (V(π), E(π), φ∣{x}, λ) on {x} consisting of all the nodes and edges in π and associated labels. Any sub-network of N such as P is called a root-leaf path of N. Given a root-leaf path P and a node v belonging to it, any weighted path formed by all the ancestors [descendants] of v in P is a prefix [suffix] of P. Note that a prefix [suffix] only consists of one node when v is the root [leaf] of P. A wishbone of N is any sub-network of N formed by taking the union of two root-leaf paths that have in common only a prefix. A crack of N is any sub-network of N formed by taking the union of two root-leaf paths that have in common only a prefix and a suffix.
Fig. 14 illustrates the definitions above. Note that any root-leaf path P is both a wishbone and a crack, as P is the result of the union of P with itself, and P has a common prefix and a common suffix with P. Moreover, any sub-network R that can be obtained from a root-leaf path by attributing two lengths to one of its edges e is a crack. Finally, note that wishbones and cracks are networks, and thus the notion of isomorphism (Definition 5 in S1 Text) can be applied to them.
The proof of part (ii) in Theorem 1 depends on two important results (Propositions 1 and 2 below), whose proofs can be found in S1 Text. The first states that a network with the NELP property is uniquely determined by the wishbones and cracks it contains.
Proposition 1. Two networks N1 and N2 with the NELP property are isomorphic if and only if they contain the same wishbones and cracks (up to isomorphism).
Proposition 1 is interesting on its own as it suggests an enumerative algorithm to verify whether two networks with the NELP property are isomorphic. Unfortunately this algorithm would be impractical, as the number of wishbones (or cracks) in a network is not polynomial in the size of the network. Also note that we require N1 and N2 to satisfy the NELP property because there exist non-isomorphic networks containing the same wishbones and cracks: for example the networks in the bottom line of Fig. 10. The second result that we need is the following:
Proposition 2. Let N1 and N2 be two indistinguishable funnel-free networks, satisfying the NELP property. Then they contain the same wishbones and cracks (up to isomorphism).
Proof of part (ii) of Theorem 1. Let N be a network with the NELP property and N′ a canonical form of N obtained by applying the reduction algorithm. By Lemma 1, N′ satisfies the NELP property. Now suppose that there exists another canonical form of N, called N′′, satisfying the NELP property. By transitivity, N′ and N′′ are indistinguishable. Because N′ and N′′ are indistinguishable, funnel-free and with the NELP property, N′ and N′′ must contain the same wishbones and cracks (because of Proposition 2). But then, because of Proposition 1, N′ and N′′ are isomorphic. □
We note that some of our arguments in S1 Text lead us to conjecture that a funnel-free network satisfying the NELP property cannot be indistinguishable from a funnel-free network violating the NELP property. This claim would allow us to simplify the statement of Theorem 1: networks with the NELP property would be guaranteed to have a unique canonical form (not just among networks with the NELP property, but among all networks). Unfortunately, to this date, we were unable to prove this conjecture. Nonetheless, note that the reduction algorithm returns, for any network with the NELP property, its unique canonical form with the NELP property (by Lemma 1).
It remains to prove the two corollaries at the end of the Results section. The first one states that two networks N1 and N2 satisfying the NELP property are indistinguishable if and only if their unique canonical forms with the NELP property, N 1 ′ and N 2 ′ respectively, are isomorphic. By Lemma 1, N 1 ′ and N 2 ′ can be obtained by applying the reduction algorithm to N1 and N2.
Proof of Corollary 1. The if part trivially follows from the transitivity of indistinguishability. As for the only if part, note that (again by transitivity) N 1 ′ is indistinguishable from N2. As it is also funnel-free, N 1 ′ is a canonical form of N2. Because N2 can only have one canonical form satisfying the NELP property (by Theorem 1 (ii)), N 1 ′ and N 2 ′ must be the same network (up to isomorphism). □
As for Corollary 2, we recall that it states that a canonical network N with the NELP property is uniquely determined by the trees it displays.
Proof of Corollary 2. Let N and N′ be indistinguishable canonical networks satisfying the NELP property. Then, N and N′ are both canonical forms of N satisfying the NELP. But then, by Theorem 1(ii), N and N′ must be the same network (up to isomorphism). □
|
10.1371/journal.pgen.1005818 | Innovation of a Regulatory Mechanism Modulating Semi-determinate Stem Growth through Artificial Selection in Soybean | It has been demonstrated that Terminal Flowering 1 (TFL1) in Arabidopsis and its functional orthologs in other plants specify indeterminate stem growth through their specific expression that represses floral identity genes in shoot apical meristems (SAMs), and that the loss-of-function mutations at these functional counterparts result in the transition of SAMs from the vegetative to reproductive state that is essential for initiation of terminal flowering and thus formation of determinate stems. However, little is known regarding how semi-determinate stems, which produce terminal racemes similar to those observed in determinate plants, are specified in any flowering plants. Here we show that semi-determinacy in soybean is modulated by transcriptional repression of Dt1, the functional ortholog of TFL1, in SAMs. Such repression is fulfilled by recently enabled spatiotemporal expression of Dt2, an ancestral form of the APETALA1/FRUITFULL orthologs, which encodes a MADS-box factor directly binding to the regulatory sequence of Dt1. In addition, Dt2 triggers co-expression of the putative SUPPRESSOR OF OVEREXPRESSION OF CONSTANS 1 (GmSOC1) in SAMs, where GmSOC1 interacts with Dt2, and also directly binds to the Dt1 regulatory sequence. Heterologous expression of Dt2 and Dt1 in determinate (tfl1) Arabidopsis mutants enables creation of semi-determinacy, but the same forms of the two genes in the tfl1 and soc1 background produce indeterminate stems, suggesting that Dt2 and SOC1 both are essential for transcriptional repression of Dt1. Nevertheless, the expression of Dt2 is unable to repress TFL1 in Arabidopsis, further demonstrating the evolutionary novelty of the regulatory mechanism underlying stem growth in soybean.
| Similar to the “green revolution” semi-dwarf cereals, semi-determinate soybean varieties are lodging-resistant and particularly suitable for planting in high fertility and irrigated environments. Nevertheless, molecular mechanisms underlying semi-determinate stem growth have not been deciphered in any flowering plants. We demonstrate that semi-determinacy is originated from an innovation of spatiotemporal expression of an ancient MADS-box gene and consequent changes of spatiotemporal expression of its interacting genes in soybean, which occurred post-domestication of soybean and selected by breeding. The findings from this study not only provides new insights into the evolutionary novelty of molecular mechanisms regulating stem growth habit reshaped by artificial selection, but also exhibited potential application of such an innovative mechanism for molecular design of stem architecture in other crops towards enhanced adaptability and yield potential.
| Stem growth habit is an important morphological and adaptation trait in flowering plants, which is primarily shaped by regulatory processes converting the vegetative shoot apical meristems (SAMs) that form leaves to the inflorescence meristems (IMs) and then floral meristems (FMs) that form flowers [1–3]. Such processes have been best studied in Arabidopsis [4, 5]. Upon floral induction by both environmental signals (e.g., day length, light quality, and temperature) and endogenous cues (e.g., age and hormone), the primary shoot meristems of Arabidopsis begin to produce the determinate IMs on its flanks, where the floral identity genes such as LEAFY (LFY) and APETALA1 (AP1) are expressed to develop flowers [6–8]. However, the SAMs in the center of the stem tips sustain the indeterminate growth due to the spatial expression of a floral repressor Terminal Flower1 (TFL1) [1, 9, 10], which represses the expression of LFY and AP1 and thus prevents the formation of FMs [11–12]. As a result, the wild-type (TFL1) Arabidopsis produces the indeterminate apical stems that grow indefinitely. By contrast, LFY and AP1 are expressed in the center of stem tips of the loss-of-function tfl1 mutants to give rise to determinate growth, with terminal flowers on the stem tips [1, 8, 11, 13, 14]. In addition, both LFY and AP1 are able to bind to the TFL1 locus to suppress its expression in floral meristems [15, 16].
Although the full functions of TFL1 in Arabidopsis remain to be elucidated, it is documented that the putative orthologs of TFL1 are widely conserved among diverse plant species including many leguminous and solanaceous species, and in particular, their roles as floral repressors, such as Dt1 in soybean (Glycine max) [17, 18], PvTFL1y in common bean (Phaseolus vulgaris) [19], Det in pea (Pisum sativum) [20], Sp in tomato (Solanum lycopersicum) [21], and CaSP in peppers (Capsicum annuum) [22], in producing indeterminate stems are conserved. In general, the wild progenitor species of these individual crops and the majority of the cultivated varieties from these species exhibit indeterminate stem growth. Nevertheless, determinate growth habit in these crops was also selected through domestication or modern breeding, and adapted to specific eco-regions for agricultural production [23–25].
The determinate soybean varieties rose originally from independent human selections of four distinct single-nucleotide substitutions in the Dt1 gene during soybean domestication from its wild progenitor Glycine soja, each of which led to a single amino acid change that resulted in a recessive dt1 allele specifying determinate stem growth [17]. In general, determinate soybean cultivars have distinctly separate vegetative and reproductive stages and are relatively late maturing and grown in the southern eco-regions of both the United States and China. By contrast, the indeterminate cultivars have more overlap of vegetative growth with reproductive development, providing better adaptation to shorter growing season in the north [26]. In addition to these two major types of stem growth habit, semi-determinate cultivars, which produce stems with terminal racemes similar to those observed in determinate cultivars but show an intermediate phenotype have been developed through breeding in the past few decades and deployed for production in the north. While semi-determinate cultivars usually produce slightly fewer stem nodes than indeterminate cultivars, the former are somewhat shorter than the latter and thus provide some degree of lodging resistance that is desirable for production in the high fertility and irrigated environments [27], representing an alternative for enhancement of soybean yield potential, similar to that achieved by the “green revolution” gene in cereals.
Classic genetic analysis demonstrated that semi-determinacy in soybean is specified by a dominant allele, designated Dt2, in the Dt1 genetic background [23]. As shown in S1 Table, the dt2dt2;Dt1Dt1 and Dt2Dt2;Dt1Dt1 genotypes produce indeterminate and semi-determinate plants respectively, whereas both the dt2dt2;dt1dt1 and Dt2Dt2;dt1dt1 genotypes produce determinate plants, indicating a recessive epistatic effect of dt1 on the Dt2/dt2 locus. Semi-determinate stem growth habit was also observed and genetically investigated in tomato [21, 28–30] and two other leguminous crops, pigeon pea (Cajanus cajan) [24] and chickpea (Cicer arietinum) [25]. However, unlike observed in soybean, semi-determinacy in tomato is specified by a recessive allele sdt in the recessive sp genetic background, and the dominant allele Sp, the functional equivalent of TFL1/Dt1, exhibits dominant epistatic effect on the Sdt/sdt locus [21, 28–30] (S1 Table). More intriguingly, the legume crops pigeon pea and chickpea, two close relatives of soybean, both show an inheritance pattern of stem growth habit and a digenic epistasis distinct from observed in soybean but similar to observed in tomato [24, 25]. A more recent study demonstrated that the genetic control of stem growth habit in pea is also distinct from observed in soybean [31] (S1 Table). These observations reflect the uniqueness and evolutionary novelty of genetic control of stem growth habit in soybean.
Recently, Dt2 has been isolated by a map-based cloning approach using a segregating population derived from a high-yielding semi-determinate elite soybean cultivar NE3001 and a high-yielding indeterminate elite cultivar IA3023 [32] (S1 Fig). Dt2 was demonstrated to be a dominant gain-of-function MADS-domain factor gene belonging to an AP1/SQUAMOSA subfamily that includes floral identity genes AP1, CAULIFLOWER (CAL), FRUITFUL (FUL) in Arabidopsis [33–35]. It was also found that the causative mutation that converting dt2 into Dt2 is located in the non-coding regulatory region of the gene. Quantitative real time-polymerase chain reaction (qRT-PCR) analysis revealed that Dt2 is primarily expressed in the stem tips at vegetative 2 (V2) stage, when the first trifoliate leaflets at node 2 are fully expanded but the second trifoliate leaflets at node 3 are not yet unfolded. It was proposed that, at this stage, floral induction occurs in all meristems (apical and lateral), abruptly in the case of the determinants, less abruptly in the case of semi-determinants, but not in the terminal apical meristems in indeterminants, suggesting the essential role of Dt2, as a floral activator, in promoting terminal flowering with the presence of Dt1. However, except of the observed phenotypic epistasis, it is not yet known how this recently selected dominant gain-of-function Dt2 allele interacts with Dt1 and other genes to modulate the semi-determinate growth habit. Here, we report molecular dissection of the Dt2-mediated molecular mechanism regulating stem growth habit in soybean, with an emphasis on the evolutionary novelty of the regulatory pathways reshaped by artificial selection.
The expression patterns of the Dt1 and Dt2 loci in the main stem tips of NE3001 (Dt2Dt2;Dt1Dt1) and IA3023 (dt2dt2;Dt1Dt1) have been previously examined by qRT-PCR [32] (Fig 1A). The expression level of either the Dt2 or dt2 allele was increased from the V0 (when the cotyledons at node 0 are fully extended but the unifoliate leaflets at node 1 are not yet unrolled) to V2 stages and then decreased at the V3 stage (when the second trifoliate leaflets are fully expanded but before the third trifoliate leaflets are still unrolled). By contrast, the expression level of Dt1 in either the Dt2 or dt2 backgrounds was consistently reduced over these developmental stages. In addition, the expression level of Dt1 in the Dt2 background was lower than detected in the dt2 background. Such an expression pattern, particularly at the V2 stage, together with the epistatic interaction between the two genes as deduced from the phenotypes [23], suggest that Dt2 may be a transcriptional repressor of Dt1. However, because the apical meristems only made up a small portion of the main stem tips, the relative abundance of the transcripts from the two genes in apical meristems could not be precisely reflected by qRT-PCR analysis. Therefore, how the difference in levels of Dt2 and dt2 expression determines indeterminate or semi-determinate stems was not understood.
To further elucidate the effects of the Dt2 expression on the transcription of Dt1, we analyzed the expressional changes of Dt1 under ectopic expression of Dt2 driven by the Cauliflower Mosaic Virus (CaMV) 35S promoter. As shown in Fig 1B, the expression level of Dt1 in stem tips at the V2 stage was significantly reduced upon the ectopic expression of the transgene Dt2 in the Throne (dt2dt2;Dt1/Dt1) genetic background, which resulted in an conversion from indeterminate stems to semi-determinate stems [32].
Because the expression level of Dt1 is extremely low relative to that of Dt2 or dt2 in the stem tips [32], it is quite difficult to accurately determine the extent of repression of the Dt1 expression. Previous work has demonstrated that Dt1 is expressed at the highest level in soybean roots [17], where both the Dt2 and dt2 alleles are expressed at extremely low levels [32]. Thus, in the soybean root system, there appear to be little or no effects of the native Dt2/dt2 locus on the expression of Dt1, making the system ideal for investigation of the effect of Dt2 on the expression of Dt1. Using an indeterminate soybean cultivar Kefeng 1 (dt2/dt2;Dt1/Dt1), we generated the Dt2 over-expression transgenic hairy roots. As shown in Fig 1C, the Dt2 transgene driven by the 35S promoter in the transgenic hairy roots was expressed at a level ~30 times higher than the native dt2. By contrast, the expression level of the native Dt1 was reduced ~10 times in the Dt2 transgenic roots, indicating that the level of the Dt2 expression is a key factor controlling Dt1 expression.
Since Dt2 is a MADS-box domain transcription factor (TF) localized in nucleus [32], we wondered whether Dt2 could directly interact with the Dt1 promoter to inhibit the transcription of Dt1. To this end, we created a fusion of the Dt2 protein to the hormone-binding domain of the rat glucocorticoid receptor (GR), under the control of the constructive 35S promoter (Pro35S:Dt2-GR), and the construct was transformed into the hairy roots of the soybean cultivar Kefeng 1. To determine the effect of Dt2 activation on the expression of Dt1 in the Pro35S:Dt2-GR transgenic roots, we treated the roots with the steroid hormone dexamethasone (DEX), the protein synthesis inhibitor cycloheximide (CHX) or both and then measured the changes of Dt1 expression by qRT-PCR, using non-transgenic hairy roots as a control. As shown in Fig 2A, the level of the Dt1 mRNA was reduced significantly after the treatment with DEX, indicating that DEX activation of the Dt2-GR fusion protein by nuclear translocation [13, 36] resulted in repression of Dt1. The level of Dt1 mRNA was also reduced significantly in the presence of both DEX and CHX, but not reduced in the presence of CHX, which has been proven to be able to terminate de novo protein synthesis [13, 36], suggesting that Dt2 can transcriptionally repress Dt1 directly without the requirement for protein synthesis.
As MADS box domains are generally able to bind to DNA sequences of high similarity to the motif CC[A/T]6GG termed the CArG-box [37], we first examined the Dt1 sequence and its flanking sequences from NE3001 and identified five putative CArG-boxes within ~2kb upstream of the Dt1 coding sequence (CDS) and one putative CArG-box at ~1.5kb position downstream of the CDS (Fig 2D). To determine whether Dt2 directly binds to the CArG-box sequences in the regulatory region of Dt1, we performed chromatin immunoprecipitation (ChIP). We first raised a Dt2-specific antibody, anti-Dt2 (Fig 2B) based on a highly unique peptide composed of 19 amino acids from Dt2 (S2 Fig), according to the soybean reference genome sequence [38], and its specificity was further indicated by a substantially higher level of the Dt2 protein detected in the stem tips of NE3001 than that of dt2 detected in IA3023 at the V2 stage (Fig 2C). We then used the anti-Dt2 antibody to enrich DNA fragments bound by Dt2 in NE3001 and then measured the relative enrichment by quantitative PCR (qPCR). As shown in Fig 2D, fragments containing the 1st, 2nd, and 5th putative CArG-boxes, respectively, were enriched by >5–9 fold compared with the control DNA fragment amplified from the soybean ATP binding cassette transporter gene Cons4 [39] in the same genome. By contrast, the 3rd and 4th putative CArG-boxes, and the 7th one downstream of the Dt1 gene were not enriched compared with the control, indicating that these three of the six CArG-boxes are recognized and bound by Dt2. These results were consistent with the observations from the electrophoretic mobility-shift assay (EMSA) analysis (Fig 2E), which reveals that only 1st, 2nd, and 5th CArG-boxes can be bound by a Dt2-6×His fusion protein isolated from an Escherichia coli strain BL21 (Fig 2B). The essentiality of the three CArG boxes for the repression of Dt1 transcription by Dt2 was further demonstrated by the observed ineffectiveness of the repression activity upon the truncation of the Dt1 promoter region involving these CArG-boxes or point mutations within each of the three CArG boxes using a luciferase (LUC) as a reporter (Fig 2F and 2G). Together, these observations suggest that Dt2 functions as a repressor of Dt1 expression by binding directly to the three CArG boxes in the Dt1 promoter region through its MADS-box domain, and that all these three CArG boxes, bound by Dt2, are essential for repression of the activity of the Dt1 promoter.
In addition to the MADS-box domain that binds to the three CArG-boxes in the Dt1 promoter region, a Keratin (K)-box domain, I domain, and a C-terminal or C domain were predicted in the Dt2 protein based on the homolog searches against the conserved domain database (Fig 3A). Compared with MADS-box domains, K-domains are generally less conserved and often involved in protein-protein interactions to form heterodimers for performing their functions [40–43]. It was reported that the C-domains of some MADS-box factors are also important for translational regulation [13]. To test if the K-domain and C-domian in Dt2 are required for fulfillment of the Dt2 function of repressing Dt1 expression, we investigated the effects of constitutive expression of the intact Dt2 protein, an incomplete Dt2 protein without the K-domain (Dt2ΔK), and an incomplete Dt2 protein without a C terminal domain (Dt2ΔC), driven by the 35S promoter, respectively (Fig 3A), on the activity of the Dt1 promoter in Arabidopsis using GUS as a reporter. As shown in Fig 3B and 3C, the activity of the Dt1 promoter was inhibited under the constitutive expression of Dt2, was partially inhibited under the constitutive expression of Dt2ΔC, and was not inhibited under the constitutive expression of Dt2ΔK. By contrast, the expression levels of Dt2, Dt2ΔC, Dt2ΔK did not show obvious difference in levels of expression under the control of the 35S promoter (Fig 3D). These observations suggest that the K-domain is essential, and perhaps, so are its interacting proteins, for the Dt2 function of repressing Dt1 expression.
We then carried out Yeast Two Hybrid (Y2H)-screening of a cDNA library constructed with V2-stage stem tips of soybean using Dt2 as the bait, and identified eight unique cDNA clones each with an insert from a soybean gene (S2 Table), including a putative orthologs of the Arabidopsis SUPPRESSOR OF OVEREXPRESSION OF CONSTANS 1 (SOC1), dubbed GmSOC1. SOC1 in Arabidopsis encodes a MADS-domain factor protein, which integrates multiple flowering signals derived from photoperiod, temperature, hormone, and age-related signals [44, 45]. However, similar to AP1, SOC1 is not expressed in the main shoot of Arabidopsis to maintain the stem’s indeterminate growth [2]. The interaction between Dt2 and GmSOC1 was further validated by Y2H (Fig 4A) and bimolecular florescence complementation (BiFC) using leaf cells of Nicotiana benthammiana (Fig 4B). The interaction signals between Dt2ΔC and GmSOC1 were detected by BiFC, whereas no interaction signals between Dt2ΔK and GmSOC1 were detected. Because both the Dt2ΔK and GmSOC1 proteins were expressed at substantially high levels (Fig 4C), the lack of interaction signals between Dt2ΔK and GmSOC1 would be indicative of the lack of interaction between the two proteins. These results are consistent with the ineffectiveness of Dt2ΔK expression on suppression of Dt1 expression, as illustrated in Fig 3, suggesting that Dt2 interacts with GmSOC1 via its K-domain, and that GmSOC1 is important for fulfillment of the Dt2 function. As a MADS-domain factor, GmSOC1, as expected, was localized to the nucleus (Fig 4D).
To test whether GmSOC1 interacts with Dt1, a GmSOC1-GR fusion protein was created and expressed in the hairy roots of Kefeng 1 directed by the 35S promoter to evaluate the effect of GmSOC1 activation in the GmSOC1-GR fusion protein on the expression of Dt1. The levels of Dt1 mRNA was reduced significantly in the presence DEX, and in the presence of DEX and CHX, but was not changed significantly in the presence of CHX, compared with the transgenic roots without any treatment (Fig 5A), suggesting that GmSOC1 was involved in repression of Dt1 transcription by direct binding to the promoter region of Dt1. Nevertheless, the effect of constructive expression of GmSOC1 on repression of the Dt1 promoter activity is not as strong as the effect of constructive expression of Dt2 on repression of the Dt1 promoter activity (Fig 5C and 5D). This is also consistent with the observed effects of GR-fusion proteins on Dt1 expression (Figs 2A and 5A).
A GmSOC1-6×His fusion protein was isolated from the Escherichia coli strain BL21 and used to examine whether the MADS-domain of GmSOC1 can bind to any of the CArG-boxes in the promoter region of Dt1 by EMSA. As shown in Fig 5B, only binding of GmSOC1 with the 5th CArG box in the Dt1 promoter region was detected.
Expression analysis by qRT-PCR using the stem tips from NE3001 and IA3023 revealed consistent expression patterns between GmSOC1 and Dt2 in NE3001 from the V0 stage through the V5 stage (when the fourth trifoliate leaflets are fully expanded but before the fifth trifoliate leaflets are still unrolled) (Fig 6A). It is particularly noticeable that both GmSOC1 and Dt2 were expressed at the highest levels in NE3001 at the V2 stage. By contrast, GmSOC1 in IA3023 showed an expression pattern distinct from dt2. In particular, the expression levels of GmSOC1 in IA3023 continued to be elevated after the V2 stage through the V5 stage, suggesting that GmSOC1 may have different regulatory roles between the Dt2 and dt2 backgrounds.
Given that Dt2 is primarily expressed in the stem tips, we thus performed in situ hybridization to localize the transcripts of Dt2/dt2, GmSOC1, and Dt1 alleles in specific sections within the V2-stage stem tips in semi-determinate NE3001 and indeterminate IA3023 (Fig 6B). It was found that the Dt2 transcripts were concentrated in the central zone of the SAMs at the V2 stage in NE3001, where the Dt1 expression was not detected. By contrast, the dt2 transcripts were not detected in the central zone of SAMs in IA3023, where abundant Dt1 transcripts were observed. In the Dt2 (i.e., NE3001) background, the GmSOC1 transcripts were detected in SAMs, whereas no expression of GmSOC1 was detected in SAMs in the dt2 (i.e., IA3023) background. The spatiotemporally specific co-expression of Dt2 and GmSOC1 in NE3001 and absence of dt2 and GmSOC1 transcripts in SAMs of IA3023 suggest that the observed expression of GmSOC1 in SAMs and thus its novel function are Dt2-depedent, and that both Dt2 and GmSOC1 are involved in transcriptional repression of Dt1, responsible for the formation of semi-determinacy.
It is also noticeable that dt2 was expressed in lateral meristems in IA3023, where the transcripts of Dt1 was not detected, and that GmSOC1 was detected in lateral meristems in both NE3001 and IA3023, where no or minimal expression of Dt1 was observed (Fig 6B), suggesting that the lateral meristems at the stem tips of the V2 stage may be in the state of transition from IMs to FMs in both NE3001 and IA3023, and that both dt2 and GmSOC1 may be involved in floral induction in the lateral meristems of both NE3001 and IA3023, most likely, by suppressing Dt1 transcription, as Dt2 and GmSOC1 do in the SAMs of NE3001 (Fig 6C).
Our previous study demonstrated that the TFL1/tfl1 promoter was able to drive the expression of Dt1 in the Arabidopsis tfl1 mutant to convert the mutant phenotypes (determinate stem and early flowering) to the wild-type phenotypes (indeterminate stem and late flowering) [17]. Also, since the activity of the Dt1 promoter could be detected by the GUS gene in Arabidopsis (Fig 3B), we were thus curious about whether the expression of Dt1 driven by its own promoter in Arabidopsis tfl1 mutants could recover the wild-type phenotype, and if so, whether ectopic expression of Dt2 alone or in combination with Dt1 in the tfl1 mutant could produce semi-determinate stem growth habit that has not been observed in Arabidopsis.
To address these questions, we created a ProDt1:Dt1 construct comprised of the promoter of Dt1 and its rest genomic sequence, and introduced it to an Arabidopsis tfl1 mutant line to produce ProDt1:Dt1 transgenic lines. We also developed Pro35S:Dt2 transgenic lines with the wild-type (TFL1) Arabidopsis. These two transgenic lines were crossed to generate progeny lines containing both ProDt1:Dt1 and Pro35S:Dt2 in the tfl1 background. We found that the ProDt1:Dt1 transgenic line with the tfl1 background recovered the wild-phenotypes that are typically shown by the wild-type Arabidopsis (Fig 7D and 7J), suggesting that the Dt1 promoter functions as the TFL1/tfl1 promoter in driving Dt1 expression to fulfill the TFL1 function. The ectopic expression of Dt2 in the wild-type genetic background did not affect the indeterminate stem growth determined by TFL1. By contrast, the ectopic expression of Dt2 expression in the ProDt1:Dt1 transgenic line with the tfl1 background exhibited semi-determinate stem growth habit similar to shown by semi-determinate soybean (Fig 7E and 7K), indicating that the TFL1 promoter activity was not repressed by Dt2.
The spatiotemporal co-expression of Dt2 and GmSOC1 and their co-binding to the Dt1 promoter suggests the essential role of GmSOC1 in the formation of semi-determinacy. To further test this hypothesis, we crossed the semi-determinate Arabidopsis transgenic line (Fig 7E and 7K) with the Arabidopsis soc1 mutant [44,45], and obtained F2 plants containing both ProDt1:Dt1 and Pro35S:Dt2 in the tfl1 and soc1 background. As exemplified in Fig 7F and 7L, the plants with such a combination of genes showed indeterminate stem growth, indicating that SOC1 is indeed essential to repress Dt1 transcription in the Dt2 background in SAMs.
We present a novel mechanism underlying plant stem terminal flowering and semi-determinate growth habit, a key adaptation and agronomic trait that was formed post-domestication of soybean and was artificially selected for soybean production by breeding [23]. We demonstrate that the spatiotemporal expression of a recent gain-of function mutation allele Dt2 in apical meristems of main stems triggers co-expression of the putative soybean floral integrator gene GmSOC1, and the proteins encoded by these two genes directly interact to form co-repressors to directly target and repress Dt1 transcription, resulting in the formation of apical floral meristems that is essential for semi-determinate stem growth habit. In Arabidopsis and all other plant species that have been investigated to date, terminal flowering is achieved by the null mutations of TFL1 or its functional equivalents [9, 10, 17–22, 30]. Thus, our findings represent a unique mechanism reshaped by artificial selection. Similar to the “green revolution” semi-dwarf trait in cereals [46], semi-determinacy makes soybean plants more lodging-resistant and thus is desirable for production in high-yield lodging-prone environments. In addition, we demonstrate potential application of this mechanism for modification of stem growth habit in other species.
As reflected by the timing and spatial patterns of it’s expression, Dt1 in indeterminate soybean appears to function in a way similar to what TFL1 does in the wild type Arabidopsis to repress stem terminal flowering. In Arabidopsis, TFL1 starts to be expressed weakly in the center of the SAMs during the vegetative phase and its expression level is up-regulated at the stage when the SAMs make cauline leaves that bear shoot meristems in their axils [47, 48]. The expression of TFL1 remains in SAMs of the main shoot afterwards to repress the expression of the floral identity genes such as AP1, LEAFY and thus sustain the indeterminate growth until the plant cease to grow [35]. In emerging floral meristems on the flanks, AP1 and LFY were found to repress TFL1 by direct binding to its 3’ regulatory region [15, 16]. A more recent study demonstrated that AP1 recruits SEP4, SOC1, AGL24, and SVP to form a regulatory complex that represses the expression of TFL1 to initiate lateral flowering [49]. In soybean, the highest level of Dt1 in main stem tips was detected at the V0 stage, which seems to be equivalent to the Arabidopsis stage when TFL1 is up-regulated. Because the functional equivalents of the Arabidopsis floral identity genes such as AP1 and LEAFY in soybean have not been identified, it remains unclear when the floral meristems at the flanks are exactly initiated in soybean. Nevertheless, Dt1 expression was detected in the center zone of SAMs of IA3023, but not detected in the lateral meristems in either NE3001 or IA3023, suggesting that floral induction may have occurred in the lateral meristems in both varieties.
Several lines of evidence indicate that Dt2 is not the functional equivalent of the Arabidopsis of AP1, although the functional orthologs Dt1 and TFL1 are their respective direct targets. Firstly, Dt2 is a rare gain-of-function allele that is present only in semi-determinate soybean, while AP1 is a floral identity gene in the wild-type Arabidopsis; Secondly, Dt2 is not one of the four soybean duplicates orthologous to the Arabidopsis AP1 [32]. Instead, Dt2 appears to be an ancestral copy of MADS-box factor gene proceeding the divergence of AP1 from FUL that had occurred before the split of Arabidopsis from soybean; Thirdly, Dt2 is expressed in the central zone of SAMs to repress the expression of Dt1 (Fig 6B), whereas AP1 is not expressed in SAMs of the main shoots of Arabidopsis [35]. Fourthly, Dt2 binds to the promoter region of Dt1 (Fig 2E, 2F and 2G), whereas AP1 binds to the 3’ regulatory region of TFL1 to achieve suppression of the transcription of the two target genes [13]. Nevertheless, Dt2 appears to be responsible for initiation of floral meristems in SAMs similar to that was achieved by AP1 in the lateral meristems in Arabidopsis.
The transcripts of GmSOC1 were detected not only in the SAMs of NE3001, but also the lateral meristems of both NE3001 and IA3023 by in situ hybridization (Fig 6B). Further, the expression level of GmSOC1 in the main stems with both apical and lateral meristems continues to increase in IA3023 but that starts to decrease in NE3001 after the V2 stage (Fig 6A). Because the main stems of IA3023 continue vegetative growth at their apical meristems and floral induction at their flanks until all meristems are consumed and the plants get matured, whereas the main stems of NE3001 appear to have undergone the transition from IMs to FMs at the V2 stage (Fig 6B), the elevated expression levels of GmSOC1 in IA3023 after the V2 stage would be considered as additional evidence in support of the role of GmSOC1 as a floral identity gene in soybean. In Arabidopsis, TFL1 expression is repressed by SOC1 in an AP1-dependent manner [49]. Contrastingly, the repression of Dt1 by GmSOC1 appears to be Dt2-dependent. Such a similarity reflects not only the functional conservation between GmSOC1 and SOC1 as repressors of Dt1/TFL1, but also the way in which they function.
Although the dt2 transcripts were also detected in the lateral meristems of IA3023 (Fig 6B), the expression level of dt2 was declined after the V2 stage in IA3023, suggesting that, unlike GmSOC1, dt2 may not be essential for floral induction in the lateral meristems. If GmSOC1, indeed, is the functional equivalent of the Arabidopsis SOC1, as directly indicated by the recovered indeterminacy by Dt1, with the constitutive expression of Dt2, in the tfl1 and soc1 double mutants of Arabidopsis (Fig 7F and 7L), the expression of GmSOC1 would be essential for initiating terminal flowering through suppressing Dt1 expression and perhaps through activating the expression of other flowering identity genes in soybean such as functional equivalents of the Arabidopsis LEAFY and FUL in SAMs.
Several observations obtained in this study, such as the spatially specific and co-expression pattern of GmSOC1 and Dt2 (Fig 6), their direct interaction (Fig 4A and 4B), the lack of Dt2Δk for interacting with GmSOC1 and repressing Dt1 expression (Fig 3B and 3C), and the repressive effect of GmSOC1 on Dt1 expression (Fig 5C and 5D), suggest that GmSOC1 plays an essential role in forming the semi-determinate stem growth habit, and this role was likely fulfilled by its dimerization with Dt2. As the Arabidopsis SOC1 is not expressed in the shoot meristems, such a pattern of GmSOC1 expression would indicate its novel function, which was specifically triggered by Dt2. As similarly observed in Arabidopsis, rice SOC1, AGL24, SVP, and SEP4 orthologs regulate panicle branching through suppressing the TFL1 orthologs in rice, indicating the genetic pathways underlying inflorescence architecture are highly conserved between monocot and dicot species [49]. Intriguingly enough, such an inter-specifically conserved pathway is not conserved between the apical and lateral meristems in initiating flowering in semi-determinate soybean due to the spatiotemporal expression of Dt2 (Fig 6). It is possible that the activation of GmSOC1 was initiated through suppression of Dt1 by Dt2. Alternatively, GmSOC1 could be directly activated by Dt2. It would be interesting to further investigate how such specific expression of Dt2 was achieved simply through the gain-of-function mutation(s) that occurred outside of its CDS [32], and how the spatiotemporal expression of GmSOC1 was triggered by the Dt2 mutation, and to what extent the regulatory networks underlying soybean stem growth habit was reshaped by the Dt2 mutation.
The formation of the semi-determinate Arabidopsis by heterologous expression of Dt2 and Dt1 in the tfl1 mutants is an applausive observation (Fig 7), which suggests that all the functions of the soybean genes involved in the regulatory complex suppressing Dt1 expression can be fully provided by the Arabidopsis genes. By contrast, the overexpression of Dt2 in wild Arabidopsis did not result in any stem architectural changes, suggesting that Dt2 did not interact with TFL1. This may be explained by the absence of any CArG-boxes in the promoter region of TFL1 as potential target sites of Dt2. Therefore, the heterologous expression experiment demonstrated both conservation and divergence of regulatory sequences between TFL1 and Dt1 for precise control of their switch-on and switch-off. It would be important to further identify floral identity genes in FMs developed from the apical IMs and secondary IMs, respectively, towards a more in-depth understanding of the spatiotemporal specificity and commonality of floral regulation that determine that plant’s inflorescence architecture.
Given such a long period of divergence of soybean and Arabidopsis from a common ancestor, the formation of semi-determinacy by Dt2 and Dt1 in Arabidopsis would suggest a feasibility and potential application of this novel regulatory mechanism for modification of stem growth habit in many other plants, particularly, the legume crops, towards optimizing plant architecture for enhanced yield potential and adaptability.
Semi-determinate elite soybean line NE3001 (Dt2Dt2;Dt1Dt1), indeterminate soybean elite lines IA3023 and Thorne (dt2dt2;Dt1Dt1), and a Dt2 over-expression transgenic line (#2) in the Thorne genetic background were previously described [32]. An indeterminate elite line Kefeng 1 (dt2dt2;Dt1Dt1), used for hairy root transformation, was obtained from the USDA Soybean Germplasm Collection. The tfl1 mutant (tfl1-1) was obtained from The Arabidopsis Information Resource (TAIR) Arabidopsis Stock Centers. The Arabidopsis seeds were surface-sterilized with 10% bleach plus 0.01% Triton X-100 for 12 min, followed by washing five times with sterile water. The sterilized seeds were stratified at 4°C for 2 days and transferred to culture media or soil for further growth at 22°C under the condition of 16 h of 120 μE·m−2·s−1 light and 8 h of dark.
Genomic DNA isolation, PCR primer design, PCR amplification with genomic DNA, PCR product purification, RNA isolation, cDNA synthesis by reverse transcription-PCR (RT-PCR), quantitative real-time-PCR (qRT-PCR), and sequencing of DNA and cDNA fragments were performed using protocols previously described [17, 32]. In the qRT-PCR experiments, three biological replicates were analyzed to quantify the levels of gene expression in NE3001, IA3023, Kefeng 1 and three technical replicates were performed to measure the levels of gene expression in Thorne and the Thorne transgenic line, and the soybean ATP binding cassette transporter gene Cons4 [39] was used as the internal control, and we normalized the relative expression levels of the genes/alleles Dt2, dt2, Dt1, GmSOC1 in each experiment by setting the lowest expression level as 1.0. Primers used for PCR, RT-PCR, qRT-PCR and sequencing are listed in S3 Table.
The full-length or portions of the CDSs of Dt2, and GmSOC1 were amplified by RT-PCR using KOD hot start DNA polymerase (Novagen catalog no. 71087), and the Dt2ΔK fragments were obtained by overlapping PCR with two overlapped CDS fragments as templates in a same reaction. The CDS and fused CDS fragment were then inserted to pCR8/GW/TOPO vector (Invitrogen, catalog no. K2500-20) and verified by sequencing. Subsequently, the verified inserts were cloned into the binary vector pGWB17 [50] to obtain the three constructs Pro35S:Dt2, Pro35S:Dt2ΔC, Pro35S:Dt2ΔK, and Pro35S:GmSOC1, and the verified inserts were cloned into pBI-ΔGR-GW [51] to generate the Pro35S:Dt2-GR and Pro35S:GmSOC1-GR constructs.
The ~2.4-kb upstream sequence from the start codon (dabbed the promoter region or pProDt1), the truncated promoter without the cluster of the five putative CArG-boxes (dabbled ProDt1Δ), the CDS, and the ~1.1kb downstream sequence from the stop codon (dabbed terminator region) of Dt1 were amplified from the indeterminate soybean cultivar Williams 82. The obtained PCR fragments were cloned into the pGEM-T Easy Vector (Promaga, catalog no. A1360) and then sequenced. The verified clones with the promoter region, the CDS, and the terminator region, were digested by PstI and SalI, SalI and XbaI, and XbaI and BamHI, respectively, and then integrated into pPZP212 [32]. The construct of ProDt1Δ:LUC was made by integrating pProDt1 into pGWB435 [50]. The Dt1 promoter sequences with point mutations within each of the three CArG boxes were created by using QuikChange II Site-Directed Mutagenesis Kit (Agilent Technologies, Catalog #200523) with specifically designed primers (S3 Table). These constructs were introduced into Agrobacterium tumefaciens strain GV3101 or Agrobaterium rhizogenes strain K599. The Arabidopsis transgenic lines each from a single construct were obtained by A. tumefaciens-mediated transformation, and the transgenic lines with genes from two distinct constructs were generated by crossing two transgenic lines with respective transgenes and subsequent screening of the progeny lines. The soybean transgenic hairy roots were produced by A. rhizogenes-mediated transformation following a protocol previously described by Kereszt et al. [40]. The seedlings with transgenic roots were sprayed a solution of 0.03mM DEX with 0.005% Silwet L-77, a solution of 1.8mM CHX with 0.005% Silwet L-77 or a solution of 0.03mM DEX and 1.8mM CHX with 0.005% Silwet L, and levels of gene expression was measured four hours after each treatment.
Thermo Scientific Antigen Profiler, a bioinformatics protein sequence analysis tool and custom peptide design algorithm, was provided by Pierce Biotechnology, Inc. and employed to design a unique 19 amino-acid peptide from Dt2, which was then used to raise a Dt2-specific antibody from a rabbit (dabbed anti-Dt2, Pierce Biotechnology, Inc.). The specificity of the anti-Dt2 was tested by Western blot using total proteins isolated from the stem tips of NE3001 at the V2 stage.
ChIP assays with anti-Dt2 were performed following the protocols described previously [52, 53], with minor modification. Stem tips of NE3001 collected at the V2 stage were immersed in 1×Phosphate Buffered Saline (PBS) buffer containing 1% formaldehyde (Macron, catalog no. K15754) for cross-linking. Enrichment of the precipitated DNA by anti-Dt2 relative to DNA recovered from the control treatment without anti-Dt2 was measured by qRT-PCR with three biological replicates. The primers used in ChIP-PCR are listed in S3 Table.
The CDSs of Dt2 and GmSOC1 were cloned into the expression vector pET-DEST42 containing a 6×His tag (ThermoFisher Scientific, catalog no. 12276–010), separately, to generate the Dt2-6×His and GmSOC1-6×His constructs. The two constructs were transformed into the Escherichia coli strain BL21 and Rosetta (DE3), respectively, The Dt2-6×His and GmSOC1-6×His fusion proteins were then induced in the transformed cells by growing at 37°C for 5h in the 2×YT medium with 1 mM isopropyl β-D-1- thiogalactopyranoside and then extracted and purified with Ni-NTA Agarose (Qiagen, catalog 30210). EMSAs were performed using digoxigenin-labeled probes and the DIG Gel Shift Kit (Roche, 3353591910) following the manufacturer’s instructions.
Ten transgenic plants for each construct were mixed for protein extraction and histochemical staining. The Gus activities were measured in a method described earlier [54].
The stem tips from Williams 82 at the V2 stage were used to isolate total RNA, which was used to synthesize cDNA by reverse transcription. The pool of cDNA fragments were cloned into pGADT7 (Clontech, catalog no. 630442) and then transformed into the yeast strain Y187 (Clontech, catalog no. 630457), following the manufacturer's instructions The CDS of Dt2 was inserted into vector pGBKT7 as a bait and introduced into the yeast strain Y2H Gold (Clontech, catalog no. 630498). Mating between the Dt2 strain and the cDNA library were screened on the quadruple dropout medium (QDO) with SD/–Ade/–His/–Leu/–Trp (Clontech, catalog no. 630322). The positive cDNA clones were sequenced and the interaction between Dt2 and one of the positive clones, which contains a fragment from the CDS of GmSOC1, was further validated by co-transformation of pGBKT7 with the Dt2 CDS and pGADT7 (Clontech, catalog no. 630442) with the GmSOC CDS into Y2HGold and grown on the selection medium QDO supplemented with X-a-Gal (Clontech, catalog no. 630462) and Aureobasidin A (Clontech catalog no. 630466) following the manufacturer's instructions.
For BiFC assays, the Dt2, Dt2ΔC, and Dt2ΔK were cloned into the pEarleyGate201-YN vector [55] and the GmSOC1 was cloned into the pEarleyGate202-YC vector [55]. These constructs were introduced into the A. tumefaciens strain GV3101, together with the p19 strain, and the strains carrying GmSOC1 and Dt2, Dt2ΔC, or Dt2ΔK were co-infiltrated into leaf epidermal cells of 3- to 4-week old tobacco (Nicotiana benthamiana), following a protocol described previously [56]. The transformed cells were observed and photographed using a confocal scanning microscope (Nikon 90i) 24 h after infiltration.
For subcellular localization, the CDS of GmSOC1 was cloned into the plasmid pGWB405 to form a fusion protein of GmSOC1 and a green fluorescent protein (GFP) under the control of 35S promoter, which are provided by the plasmid. The construct was introduced into leaf epidermal cells of 3- to 4-week old tobacco by A. tumefaciens infiltration. The transformed cells were observed and photographed using a confocal scanning microscope (Nikon 90i) 24 h after infiltration.
RNA in situ hybridization was performed according to a previously described protocol [57]. A 120-bp fragment specific to the Dt2/dt2 cDNA, a 123-bp fragment specific to the Dt1 cDNA, and a 188-bp fragment specific to the GmSOC1 cDNA were amplified with respective primer sets (S3 Table), and then integrated into the pGEM-T Easy vector, respectively. Digoxigenin-labeled sense and anti-sense probes were obtained from EcoR-digested linear pGEM-T Easy Vectors with 120-bp, 123-bp, or 188-bp inserts by in vitro transcription with SP6 or T7 RNA polymerase (Roche, catalog no. 11175025910) according to the manufacturer’s protocol. Hybridization signals were detected and photographed using a confocal scanning microscope (Nikon A1R).
|
10.1371/journal.pntd.0006078 | Biological and phylogenetic characteristics of West African lineages of West Nile virus | The West Nile virus (WNV), isolated in 1937, is an arbovirus (arthropod-borne virus) that infects thousands of people each year. Despite its burden on global health, little is known about the virus’ biological and evolutionary dynamics. As several lineages are endemic in West Africa, we obtained the complete polyprotein sequence from three isolates from the early 1990s, each representing a different lineage. We then investigated differences in growth behavior and pathogenicity for four distinct West African lineages in arthropod (Ap61) and primate (Vero) cell lines, and in mice. We found that genetic differences, as well as viral-host interactions, could play a role in the biological properties in different WNV isolates in vitro, such as: (i) genome replication, (ii) protein translation, (iii) particle release, and (iv) virulence. Our findings demonstrate the endemic diversity of West African WNV strains and support future investigations into (i) the nature of WNV emergence, (ii) neurological tropism, and (iii) host adaptation.
| The West Nile virus (WNV) can cause severe neurological diseases including meningitis, encephalitis, and acute flaccid paralysis. Differences in WNV genetics could play a role in the frequency of neurological symptoms from an infection. For the first time, we observed how geographically similar but genetically distinct lineages grow in cellular environments that agree with the transmission chain of West Nile virus—vertebrate-arthropod-vertebrate. We were able to connect our in vitro and in vivo results with relevant epidemiological and molecular data. Our findings highlight the existence of West African lineages with higher virulence and replicative efficiency in vitro and in vivo compared to lineages similar to circulating strains in the United States and Europe. Our investigation of four West African lineages of West Nile virus will help us better understand the biology of the virus and assess future epidemiological threats.
| West Nile virus (WNV) is a member of the Japanese Encephalitis virus (JEV) serocomplex and is a part of the genus Flavivirus of the family Flaviviridae. The WNV is a single-stranded, positive-sense RNA virus. The genomic RNA is about 11 kilobases (kb), containing one long open reading frame (ORF) flanked by 2 non-coding regions. This ORF encodes for a polyprotein, which is processed into three individual structural (Capsid, pre-Membrane, Envelope), and seven non-structural (NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5) proteins [1–4].
West Nile fever disease (WN fever) is caused by the WNV. WN fever in humans can range from asymptomatic infections or mild acute febrile illness, to neurological diseases including meningitis, encephalitis, and acute flaccid paralysis [5–7]. WNV’s host range is extensive: it has been detected in over 65 species of mosquitoes and ticks, 225 species of birds, and 29 different animals [8,9]. A human vaccine or specific antiviral treatment for WN fever is currently unavailable.
WNV was first discovered and isolated from the blood of a woman suffering from febrile illness in 1937 in Uganda [10]. Cases of WN fever were documented in Israel and Egypt in the early 1950s, France in the 1960s, and South Africa in the 1970s [11]. The global awareness of WN fever increased in the 1990s, as sporadic and major outbreaks occurred, primarily in the Mediterranean Basin and occasionally in Europe [5]. In 1999, WNV unexpectedly emerged in New York City, signifying the first confirmed incidence of WNV in the Western Hemisphere. Since then, WNV has spread throughout the Americas, causing over 20,265 cases of neurological disease and 1,783 case fatalities in humans and even higher rates of mortality among birds in the United States [12–16].
Meanwhile, WNV continued to spread and cause WN disease and encephalitis in Europe, Asia, and Oceania [17]. In the 1990s, the largest outbreaks occurred in Romania in 1996 [18], and Russia in 1999 [19], with 17 and 40 human fatalities, respectively. In the 21st century, emergences of WN fever and encephalitis have been reported in Europe [20], with a hallmark human neurological outbreak in Greece in 2010 [21], and several noteworthy outbreaks in Italy [22–24], Hungary [25] and Serbia [26].
WNV is biologically diverse; up to nine lineages have been proposed [27–30]. However, most human outbreaks of WN encephalitis have been attributed to lineages 1 and 2. Lineage 1 is globally spread and exists in distinct clades. Clade 1a comprises of strains isolated from Europe, Africa, and the Americas. Clade 1b, also referred to as Kunjin virus, has been restricted to Oceania. Major outbreaks in Europe, Africa, and the Americas with neurological diseases are caused by strains belonging to lineage 1, with an exception to clade 1b where neurological disease is rarely reported [12,31,32].
Lineage 2 was exclusively reported in Africa up until 2004, until it was isolated from humans and bird populations in Hungary, Greece, and Italy [4,21,23,33]. Lineage 2 was also considered to be less pathogenic than lineage 1, until it caused severe disease in South Africa and encephalitis among birds and humans in Europe [4,21,23,33,34]. Both lineages include strains with varying degrees of neuroinvasiveness in humans [35].
Besides lineages 1 and 2, there are lineages that are less widespread. Lineage 3, also referred to as Rabensburg virus, was repeatedly isolated in the Czech Republic [36–38]. Lineage 4 has been isolated and reported from Russia [39]. The 5th lineage was isolated from India, and is often identified as a distinct clade of lineage 1 (clade 1c) [40]. A putative 6th lineage, based on a small gene fragment, has been described from Spain [27,41].
Koutango virus (lineage 7) was initially classified as a different virus, but is now a distinct lineage of WN virus [31,42]. Lineage 7 strains were isolated from ticks (this study) and rodents, a rare feature among WN virus lineages [4]. The Koutango strain virus has also been shown to have a higher virulence than the lineage 1a strain “NY99” in mice [43,44]. Although there was a report of an accident where a Senegalese lab worker was symptomatically infected with the Koutango strain, a natural human infection has yet to be confirmed [45]. Additionally, a new lineage (putative lineage 8) of WNV was isolated from Culex perfuscus in Kedougou, Senegal in 1992 [4]. Finally, a putative 9th lineage, or sublineage of lineage 4, was isolated from Uranotaenia unguiculata mosquitoes in Austria [27].
Despite the presence of lineages 1, 2, 7 (Koutango) and a putative 8th lineage circulating in Africa [4,46,47], WNV has had minor impact on human health. Sporadic outbreaks were observed in several African counties [48–50], with lower frequencies of neurological disease than that reported from outbreaks in the USA [51,52]. For example, Senegal has never had a major outbreak of WN fever, but was the source of several endemic genotypes that were identified and sequenced. Moreover, in Senegal, WNV antibody seroprevalence has been around 80% in sampled humans, horses, and birds [53–57].
A recent study on the vector competence of African WNV lineages demonstrated that local mosquito populations lack efficient transmission of WNV [4]. Besides vector competence—i.e. intrinsic genetic variations among lineages—host adaptation, movement of host populations, climate and ecological factors could play a role in viral replication, virulence, and the outcome of infection. The N-linked glycosylation site of the envelope protein may be associated with differences observed in: (i) WNV neuroinvasiveness in mice, (ii) viral replication, and (iii) transmission of WNV in mosquitoes [4,58–60]. In this regard, Senegal has been a focal point in the studies of WNV virus, where multiple lineages of WNV are co-circulating endemically, but whose biology remains poorly understood.
To address these questions, we analyzed complete coding regions (polyproteins) of four different lineages circulating in Senegal and West Africa. Using additional WNV sequences from Genbank, we performed a phylogenetic analysis using the complete polyprotein sequences of the viruses and investigated sites for positive selection. We also analyzed the biological properties of these 4 WNV lineages using in vitro and in vivo models. Ultimately, understanding the relationships among ecological and genetic differences will ameliorate our understanding of WNV emergence, epidemiology, and its maintenance in nature.
In this study, three complete polyprotein genes from Senegal isolates were sequenced: ArD76986, ArD96655, and ArD94343 (Table 1). These novel sequences are representative of lineages 1, 7 (Koutango) and 8 (putative), respectively. The lineage 1 and lineage 8 strains were isolated from Culex mosquito species, while the lineage 7 strain was isolated from a tick species.
Acknowledging previous works that have reconstructed the evolutionary history and those that have characterized novel isolates and lineages of WNV, we included seven additional complete ORF sequences to compare differences at the gene and protein level (Fig 1). Among representative sequences, the average nucleotide pairwise identity is 77.6% (s.d. = 4.1%) and the amino acid average pairwise identity is 90.1% (s.d. = 3.3%). When comparing individual sequences, the NY99 strain (Accession number: AF196835, lineage 1a, United States 1999) shared a 99.5% pairwise identity to ArD76986 (Accession number: KY703854, lineage 1a, Senegal 1990) at the amino-acid level (Fig 1A). The sequence diversity of endemic WNV lineages in Senegal (SN) is notable, as the lineage 1 strain (ArD76986) was 88.9% and 90.9% identical at the amino-acid level to the ArD96655 (Accession number: KY703855, lineage 7, SN 1993) and the ArD94343 (Accession number: KY703856, lineage 8, SN 1992) strains respectively. Between the lineage 7 and lineage 8 strains, the amino-acid pairwise identity was 87.9% (Fig 1A).
The 1937 WNV isolate of strain B956 (Accession number: AY532665, lineage 2, Uganda 1937) is of particular interest, as it is the oldest clinical isolate available with a complete ORF sequenced. Amino acid pairwise identity was 93.9% to the NY99 strain sequence and to the lineage 1a (SN) sequence, 89% to the lineage 7 strain sequence and 91.1% to the lineage 8 strain sequence. At the nucleotide level, the lineage 2 strain (UG) had a pairwise identity of 79.1% to the NY99 sequence, 79.3% to the lineage 1a strain (SN) sequence, 76.9% to the lineage 7 strain sequence and 77.7% to the lineage 8 strain sequence. Additionally, B956 contains a 12 base pair deletion at nucleotide position 1,331, corresponding to the WNV envelope glycosylation site.
We compared published sequences and published works to identify whether mutations that have been shown to influence WNV virulence and replication were present in the newly sequenced open-reading frames. For each strain, we discovered amino acid changes that were associated to a phenotypical change and many additional mutations with unknown consequences (Fig 1B). For example, the 22nd and 72nd codon sites of the pre-membrane protein (prM) have been shown to play a role in enhancing the virulence and particle secretion in WNV [61]. At this site, we found alterations in the lineage 7 strain (SN_1993_L7), and the lineage 8 strain (SN_1992_L8).
Another example is the glycosylation site found in the 154-156th positions of the envelope (Env) protein, which is considered a virulence factor [62]. We found that the lineage 1 strain from Senegal (SN_1990_L1a), the Kunjin strain (AU_1991_L1b), the NY99 strain (US_1999_L1a) and the lineage 8 strain (SN_1992_L8) harbored the NYS motif while other strains had variations or deletions in this locus. Next, the 249th codon position of the NS3 protein [63], the helicase protein, was found to increase viremia and virulence in birds, and could play a role in other hosts. We observed several variations in our data at the 249th codon position (Fig 1B).
Additionally, changes in the highly conserved 120P-E-P-E123 region of the NS4A protein can attenuate or even impair virion replication and release [64], which we found present in the lineage 8 strain. Finally, a mutation in the NS5 protein, serine (S) to phenylalanine (F) at the 653rd position in the NS5 protein, is associated with an increased resistance to interferon [65], a mutation that is shared by the lineage 7 strain (SN_1993_L7) (Fig 1B). We also found several synonymous changes in positions corresponding to known virulence motifs, such as variability in the third codon site position (the wobble base) during the translation of serine (S) at the 156th codon site. We also investigated sites within the NS2A [66], NS4B [67], and additional sites within the NS5 region that are known to impact on infectivity and virulence [65], but no mutations were present in our sequences.
The phylogenetic analysis revealed a similar topology to the ones obtained from previous maximum likelihood trees [27,40,42,68,69]. Currently, up to 9 distinct lineages have been suggested.
A total of 95 sequences, including 3 novel polyprotein sequences from Senegalese isolates (Table 1), were used to estimate a maximum-likelihood tree with FastTree (S1 Fig) and a very similar relaxed clock Bayesian maximum-clade credibility (MCC) tree (Fig 2), summarizing the MCMC runs with BEAST. The MCC tree was scaled to time (years) and branch tip-nodes were colored to identify previously classified lineages [27]. Here, the time to the most recent common ancestor (tMRCA) with its corresponding 95% highest posterior density (HPD) interval for WNV was estimated in the unit of years. The tMRCA of WNV is predicted to have originated in the late 16th/early 17th century (95%HPD: 1476–1765), a major split that diverges lineages 1, 5 and 7 from lineages 2, 3, 4, 8, and 9. Both lineage 1 and 2 show multiple introductions into Europe and other New World countries. Additionally, we see that lineages 1, 2, 7, and 8 have been isolated in West Africa, yet only lineages 1 and 2 have emigrated.
Infection, viral proliferation, and virulence in each cell type were measured by 4 different tests over a 146 hours post-infection period: quantitative reverse transcriptase PCR (qRT-PCR) of the lysed cell fracture to measure genome replication (Fig 3A and 3B), qRT-PCR of the supernatant fraction to detect genome replication dynamics (i.e., total number of particle release) (Fig 3C and 3D), immunofluorescence staining of the cells to visualize the infectivity of cells and estimate protein translation efficiency (Fig 3E and 3F), and plaque assays to determine the amount of infectious viral particles (PFU/ml) from the supernatant fraction (Fig 3G and 3H). Using Ap61 and Vero cells, our goal was to replicate the biology of WNV in a mosquito vector and its vertebrate host.
We found that African lineages have different growth dynamics in mosquito and mammalian cell lines. In Aedes pseudoscutellaris cells, growth dynamics were similar for all lineages, (Fig 3, left column) where all lineages exhibited successful replication and generation of infectious particles. In Vero cells (Fig 3, right column), lineages 1, 2, and 7 showed exceptional growth, with lineage 2 strain exhibiting the highest replication and particle release capabilities, and lineage 7 strain having exceptional translational dynamics and highest PFU/ml during the infection interval.
We observed cell-specific growth differences among different WNV strains. For example, Fig 3A and 3B showed differences in genome replication dynamics in the cells with respect to host cells. Interestingly, lineage 1 strain had higher genome replication in Ap61 cells (p-value ranging from 2.22x10-16 to 0.002) while the lineage 2 strain had higher genome replication in Vero cells (statistically comparable). Lineage 8 showed a lower significant replication profile in Vero cells (p-value ranging from 8.81x10-13 to 0.031). Furthermore, differences in growth at T0 further supports that WNV lineages could have a preference to a specific cellular environment. The rate of viral attachment, entry and replication initiation can all depend on the genetics of the infecting strain [70].
We estimated the total number of released particles at different times post infection by measuring the WNV RNA copy number in the cell supernatant. All tested lineages had comparable genome copy numbers in Ap61 supernatants (Fig 3C). However, we found a significantly higher copy number of total particles released for the lineage 2 strain at 22, 28, and 50 hours post-infection (hpi) in both in vitro models (p-value ranging from 2.22x10-16 to 0.023). Lineage 8 strain showed significantly lower genome copy numbers in Vero supernatants (p-value ranging from 8.81x10-13 to 0.031).
Next, we approached differences in protein translation efficiency between lineages by detecting viral proteins using an immunofluorescence assay (IFA) (Fig 3E and 3F). The lineage 7 strain displayed more efficient protein translation in both cells (p-value ranging from 3.98x10-13 to 0.011), while lineage 8 strain had significantly lower levels of protein translation in Vero cells. Nevertheless, the translation rate in lineage 8 increased significantly from T124-146 in Ap61 cells. We also noticed a delay on translation detection in both cells, with no detectable protein production until T99 hours (Fig 3E) and T50 hours (Fig 3F) respectively.
To quantify the infectious particles of different WNV strains, we used plaque assays to estimate the amount of infectious viral particles (PFU/ml) in the supernatant fractions. In Ap61 cells, we found a similar profile of infectious particles production for all lineages, with significant higher rates at 124 hpi and 146 hpi for the lineage 7 strain (p-value ranging from 2.22x10-16 to 0.028) (Fig 3G). In Vero cells, lineage 1 and lineage 7 strains had higher number of PFU/ml, while lineage 2 had an intermediate profile and lineage 8 had the lowest amount of infectious viral particles, with significant differences from 28 to 124 hpi (p-value ranging from 2.22x10-16 to 0.049) (Fig 3H).
Finally, we approximated the replication efficiency by finding the ratio of the number of virions released in the supernatant–particles that completed the infectious cycle–divided by the number of plaque forming units (PFU) [71–73]. We estimated the ratio for each strain in each cell and found significant differences in replication efficiency (p-value ranging from 5.49x10-16 to 0.0223) (Fig 4). There are some consistencies with Fig 3, where the lineage 7 strain was the most efficient and the B956 strain was the least efficient in vitro. Lineage 1 and lineage 7 strains seem to be more cell-specific; both replicated less efficiently in Ap61 cells.
To determine the virulence of WNV strains (Table 1), we challenged five- to six-week-old mice with three different viral doses and observed their overall survival for 21 days. Depending on the strain and dose used, several mice developed clinical disease and died (Table 2). Clinical signs included tremors, reduced activity and reluctance to move, hind leg paralysis and closed eyes. The PBS-inoculated control groups exhibited no signs of disease throughout the experiment.
The lineage 7 strain was the most virulent of the strains at all administered doses (Wilcoxon rank sum test, p-values < 0.05). In fact, the lineage 7 strain induced the shortest survival time compared to the other strains and always resulted in 100% mortality in every experiment (Fig 5 and Table 2). Interestingly, in most cases, mice inoculated with the lineage 7 strain died without showing any clinical signs.
Comparatively, mice inoculated with the lineage 1 and lineage 2 strains usually showed signs of disease at least 1 day before dying. However, lineage 8 showed no virulence (100% survival) at 100 and 1000 PFU doses (Fig 5B and 5C). In fact, only one mouse mortality was observed at 10000 PFU (Fig 5A and Table 2).
To determine the evolutionary pressures acting on the WNV ORF, we estimated the ratio of nonsynonymous (dN) to synonymous (dS) substitutions per codon site (where dN—dS > 0, signifies positive selection) using 95 sequences, which represent all investigated WNV lineages. Our investigation on selection regimens acting on all WNV complete ORF sequences—with the FUBAR method—revealed 3313 well supported (posterior probability ≥ 0.9 and Bayes Factor < 3.0) sites under purifying selection (S1 Table and S2 Fig). However, we found 95 statistically significant sites (p-value ≤ 0.1) under diversifying episodic selection (S2 Table and S3 Fig), using MEME method.
Despite the presence of at least four different lineages in West Africa, there has never been a major outbreak, nor a large frequency of encephalitic cases connected with WNV. The lack of a WN disease “burden” within Senegal could suggest that WNV is endemic, which could explain the high seroprevalence, and therefore, few susceptible hosts [54]. However, the threat of WNV emerging to places where the population’s seroprevalence is much lower or even naive is a serious concern. Avian migratory routes could have played a role in the emigration of WNV strains from Africa [53], and for other African-borne arborviruses such as Usutu virus [74]. The extensive genetic diversity (Fig 1A) and broad host range of WNV [8,9] could have also contributed to its global dissemination (Fig 2), as certain mutations have been previously prosecuted with lineage 1’s entrance in the United States [32] and lineage 2’s emergence in Europe [75]. As a consequence, several groups have investigated how specific genetic changes and selective pressures within the WNV ORF can affect the phenotypical behavior of a WNV strain.
In our study, the growth kinetics of the different West African WNV lineages were explored in Aedes pseudoscutellaris (Ap61) and African green monkey kidney cells (Vero) to reflect infection dynamics in two common classes of WNV hosts (insect vector and primate) (Figs 3 and 4). The virulence of these lineages in mice was also analyzed (Fig 5 and Table 2). We found that these 4 West African lineages have significant differences in their ability to proliferate in our tested cell lines and their degree of virulence in mice (Figs 3, 4 and 5). We also explored how our in vitro and in vivo results could be explained by their evolutionary (Fig 2) and individual genetic variations (Fig 1).
In agreement with other viruses that use alternate hosts (vertebrate-arthropod-vertebrate) and cause acute infections, we found that the majority of WNV codon sites are undergoing purifying selection [76]. Nevertheless, some of the significant episodic diversifying sites that we found are related to virulence, like the 444th and 446th codons in polyprotein gene (154-156th positions of the Env protein) that encode the N-glycosylation motif (NYS). This site is present in many lineage 1 strains, some neuroinvasive lineage 2 strains [34,77], and the lineage 1 and 8 strains from this study (Fig 1B). We found significant diversifying selection acting on these codon sites in 3% and 24% in all of the WNV lineages, respectively (S1 Table). These episodic non-conservative changes could have resulted in the loss of the N-linked glycosylated site motif, which is related to less efficient replication in Culex cells [59] and better replication in Aedes albopictuscells [60]. This N-linked site is also associated with neuroinvasiveness in mice [62]. Second, we found that 1754th codon in polyprotein (249th in NS3) was under diversifying selection (ω = 33.1) 13% of the time and under purifying selection (ω = -0.91) 87% of the time (S1 Table) in our WNV dataset. As this site was discovered to increase viremia and virulence in birds [63], further experiments in avian cell lines should be explored to see if our discovered substitutions effect replication in avian hosts and transmission dynamics.
Although no significant diversifying selection was observed on the cleavage site in the NS4A protein (120PEPE123 motif), we did discover the P122S substitution in the lineage 8 strain (Fig 1B). Crucially, induced mutations in this motif are related to low rates of replication and protein production in Vero cells [64], which we expected and observed for the lineage 8 strain in vitro (Fig 3B, 3D and 3F), and could help explain its low virulence in vivo (Fig 5). In general, we observed little change in viral replication and protein production between West African strains in Ap61 cells (Fig 3A, 3C and 3E). This could suggest that the conservation of PEPE motif may have a lesser role in replication in mosquito cells (lineage 8) or that the strains may have been “pre-adapted” prior to our experiment.
The lineage 7 strain has the S653F NS5 mutation that is associated with an increased resistance to interferon [65], which could help explain its phenotypical virulence in vitro and in vivo (Figs 1B, 3, 4 and 5). However, because Vero cells are known to be interferon-deficient, we could not associate this mutation to our in vitro results for lineage 7 in Vero cells (Fig 3, right column). Nevertheless, all three West African strains contained a non-synonymous change in a locus that was previously explored by site-directed mutagenesis experiments (Fig 1B). Interestingly, we also detected synonymous changes in the “wobble” base position of the codons in “sites of interest”. However, our knowledge of how synonymous changes impact infectivity, virulence, and replication of WNV is still limited.
As previously described, lineages 1 and 2 originated in Africa and emerged as a New World pathogen over the last 60 years [68]. Lineage 7 could be following a similar path; besides Senegal, it has been detected in Somalia, Gabon and possibly in Italy [78–80]. Our study supports that there are other lineages besides 1 and 2—such as lineage 7—that can exhibit high virulence in mice and efficient replication in mammalian cells (Figs 3, 4 and 5). This high virulence of lineage 7/Koutango strains in mice has been explored in two other studies, where the high virulence is suggested to be a result of delayed viral clearance and a weak neutralizing antibody response [43,44]. All three doses tested resulted in 100% mice mortality (Table 2), which agreed with previous results. The differences in average survival time and mortality rates compared to previous studies, could be explained by differences in the passage history of viral strains, and the age of the infected mice [81].
The Senegalese lineage 1 strain exhibited moderate virulence in mice (Fig 5) and caused comparably less mortality than the NY99 strain when compared to similar studies [44]. Differences in neuroinvasive potential and virulence among lineage 1 strains has been reported and could be explained by genetic differences [35]. Alternatively, lineage 8 showed poor growth capabilities in Vero cells (Figs 3 and 4) and almost no virulence in mice (Fig 5 and Table 2), suggesting that it may be restricted to vertical transmission or is species restrictive. Lineage 8 was described to have a similar phenotype to Rabensburg virus (lineage 3, Czech Republic 1997). Moreover, growth kinetics and vector competence studies revealed poor growth of the Rabensburg virus in mammalian cell lines and low virulence in mice [36,82]. This similarity could indicate that both lineage 8 and the Rabensburg strain may be restricted in host range and are also maintained in nature through vertical transmission. Investigating the vector competence of lineage 8 in different arthropod species (i.e. Culex, Aedes, and tick species) could lead to a better understand the transmission dynamics and maintenance cycles of WNV in nature. The low virulence phenotype of the lineage 8 strain could also be a factor for its consideration as a potential vaccine candidate for West Nile fever.
Further studies could complement our analysis, particularly, on other factors that could explain differences in WNV host and disease dynamics. Exploring variations in codon usage bias could also help explain biological differences [83], as distinct lineages have shown different degrees of natural selection and mutational bias. Site-directed mutagenesis studies may also help explain how strain-specific mutations, both synonymous and non-synonymous, could explain deviations in replication efficiency and virulence for our in vitro and in vivo results. For example, future studies in cell lines with interferon may help clarify the impact of the S653F NS5 mutation for the lineage 7 strain. Additionally, the flavivirus 5’ and 3’ untranslated regions (UTR) can affect replication and translation; certain mutations in these regions can cause complete viral attenuation [84–87]. Unfortunately, we could not investigate their impact in this study, as the majority of WNV UTR’s were publically unavailable.
Taking everything into account, especially differences in sequences, growth dynamics and virulence in vivo, the West Nile virus is a pathogen with the capability to cause severe epidemics anywhere in the globe. As complete genome sequences including the 5’ and 3’ UTR regions are currently being generated, this could lead to future studies focused on in vivo transmission and growth dynamics. As additional strains of WNV are characterized, monitoring the global diversity and distribution will aid in threat assessment and epidemiological modeling if future outbreaks are to occur.
Two cell lines have been used for virus cultivation and growth kinetics. Ap61 cells (Aedes pseudocutellaris) were grown in L15 (Leibovitz’s 15) medium (10% heat-inactivated fetal bovine serum [FBS], 1% penicillin-streptomycin, 0.05% amphotericin B [Fungizone] (GIBCO by life technologies; USA) and 10% tryptose phosphate (Becton, Dickinson and Company Sparks, USA) and incubated at 28°C without CO2. Vero cells (African green monkey kidney epithelial cells; Cercopithecus aethiops) (obtained from Sigma Aldrich, France) were grown using the same medium without tryptose phosphate and CO2. Furthermore, PS (Porcine Stable kidney cell line, American type Culture Collection, Manassas, USA) cells were grown in same conditions than Vero cells and have been used for plaque assay.
The virus strains used in this study corresponding to lineages 1, 2, Koutango (lineage 7) and 8 were described in Table 1. The virus stocks were prepared by inoculating Aedes pseudoscutellaris (Ap61) continuous cells lines for 4 days. The infection status was tested by immunofluorescence assay (IFA), real-time RT-PCR (Reverse Transcriptase-Polymerase Chain Reaction) and plaque assay. The supernatant of infected cells were aliquoted, frozen at -80°C, and used as viral stocks for growth kinetics.
A total of 862 complete WNV polyprotein gene sequences with country and year of isolation data were available and initially downloaded from Genbank for this study. A large number of sequences were from the Americas and formed a monophyletic group of lineage 1a comprising 770 sequences. To reduce computer-processing requirements while maintaining the authenticity of our results, we removed all lineage 1a sequences except for a single representative sequence denoted “NY99” (accession number: AF196835). With the addition of 3 new sequences, a total of 95 sequences were aligned using Muscle v3.8.31 [88] and manually curated using Se-Al v2 [89]. For Fig 1, the available complete polyprotein sequences representative of WNV diversity (excluding lineage 6, which there is only a partial sequence available) were included to compare genetic percent identities.
Likelihood mapping analyses for estimation of data quality were performed using Tree-Puzzle (Quartets ranged between 10,000 and 40,000) [90,91]. For each alignment we performed recombination screening (RDP, GeneConv, Chimaera, MaxChi, BootScan and SiScan) in RDP4.61 [92].
The Bayesian phylogenetic analysis was performed using Bayesian Inference (BI) using a general time-reversible with gamma-distributed rate variation and invariant sites model (GTR+Γ+I), as selected by Akaike's information criterion (AICc) in jModelTest 0.1 [93]. The evolutionary analysis was conducted assuming a relaxed Gamma clock and GMRF Bayesian Skyride coalescent tree prior. We then employed a Bayesian MCMC approach using BEAST v1.8.4 and performed five independent MCMC runs with up to 100 million generations to ensure the convergence of estimates. Trees were summarized in a maximum clade-credibility tree after a 10% burn-in [94] and used Tracer (http://beast.bio.ed.ac.uk/Tracer) to ensure convergence during MCMC by reaching effective sample sizes greater than 100.
To reduce the number of sequences from the original 862 downloaded from Genbank, a maximum likelihood tree was estimated using FastTree v2.1.7 [95] after identical alignment and curating methods. FastTree was run using GTR+Γ+I nucleotide model with 2000 Γ-rate categories, exhaustive search settings, with 5000 bootstrap replications using the Shimodaira-Hasegawa (SH) test. The analysis was repeated for the dataset of 95 sequences to compare tree topologies inferred by the Bayesian approach (S1 Fig). All alignments referred to in this manuscript can be found at https://github.com/caiofreire.
To perform this study and make it comparable with other studies [60,96], viral stocks were standardized in number of plaque forming units per milliliter (PFU/mL) for cell infections rather than copy numbers of genome. The growth kinetics assays were performed in 12-well plates using one plate per virus strain with one uninfected well as a negative control. Each well was seeded with 2.4x105 Ap61 or Vero cells in a volume of 400 μl of appropriate medium and infected with 2.4x103 PFU (plaque-forming unit) of virus in 400 μl of medium, resulting in a multiplicity of infection (MOI) of 0.01. After an incubation time of 4 hours, the medium was removed and replaced with 2 ml of new medium to set a zero point for the growth curves (T0). The harvesting of one well occurred at 22, 28, 50, 75, 99, 124, and 146 h post infection. Each harvest was performed as follows. Supernatants were removed and frozen at -80°C in small aliquots. Cells were washed once with phosphate-buffered saline (PBS) and then removed in 500 μl PBS. A volume of 20 μl of cell suspension was dried on a glass slide for a subsequent immunofluorescence assay as previously described [97] to measure viral proteins production. The remaining cell suspensions were frozen at -80°C.
RNA was extracted from cell suspensions and supernatants and copy numbers of genome were quantified by real time RT-PCR as previously described [98]. Infectious viral particles were measured in supernatants by plaque assay also as previously described [99]. This study was performed two times on each cell type. The initial titers of lineages 1, 2, 7 and 8 were respectively 3x108, 5x104, 7.5x106 and 1010 PFU/ml. For each lineage, 2.4x103 PFU were used for kinetics in mosquito and mammal cells. The ratio of particles per infectious unit in the initial viral stocks ranged from 8 to 600 [98]. Our viral stocks had a similar ratio of particles per infectious unit as that seen produced by fully infectious extracellular WN virus particles [100] and mosquito-derived replicon WN virus particles [101]. Variances in replication efficiency between studies observed during in vitro infection could be explained to differences in the viral strain and to the infection conditions i.e. very low MOIs (0.01), and distinct cell lines.
Mice were produced in the Institut Pasteur de Dakar farm, located in Mbao, approximately 15 kilometers from Dakar, Senegal. After one week of acclimatization, five-to-six-week-old Swiss mice were challenged by intraperitoneal (IP) injection with 100, 1000 and 10000 PFU of WNV lineages diluted in phosphate buffer saline + 0.2% endotoxin-free serum albumin (BSA). For each lineage and dose, two independent experiments of infection were made. Each individual experiment had 4 to 8 mice. A group of mice inoculated in parallel with an equivalent volume of phosphate buffer saline + 0.2% endotoxin-free serum albumin (BSA) was maintained as a control. Mice were kept on clean bedding and given food and water ad libitum. Infected animals were monitored daily for first signs of encephalitis (hunching, lethargy, eye closure, or hind legs paralysis) and death throughout the 21 days after infection. All statistical inferences were calculated using the Wilcoxon rank sum test.
To evaluate selection patterns on the complete coding sequences, we estimated the ratio of substitution rates (ω) per non-synonymous site (dN) over synonymous substitutions per synonymous site (dS) per codon sites. Briefly, sites with ω>1 are assumed to be under positive (diversifying) selection, and sites where ω<1 are undergoing negative (purifying) selection. When ω = 0, the site is undergoing neutral selection. To estimate ω, we applied three maximum likelihood methods: single likelihood ancestor counting (SLAC), fixed-effects likelihood (FEL), and internal fixed-effects likelihood (IFEL). We also investigated the presence of transient (episodic) selective pressures, using the mixed-effects model of evolution (MEME) [102] and fast, unconstrained Bayesian approximation (FUBAR) [103] approaches. For FEL, SLAC, IFEL, and MEME analyses, sites were identified as undergoing significant positive selection when p-value ≤0.10. For FUBAR, sites were identified as undergoing positive selection when there was a posterior probability ≥0.90. All estimations were implemented using HyPhy v2.11 [104].
Extraction of viral RNA from supernatants was performed with the QIAamp viral RNA mini kit (Qiagen, Heiden, Germany) according to manufacturer’s instructions. For cell fractions, prior to RNA extraction, cells were lysed by serial cycles of freeze/thaw. For the detection and quantification of viral RNA, a consensus WNV real-time RT-PCR assay and corresponding RNA standard were used as previously described [98]. The real-time PCR assays were performed using the Quantitect Probe RT-PCR Kit (Qiagen, Heiden, Germany) in a 96-well plate under the following conditions: 50°C for 10 min, 95°C for 15 min followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. Copy numbers of genome were calculated using Ct (Cycle threshold) and corresponding RNA standard.
Overlapping RT-PCRs were done to recover the complete genome. All primer sequences can be found in the S3 Table. The NS5, envelope and NS5-partial 3’UTR regions were first amplified using flavivirus consensus or West Nile specific primers [1,105,106], followed by amplification of NS3 region using designed WNV primers. The 5’ non-coding region of the genome was obtained using the 5’RACE kit (Invitrogen, Carlsbad, USA) and a designed consensus primer in the capsid protein for reverse primer. Finally, specific primers were designed according to the first sequences obtained and a second step of RT-PCR was done to obtain the complete genome.
The PCR fragments were obtained using AMV reverse transcription kit (Promega, Madison, USA) for reverse transcription and Go-Taq PCR kit (Promega, Madison, USA) for amplification. The RT conditions were set according to the manufacturer’s instructions, and the PCR conditions were as follows: 5 minutes at 95°C, 40 cycles of 1 minute at 95°C, 1 minute at 53°C, 1 to 4 minutes (according the size of the PCR product) at 72°C, and 10 minutes at 72°C. The PCR products were purified from the agarose gel using the Gel extraction kit (Qiagen) and sequenced by Cogenics (Beckman Coulter Genomics, Essex, UK).
Infected cells at different time points were dissolved in PBS and dropped on a glass slide. After complete drying, cells were fixed for at least 20 min in cold acetone, dried again, and then stored at -20°C until staining. Staining was done with a WNV-polyclonal mouse immune ascit diluted in PBS and incubated for 30 minutes at 37°C. After washing three times with PBS, cells were incubated with the second antibody (goat anti-mouse IgG, fluorescein isothiocyanate [FITC] conjugated Biorad), diluted 1:40 and blue Evans 1/100 in PBS, for 30 minutes at 37°C in the dark. The cells were washed again three times with PBS, dried, and covered with 50% glycerol in PBS. After dehydration, examination was done by fluorescence microscopy.
KJ831223, FJ159131, AY277251, FJ159130, FJ159129, AY765264, KY703856, DQ176636, HM147823, FJ425721, KT207791, KJ934710, KP780840, KP780839, KT359349, KC496016, KC407673, KF179639, KJ883346, KC496015, HQ537483, KF647251, KT207792, JN858070, KP109692, KF179640, KM203863, KP780838, KP780837, KM203861, KM203862, JN393308, EF429197, EF429198, HM147824, EF429199, KM052152, EF429200, GQ903680, HM147822, GQ851605, DQ256376, GQ851604, JX041632, GQ851602, KT934796, KT934801, JX123031, JX123030, KT934800, KT934802, KT934803, GQ851603, KT934797, KT934799, KT934798, JX041628, JX041629, JX041630, HM051416, KT163243, EU249803, KC601756, JX442279, JX041634, KU588135, JQ928175, JN858069, KF647253, KC954092, JQ928174, JX556213, JF707789, FJ766331, FJ766332, JF719069, FJ483549, FJ483548, JF719066, JF719067, KF234080, GU011992, JF719068, DQ786573, AY701413, HM152775, AY701412, GQ851606, GQ851607, GQ379161, AF196835, KY703855, EU082200, KY703854, AY532665.
|
10.1371/journal.ppat.1003691 | Prospective Antiretroviral Treatment of Asymptomatic, HIV-1 Infected Controllers | The study of HIV-infected “controllers” who are able to maintain low levels of plasma HIV RNA in the absence of antiretroviral therapy (ART) may provide insights for HIV cure and vaccine strategies. Despite maintaining very low levels of plasma viremia, controllers have elevated immune activation and accelerated atherosclerosis. However, the degree to which low-level replication contributes to these phenomena is not known. Sixteen asymptomatic controllers were prospectively treated with ART for 24 weeks. Controllers had a statistically significant decrease in ultrasensitive plasma and rectal HIV RNA levels with ART. Markers of T cell activation/dysfunction in blood and gut mucosa also decreased substantially with ART. Similar reductions were observed in the subset of “elite” controllers with pre-ART plasma HIV RNA levels below conventional assays (<40 copies/mL). These data confirm that HIV replication persists in controllers and contributes to a chronic inflammatory state. ART should be considered for these individuals (ClinicalTrials.gov NCT01025427).
| HIV-infected “controllers” are rare individuals who are HIV-seropositive but are able to maintain low levels of plasma HIV RNA in the absence of antiretroviral therapy (ART). There has been intense interest in characterizing these unique individuals because they have been considered as a potential model for a “functional cure” of HIV. Previously, our group has shown that controllers have elevated levels of T cell activation and accelerated atherosclerosis, suggesting that very low levels of viral replication may lead to disproportionately high levels of immune activation. However, the degree to which viral replication contributes to these outcomes is not known. We therefore conducted the first, prospective study of ART initiation in a cohort of asymptomatic HIV-infected controllers, in order to determine the virologic and immunologic effects of treating controllers with ART. Controllers had a significant decreases in ultrasensitive plasma HIV RNA, rectal HIV RNA, and markers of T cell activation/dysfunction in blood and gut mucosa with ART. Similar reductions were observed in the subset of “elite” controllers with extremely low pre-ART plasma HIV RNA levels (<40 copies/mL). These data suggest that HIV replication persists in controllers and contributes to a chronic inflammatory state.
| HIV-infected “controllers” are individuals who are HIV-seropositive but are able to maintain low levels of plasma HIV RNA in the absence of antiretroviral therapy (ART) [1]. These individuals are rare, comprising less than 1–7% of the HIV-infected population, depending upon the plasma HIV RNA criteria that are used to define the group [2], [3], [4]. Most controllers have evidence of strong host immune responses, which have been widely assumed to be responsible for durable viral control. Because knowledge regarding these protective immune responses might lead to novel interventions aimed at preventing or curing HIV infection, there has been intense interest in further characterizing these unique individuals.
Multiple groups have examined how HIV is controlled by these individuals [5], [6], 7,8,9. More recently, our group has focused on defining the potential clinical consequences of long-term, host-mediated, virologic control. We and others have shown that: (1) the vast majority of controllers have stable low-level viremia [10], [11]; (2) controllers have elevated levels of microbial translocation and T cell activation compared to HIV-negative and ART-suppressed individuals [12], [13]; (3) a minority (7–10%) of controllers with high levels of T cell activation progress immunologically to AIDS despite preservation of virologic control [12]; and (4) controllers have accelerated measures of atherosclerosis compared to HIV-negative individuals, even after adjustment for traditional cardiovascular risk factors [14], [15]. Collectively, these data suggest that very low levels of viral replication may lead to disproportionately high levels of immune activation in HIV-infected controllers, which may lead to an increased risk of AIDS- and non-AIDS defining events. However, the degree to which viral replication contributes to these outcomes is not known. No prospective ART studies have been performed in controllers, because it has generally been assumed that most controllers do not need ART due to their ability to control plasma viremia to very low levels.
We therefore conducted the first, prospective study of antiretroviral therapy in a cohort of asymptomatic HIV-infected controllers, in order to determine the virologic and immunologic effects of treating controllers with ART. We also measured changes in biomarkers of inflammation and coagulation. Multiple biomarkers (e.g., high sensitivity C-reactive protein and D-dimer) remain elevated in both untreated and treated non-controllers [16], and have been shown to be strongly predictive of morbidity and all-cause mortality in ART-treated non-controllers [17], [18], [19]. We therefore examined whether ART initiation led to a reduction in biomarkers of inflammation and coagulation in controllers, in order to assess whether low-level viral replication has any potential immunologic and clinical consequences in these individuals.
Sixteen asymptomatic controllers were prospectively treated with open-label raltegravir+tenofovir/emtricitabine for 24 weeks. Controllers were defined by the following inclusion criteria: (1) HIV-seropositive; (2) ART untreated; and (3) plasma HIV RNA <1,000 copies/mL for ≥12 months. Exclusion criteria included: (1) known rheumatologic conditions (e.g., systemic lupus erythematosus), because of the potential for biologic false-positive testing on HIV antibody tests; (2) known kidney disease; (3) known bone disease, including pathologic fractures; (4) chronic hepatitis B infection, because of the potential risk of liver abnormalities after starting and stopping tenofovir/emtricitabine in patients with chronic hepatitis B infection; (5) serious illness requiring hospitalization or parental antibiotics within the preceding 3 months; and (6) pregnant or breastfeeding women.
Subjects were seen every four weeks. Plasma and peripheral blood mononuclear cells (PBMCs) were collected and detailed interviews were conducted at the majority of visits. Thirteen out of 16 subjects consented to undergo 3 serial colorectal biopsies at weeks −2, 6, and 22. Five out of 16 subjects also underwent leukapheresis at weeks −4 and week 21 in order to obtain large PBMC samples for measurement of integrated HIV DNA. Adherence to study drug was measured at every study visit by self-report and pill-count. An independent Data Monitoring Committee comprised of three individuals from the scientific community met at 12, 24, 48, and 60 weeks after the enrollment of the first subject, and at 60 weeks after the enrollment of the last subject.
All subjects had a baseline plasma HIV RNA level <1,000 copies/mL in the absence of ART. The median baseline plasma HIV RNA level using a standard assay (Abbott Real Time assay, lower limit of detection <40 copy/mL) was 77 copies/mL; 4/16 subjects had an “undetectable” (<40 copies/mL) baseline plasma HIV RNA level with this assay. The median baseline plasma HIV RNA level using an ultrasensitive “single copy assay” (lower limit of detection <0.3 copy/mL) was 23 copies/mL. The median baseline age was 49 years; most subjects (88%) were men. The median baseline CD4+ and CD8+ T cell counts were 616 and 897 cells/mm3, respectively; the median baseline CD4+ to CD8+ T cell count ratio was 0.71. The median nadir CD4+ T cell count was 590 cells/mm3. The median self-reported duration of known HIV diagnosis was 10 years.
Antiretroviral therapy was well tolerated and all subjects completed 24 weeks of ART. No significant adverse events occurred during the study. The majority of controllers (11/16) elected to continue ART after the 24-week study period. Five out of 16 subjects elected to discontinue ART after the 24-week study period at various times (median 10.0 weeks, interquartile range [IQR] 1.0 to 19.0 weeks) after the end of the treatment study. They have subsequently been followed for a median 63.0 (IQR 47.4 to 66.3) weeks after discontinuing ART, and at the time of last follow-up the plasma HIV RNA level using a standard assay (Abbott Real Time assay, lower limit of detection <40 copy/mL) was a median <40 (IQR<40 to 73) copies/mL. Of the 5 subjects who elected to discontinue ART after the end of the treatment study, 4/5 of the subjects had an “undetectable” pre-ART plasma HIV RNA level at baseline using a standard assay (Abbott Real Time assay, lower limit of detection <40 copy/mL).
Controllers did not have a statistically significant increase in peripheral CD4+ T cell counts (mean 1.00-fold increase in CD4+ T cells at week 24, 95% confidence interval [CI] 1.05-fold decrease to 1.06-fold increase, p = 0.93) (Fig. 1A). Similarly, controllers did not have a statistically significant increase in %CD3+CD4+ T cells in the rectum (mean +0.4%, 95% CI −0.8% to 1.6%, p = 0.50) (Fig. 1B).
Despite having low pre-ART plasma HIV RNA levels by conventional assays, controllers had an early and persistent decrease in ultrasensitive plasma HIV RNA levels after initiation of ART (mean 66-fold decrease in S/Co at week 24, 95% CI 155 to 28-fold decrease, p<0.001) (Fig. 2A). In addition, we examined change in HIV antibody levels as a surrogate measure of antigenic stimulation and viral persistence [10], [20], [21], [22], [23]. Controllers also had an early and persistent decrease in HIV antibody levels (mean −7.2 S/Co at week 24, 95% CI −9.6 to −4.8, p<0.001) (Fig. 2B).
At baseline, the median (IQR) levels of cell-associated HIV RNA and total HIV DNA in PBMCs were 6.9 (3.5, 45.7) S/Co per million CD4+ T cells and 57 (34, 138) copies/million CD4+ T cells, respectively. In PBMCs, controllers did not have a substantial decrease in cell-associated HIV RNA (mean 1.20-fold decrease in S/Co per million CD4+ T cells, 95% CI 2.4-fold decrease to 1.62-fold increase, p = 0.58) or total HIV DNA (mean 1.22-fold decrease in copies/million CD4+ T cells, 95% CI 1.95-fold decrease to 1.32-fold increase, p = 0.41) at week 24. However, controllers did have an early and persistent decrease in rectal cell-associated HIV RNA after initiation of ART, with a mean decrease of 0.61 log10 copies/million CD4+ cells, which corresponded to a 4.1-fold decrease (95% CI 12.0 to 1.40-fold decrease, p = 0.010) at week 22 (Fig. 3A). There was a similar trend towards a decrease in rectal total HIV DNA, with a mean decrease of 0.28 log10 copies/million CD4+ cells, which corresponded to a 1.91-fold decrease (95% CI 5.1-fold decrease to 1.38-fold increase, p = 0.19) at week 22 (Fig. 3B). We also measured integrated HIV DNA levels in PBMCs obtained through leukapheresis in 5 controllers. In these subjects, there was a statistically significant decrease in integrated HIV DNA after initiation of ART, with a mean decrease of 0.32 log10 copies/million PBMCs, which corresponded to a 2.1-fold decrease (95% CI 2.7 to 1.13-fold decrease, p = 0.027) at week 21 (Fig. 4).
Markers of T cell activation/dysfunction in blood and gut also decreased substantially with ART. In PBMCs, controllers had a mean decrease of 1.9% in %CD38+HLA-DR+ CD4+ T cells (95% CI −2.8% to −0.9%, p<0.001) and a mean decrease of 9.0% in %CD38+HLA-DR+ CD8+ T cells (95% CI −12.4% to −5.6%, p<0.001) at week 24 (Fig. 5). Controllers also had a mean decrease of 1.6% in %PD-1+ CD4+ T cells (95% CI −3.1% to −0.1%, p = 0.04) and a mean decrease of 4.5% in %PD-1+ CD8+ T cells (95% CI −6.4% to −2.6%, p<0.001) in PBMCs at week 24. In the rectum, controllers had a trend towards a decrease in %CD38+HLA-DR+ CD4+ T cells (mean −0.9%, 95% CI −2.3% to +0.5%, p = 0.20) and a statistically significant mean decrease of 12.2% in %CD38+HLA-DR+ CD8+ T cells (95% CI −21.9% to −2.5%, p = 0.014) at week 22 (Fig. 6).
At baseline, the median (IQR) levels of high sensitivity C-reactive protein (hsCRP), interleukin-6 (IL-6), soluble CD14 (sCD14), and D-dimer were 1.20 (0.55, 3.03) ug/mL, 1.71 (1.28, 4.42) pg/mL, 1696 (1446, 1971) ng/mL, and 0.37 (0.28, 0.56) ug/mL, respectively. After ART initiation, there was a trend towards a decrease in hsCRP, with a mean 1.74-fold decrease (95% CI 3.2-fold decrease to 1.04-fold increase, p = 0.069) at week 4, and a mean 1.67-fold decrease (95% CI 3.0-fold decrease to 1.09-fold increase, p = 0.093) at week 24 (Fig. 7). We also observed similar trends in IL-6 (mean 1.34-fold decrease, 95% CI 2.1-fold decrease to 1.19-fold increase, p = 0.22), sCD14 (mean −44.2, 95% CI −138.6 to +50.3, p = 0.36), and D-dimer (mean 1.30-fold decrease, 95% CI 2.1-fold decrease to 1.25-fold increase, p = 0.29) levels after 24 weeks of ART, although these trends did not reach statistical significance.
At baseline, the median (IQR) levels of percentage of Gag-specific IFNγ+IL2+ CD4+ and CD8+ T cell responses in PBMCs were 0.10% (0.03%, 0.17%) and 1.53% (0.37%, 2.55%), respectively. Controllers did not have a substantial change in the percentage of Gag-specific IFNγ+IL2+ CD4+ T cell responses in PBMCs (mean 1.07-fold decrease, 95% CI 1.40-fold decrease to 1.22-fold increase, p = 0.62), although there was a trend towards a decrease in the percentage of Gag-specific IFNγ+IL2+ CD8+ T cell responses in PBMCs (mean 1.28-fold decrease, 95% CI 1.68-fold decrease to 1.03-fold increase, p = 0.075) at week 24. At baseline, the median (IQR) levels of percentage of total (IFNγ, IL-2, TNFα, and/or CD107a) Gag-specific CD4+ or CD8+ in the rectum were 0.56% (0%, 1.3%) and 0.22% (0.08%, 0.32%), respectively. Controllers did not have a substantial change in the percentage of total Gag-specific CD4+ (mean 1.26-fold increase, 95% CI 1.82-fold decrease to 2.9-fold increase, p = 0.58) or CD8+ (mean 1.61-fold decrease, 95% CI 3.6-fold decrease to 1.37-fold increase, p = 0.24) T cell responses in the rectum at week 22.
At baseline, 4/16 controllers had an “undetectable” pre-ART plasma HIV RNA level using a standard assay (Abbott Real Time assay, lower limit of detection <40 copy/mL). Despite having this extremely low pre-ART plasma HIV RNA level, this subset of so-called “elite” controllers had a statistically significant decrease in ultrasensitive plasma HIV RNA levels after initiation of ART (mean 14-fold decrease in S/Co at week 24, 95% CI 115 to 1.74-fold decrease, p = 0.013) (Fig. 8A). Similarly, this subset of controllers had a statistically significant decrease in HIV antibody levels (mean −4.2 S/Co at week 24, 95% CI −7.9 to −0.5, p = 0.027) (Fig. 8B). Finally, we observed similar trends in immune activation in these 4 controllers after initiation of ART. There was a mean decrease of 6.0% in %CD38+HLA-DR+ CD8+ T cells in PBMCs at week 24 (95% CI −13.0% to +0.10%, p = 0.091, Fig. 8C) and a mean decrease of 24.3% in %CD38+HLA-DR+ CD8+ T cells in the rectum at week 22 (95% CI −54.1% to +5.6%, p = 0.11, Fig. 8D).
In this first prospective study of antiretroviral therapy initiation in asymptomatic HIV-infected controllers, 24 weeks of ART was safe and well-tolerated. Despite being able to maintain very low plasma HIV RNA levels in the absence of ART, controllers had readily measurable levels of HIV RNA and DNA in the gut. Antiretroviral therapy led to statistically significant decreases in ultrasensitive plasma HIV RNA levels, HIV antibody levels, rectal cell-associated HIV RNA, and immune activation in the blood and gut. Collectively, these data suggest that HIV in most controllers is replication-competent [24], [25], [26], and that host rather than virologic factors account for the remarkable degree of viral control in these unique individuals. We also observed a statistically significant decrease in levels of integrated HIV DNA with ART, while total HIV DNA levels remained stable. These findings may be due to an excess of unintegrated HIV DNA in controllers, as previously reported by our group [27].
We observed that measures of immune activation/dysfunction decreased as measures of virologic burden and HIV antigenic stimulation decreased with ART. We also observed a trend towards a decrease in hsCRP (a measure of systemic inflammation) with ART; similar trends were observed with IL-6, sCD14, and D-dimer. These biomarkers have been shown to be strong, consistent, and independent predictors of increased morbidity and mortality in HIV infection [18], [19], [28]. Because the confidence intervals were wide, however, we cannot assess with certainty whether the observed decrease in hsCRP levels has any clinical relevance; it would be important to pursue these findings in future, larger studies. Taken together, however, these data suggest that there may be immunologic consequences to even very low levels of viral replication. This latter finding may have important implications for HIV-infected non-controllers as well [29], [30], [31], [32], [33].
Importantly, our study also shifts the field's working definition of a “functional cure.” On one hand, our data suggest that a complete block of viral replication is not necessary to achieve long-term virologic control. However, natural long-term virologic control appears to be coming at an immunologic and/or clinical “cost,” at least as defined by increased levels and manifestations of immune activation. Thus, although further study of controllers is warranted, untreated HIV-infected controllers may not represent the best model of a functional cure, if we believe that a cure should require a disease-free (and not just treatment-free) state.
Several limitations of our study deserve comment. First, this was a small pilot study and our findings should be replicated in a larger study of HIV-infected controllers with a longer duration of follow-up. In our study of controllers who had relatively high baseline CD4+ T cell counts, 24 weeks of ART did not appear to confer a CD4+ T cell count benefit. In studies of HIV-infected non-controllers, a greater absolute decrease in plasma HIV RNA levels during the early period after ART initiation has been shown to be a consistent predictor of an early increase in CD4+ T cell counts (with much of the increase assumed to be due to redistribution) [34]. In our study, although we did observe a significant decrease in ultrasensitive plasma HIV RNA levels with initiation of ART, the absolute change was small compared to that observed in HIV-infected non-controllers; this may have partially accounted for the limited changes in peripheral CD4+ T cell counts. It is possible that with a much longer duration of follow-up, an increase in CD4+ T cell count may have been observed. Second, there was a trend towards a decrease in the percentage of Gag-specific IFNγ+IL2+ CD8+ T cell responses in PBMCs, although a similar trend was not observed in the rectum. This observation raises the possibility that initiation of ART in controllers may reduce host mechanisms of virologic control, leading to rebound in viremia if ART is discontinued. However, in the 5 subjects who elected to discontinue ART after the 24-week study period, there was no evidence of rebound in plasma viremia after discontinuation of ART. Nevertheless, the long-term safety of ART in controllers should be confirmed. Third, we enrolled a relatively heterogeneous group of controllers. As we and others have shown, however, controllers are a heterogeneous group with varying levels of steady-state viremia; there appears to be a continuum of viremia across controllers [1], [10], [12], . In order to determine whether there is a differential effect of ART on a spectrum of controllers, we enrolled individuals whose baseline plasma HIV RNA levels spanned from nearly 0 to 1000 copies RNA/mL for at least 12 months (median duration of HIV diagnosis 10 years, IQR 4.5 to 24 years). Thus, our study included controllers who had both low-level but detectable and undetectable pre-ART plasma HIV RNA levels using conventional assays. Remarkably, however, even amongst the latter group of “elite” controllers who had undetectable pre-ART plasma HIV RNA levels at baseline, we observed a statistically significant decrease in ultrasensitive plasma HIV RNA levels and HIV antibody levels, and a trend towards a decrease in immune activation with ART. Fourth, although 24 weeks of ART significantly decreased levels of CD4+ and CD8+ T cell activation, it did not normalize them to levels observed in HIV-uninfected individuals [41]. Thus, at least in HIV-infected controllers, low-level viral replication is unlikely to be the only factor contributing to immunologic disease. The potential role of other factors that might contribute to immune activation—including co-infections and substance abuse—could not be addressed in this pilot study, but might be addressed in future studies with larger cohorts. It would also be important to systematically assess the individual and potentially synergistic contributions of ART and lifestyle modifications towards decreasing inflammation, immune activation, and clinical disease in HIV-infected controllers [14]. Finally, it is worth noting that there may be multiple pathways to virologic control, some of which may represent an appropriate model of a “functional cure” and may not receive an additional benefit from ART.
In summary, 24 weeks of ART was safe and well-tolerated in chronically HIV-infected controllers. Antiretroviral therapy in controllers led to significant decreases in ultrasensitive plasma and rectal HIV RNA, HIV antibody levels, and markers of immune activation/dysfunction in blood and gut, confirming that HIV replication persists in controllers and contributes to a chronic inflammatory state. We acknowledge that this was a small pilot study and that our findings would be ideally replicated in a larger, randomized, clinical-endpoint study. However, the relative rarity of HIV-infected controllers may make such a study impractical, if not impossible. In the absence of such a study, clinicians will need to weigh the potential benefits of ART (suggested by the changes in immune activation and biomarkers observed in our study) with the potential risks and costs associated with long-term antiretroviral therapy.
All subjects provided written informed consent. This study was approved by the University of California San Francisco (UCSF) Committee on Human Research.
The isothermal Transcription Mediated Amplification (TMA) assay (Aptima, Gen-Probe/Hologic) was used to measure ultrasensitive plasma HIV RNA levels at weeks 0, 4, 12, and 24. This is a nucleic acid-amplification test that has been FDA-approved for the early detection of HIV infection in blood donors [42], [43], [44]. It is a highly specific and sensitive assay, with a singlicate 50% detection limit of 3.6–14 copies/mL [45], [46]. The assay was performed in triplicate on 0.5 mL plasma (1.5 mL total plasma), improving the overall 50% detection limit to <5 copies/mL. The output is a signal/cutoff (S/Co) ratio (range 0–30), with S/Co<1.0 = “negative” and S/Co≥1.0 = “positive.” Ultrasensitive plasma HIV RNA levels were also measured at weeks 0 and 12 with a “single copy assay” (lower limit of detection <0.3 copy/mL), using a median 7.3 mL of plasma [47].
A “de-tuned” or less-sensitive enzyme immunoassay (LS-VITROS) was used to measure HIV antibody levels at weeks 0, 4, 12, and 24. The VITROS (Ortho-Clinical Diagnostics) is an FDA-approved diagnostic assay for the detection of IgM/IgG antibodies to HIV-1/-2. The less-sensitive modification tests 1∶400 dilutions of plasma and calculates a S/Co ratio (range 0–80), and has been validated as a method to identify early HIV infection [48].
Cell-associated HIV RNA and total HIV DNA were measured from PBMCs at weeks 0, 4, and 24. Cell-associated HIV RNA was measured using modifications of published methods (Aptima, Gen-Probe/Hologic) [10], [49]. The output is a S/Co ratio (range 0–30), with S/Co<1.0 = “negative” and S/Co≥1.0 = “positive.” All S/Co ratios were normalized to per million CD4+ T cells. Total HIV DNA was measured using modifications of published methods with an overall sensitivity of 1 copy/3 µg of DNA (450,000 PBMCs) [10], [50], [51], [52]. All total HIV DNA levels were normalized to per million CD4+ T cells.
Integrated HIV DNA was measured from PBMCs at weeks −4 and 21. DNA was prepared (Qiagen Mid) and integrated HIV DNA was measured using a published repetitive sampling method because integration levels are known to be low in controllers [27], [53]. At least 42 Alu-gag PCR reactions were performed with 150,000 diploid genomes per PCR, for a total of 6.3 million diploid genomes assayed per subject.
PBMCs were isolated from whole blood, cryopreserved, and stored at the UCSF AIDS Specimen Bank. Markers of T cell activation/dysfunction and antigen-specific T cell responses were measured at weeks 0, 4, and 24 at the UCSF Core Immunology Laboratory, using published methods that have been optimized and validated for cryopreserved PBMCs [54]. Briefly, cryopreserved PBMCs were rapidly thawed in warm media, counted on an Accuri C6 (BD Biosciences) with the Viacount assay (Millipore), and washed and stained the same day (T cell immunophenotyping) or rested overnight (cytokine flow cytometry [CFC]). The average viability of thawed cells was 93% (range 61–98%; 80% of samples had viability >90%).
For T cell immunophenotyping, the percent of activated (CD38+/HLA-DR+/PD1+) CD4+ and CD8+ T cells were measured; these markers of immune activation/dysfunction have been shown to be strong and independent predictors of HIV disease progression [12], [41], [55], [56], [57]. Cells were stained with Aqua Amine Reactive Dye (AARD, Invitrogen) to discriminate dead cells, washed, and stained with fluorescently-conjugated monoclonal antibodies: CD3-Pacific Blue (BD Pharmingen), CD38-PE, HLA-DR-FITC, PD1- Alexa647 (BD Biosciences), CD4-PE Texas Red, and CD8-QDot 605 (Invitrogen). In each experiment a fluorescent-minus one control was included for CD38, HLA-DR, and PD-1. Stained cells were washed, fixed in 0.5% formaldehyde (Polyscience), and held at 4C until analysis.
For CFC, rested PBMCs were stimulated for 18–22 h at 37C with overlapping peptide pools corresponding to HIV-1 Con B Gag peptides (NIH 8117) in the presence of 0.5 ug/mL Brefeldin A and 0.5 ug/mL Monensin (Sigma-Aldrich). A control well with no stimulation was run in parallel for each sample. Cells were washed and stained with AARD, fixed, and permeabilized for intracellular staining with antibodies against CD3-Pacfic Blue, IFNγ-FITC, IL-2-PE (BD BioScience), CD4-PE Texas Red, and CD8-QDot 605 (Invitrogen). Cells were washed and stored at 4C until analysis. We focused on Gag-specific IFNγ+IL2+ T cell responses given that we have shown that these responses are associated with control of HIV replication in controllers [5], [35], [58].
Stained cells were run on a customized BD LSR II (BD Bioscience). 100,000 and 500,000 lymphocytes were collected for immunophenotyping and CFC samples, respectively. Data were compensated and analyzed using FlowJo (Tree Star) to determine the proportion of CD4+ and CD8+ T cells expressing each of the T cell or cytokine markers. Combinations of markers were calculated in FlowJo using the Boolean gate function. For CFC data, results from control wells with no stimulation were subtracted from stimulated results.
High sensitivity C-reactive protein (hsCRP), interleukin-6 (IL-6), soluble CD14 (sCD14), and D-dimer levels were measured on stored fasting plasma samples at weeks 0, 4, and 24 at the Laboratory for Clinical Biochemistry Research at the University of Vermont. hsCRP was measured with a BN II nephelometer (Siemens Diagnostics, Deerfield, IL), IL-6 was measured with Chemiluminescent Sandwich enzyme-linked immunosorbent assay, sCD14 with a standard ELISA (both R&D Systems, Minneapolis, MN), and D-dimer was measured with an immunoturbidometric method on the Sta-R analyzer, Liatest D-DI (Diagnostica Stago, Parsippany, NJ). Interassay coefficients of variation for a number of different control materials of different values averaged ∼10% or less for all assays.
Thirty colorectal biopsy specimens were obtained 10–20 cm from the anal verge using 3 mm jumbo forceps at weeks −2, 6, and 22. Eighteen to 24 biopsy pieces were placed into 10 mL RPMI-1640 media containing fetal calf serum (15%), penicillin (100 U/mL), streptomycin (100 ug/mL), and L-glutamine (2 mM). Fresh colorectal cells were isolated on the same day using a modification of a published protocol designed to optimize yield and viability of mucosal lymphocytes without compromising the detection of most surface antigens [59]. Briefly, biopsy pieces underwent two rounds of digestion in 0.5 mg/mL collagenase type II (Sigma-Aldrich). Each digestion was followed by disruption of the tissue with a syringe bearing a 16-gauge blunt end needle and subsequent passage through a 70 µm cell strainer. Yields were 9.5–31×106 (mean 18×106) total rectal cells. One aliquot of cells was set aside for flow cytometry and stained with CD45-FITC, CD3-APC and CD4-PE (BD biosciences) for 15 min at 25C. Propidium iodide was added to stain non-viable cells and samples were run on an Accuri C6 to determine the total number of viable mononuclear cells and proportion and absolute number of viable CD45+ leukocytes and CD4+ T cells. Another aliquot of cells was frozen at −80C for subsequent nucleic acid extraction.
Total HIV RNA was measured from rectal cells using a published method [40]. Three replicates of up to 500 ng RNA were assayed for total processive HIV RNA transcripts using primers (HXB2 positions 522–543, 626–643) and probe (559–584) from the LTR region [60]. Genomic HIV RNA standards (2.5×100 to 2.5×105) were prepared from lab stocks of NL4-3 virions by extracting and quantifying HIV RNA using the Abbot Real Time assay. HIV RNA copy numbers were normalized to cellular input into the PCR, as determined by RNA mass (assuming 1 ng RNA = 1000 cells [61]), which has been shown to correlate with levels of GAPDH RNA [62]. Results were further normalized by the percent of cells that were CD3+CD4+ by flow cytometry and expressed as copies/106 CD4+T cells.
Total HIV DNA was measured from rectal cells using a published method [40]. Three replicates of up to 500 ng DNA were assayed for HIV DNA using a modification of a published TaqMan PCR assay that uses primers/probe from the LTR region (as above). External standards (105 to 1) were prepared from DNA extracted from known numbers of 8E5 cells (NIH AIDS Reagent Program), each of which contains one integrated HIV genome per cell. HIV DNA copy numbers were normalized to cellular input into the PCR, as determined by DNA mass (assuming 1 ug DNA = 160,000 cells). Results were further normalized by the percent of cells that were CD3+CD4+ by flow cytometry and expressed as copies/106 CD4+T cells.
Markers of T cell activation (CD38+/HLA-DR+) and total Gag-specific responses (Gag-specific CD4+ and CD8+ T cells expressing one or more of IFNγ, IL-2, TNFα, and/or CD107a [63], [64], [65]) were measured from rectal cells at weeks −2, 6, and 22. We focused on these responses given that we have shown that these mucosal T cell responses are associated with control of HIV replication in controllers [36].
For T cell immunophenotyping of freshly isolated rectal cells, similar methods were used as for PBMCs [59]. For CFC, freshly isolated rectal cells were rested overnight at 37C, 5%CO2, in R15 containing 0.5 mg/mL piperacillin-tazobactam, then similar methods were used as for PBMCs [59]. To account for the lower numbers of events and elevated baseline cytokine staining in mucosal samples, response data from peptide-stimulated wells were first compared against unstimulated controls using a published algorithm to determine statistical significance, prior to background subtraction [36], [66].
Mixed effect linear models with random slopes and intercepts were used to examine change in virologic and immunologic measurements over time. Changes in integrated HIV DNA levels were assessed by estimating the mean change and its bias-corrected and accelerated non-parametric confidence intervals, and using a paired t-test to obtain a corresponding p-value [67]. All statistical analyses were conducted with Stata version 11.1 (Stata Corp).
|
10.1371/journal.ppat.1007571 | PolyGlcNAc-containing exopolymers enable surface penetration by non-motile Enterococcus faecalis | Bacterial pathogens have evolved strategies that enable them to invade tissues and spread within the host. Enterococcus faecalis is a leading cause of local and disseminated multidrug-resistant hospital infections, but the molecular mechanisms used by this non-motile bacterium to penetrate surfaces and translocate through tissues remain largely unexplored. Here we present experimental evidence indicating that E. faecalis generates exopolysaccharides containing β-1,6-linked poly-N-acetylglucosamine (polyGlcNAc) as a mechanism to successfully penetrate semisolid surfaces and translocate through human epithelial cell monolayers. Genetic screening and molecular analyses of mutant strains identified glnA, rpiA and epaX as genes critically required for optimal E. faecalis penetration and translocation. Mechanistically, GlnA and RpiA cooperated to generate uridine diphosphate N-acetylglucosamine (UDP-GlcNAc) that was utilized by EpaX to synthesize polyGlcNAc-containing polymers. Notably, exogenous supplementation with polymeric N-acetylglucosamine (PNAG) restored surface penetration by E. faecalis mutants devoid of EpaX. Our study uncovers an unexpected mechanism whereby the RpiA-GlnA-EpaX metabolic axis enables production of polyGlcNAc-containing polysaccharides that endow E. faecalis with the ability to penetrate surfaces. Hence, targeting carbohydrate metabolism or inhibiting biosynthesis of polyGlcNAc-containing exopolymers may represent a new strategy to more effectively confront enterococcal infections in the clinic.
| Enterococcus faecalis is a microbial inhabitant of the human gastrointestinal tract that can cause lethal infections. Typically classified as a non-motile bacterium, E. faecalis can readily migrate and translocate across epithelial barriers to invade distant organs. Nevertheless, the molecular pathways driving enterococcal invasive attributes remain poorly understood. In this study, we uncover that E. faecalis produces a polyGlcNAc-containing extracellular glycopolymer to efficiently migrate into semisolid surfaces and translocate through human epithelial cell monolayers. Our work provides evidence that non-motile bacterial pathogens can exploit endogenous carbohydrate metabolic pathways to penetrate surfaces. Thus, targeting glycopolymer biosynthetic programs might be useful to control infections by Gram-positive cocci in the clinic.
| Microbes use a variety of strategies to obtain nutrients and ensure survival. While motility could be used as a means for accessing nutrient sources, non-motile bacterial species require unconventional mechanisms to accomplish this goal. Enterococcus faecalis is a non-motile, facultative anaerobic bacterium that inhabits the human gastrointestinal (GI) tract [1]. However, hypervirulent E. faecalis strains resistant to multiple antibiotics often cause hospital-acquired urinary tract, wound and abdominal infections, as well as bacteremia and infective endocarditis [1]. Enterococci can adhere to and invade host tissues in order to act as lethal pathogens. Indeed, E. faecalis translocation across the intestinal barrier enables bacterial spread and colonization of distal anatomical sites [2, 3]. Interestingly, E. faecalis extra-intestinal translocation appears to be promoted by association with epithelial cells in aggregates [3], a process that is partly mediated by the synthesis of adhesins [3–5] and additional unknown factors.
Enterococci produce diverse cell wall-anchored polysaccharides [6–8], which generally consist of repeating units of oligosaccharides that are associated with bacterial surfaces through linkage to cell membrane, peptidoglycan or other unknown mechanisms [7, 9]. E. faecalis displays an extensive surface glycome, including wall teichoic acid and lipoteichoic acid polymers, capsular polysaccharides [7, 10], and the enterococcal polysaccharide antigen (EPA), which is a rhamnopolysaccharide. EPA, mainly composed of glucose, rhamnose, N-acetylglucosamine (GlcNAc), N-acetylgalactosamine (GalNAc), and galactose that appears to be buried within the cell wall, thus precluding interaction with host cells [7]. In addition to these polysaccharides, a new glycopolymer was recently discovered on the surface of E. faecalis cells by immunofluorescence assays using the human IgG monoclonal antibody (mAb) F598, which specifically binds to β-1,6-linked GlcNAc polysaccharides [8, 11]. While this putative polyGlcNAc-like polymer has not been studied in E. faecalis, other reports have characterized similar polysaccharides that react with mAb F598 [12, 13], termed either PIA (for polysaccharide intercellular adhesin) in Staphylococcus epidermidis [14, 15] or PNAG (for polymeric N-acetyl glucosamine) in Staphylococcus aureus and other pathogens [16, 17]. Of note, these extracellular glycopolymers consist of β-1,6-linked GlcNAc residues containing 5–10% positively-charged amino groups (due to partial de-N-acetylation (GlcNH3)) as well as negative charges (resulting from O-succinylation) [8, 14–17].
E. faecalis has been shown to invade surfaces such as mammalian tissues [3, 18] and penetrate solid culture media [19], but the mechanisms driving these processes remain elusive. In the present study, we identified the molecular events and metabolic pathways that endow E. faecalis with remarkable capacity to penetrate semisolid surfaces. We found that E. faecalis produces extracellular polyGlcNAc-containing polymers to form penetrating microcolonies inside semisolid surfaces. Using diverse genetic and biochemical approaches, we determined that biosynthesis of these complex exopolymers occurs through the RpiA-GlnA-EpaX metabolic pathway. Notably, E. faecalis mutants unable to produce polyGlcNAc-containing polymers demonstrated impaired capacity to pass into semisolid surfaces and translocate through human epithelial cell monolayers.
Analysis of semisolid media penetration has been useful for identifying and characterizing virulence traits in human fungal pathogens [20–23]. We sought to exploit this approach to understand the molecular mechanisms that E. faecalis utilizes to enter surfaces. Under our conditions, an indelible bacterial “colony-print” developed inside modified MOLP (medium optimal for lipopeptide production) [24], when six-day-old colonies of the clinical isolate E. faecalis were extensively washed with water to remove adventitiously associated bacterial cells from the surface of the agar. This colony-print, indicative of penetration, was not observed at agar concentrations above 1.0% (Fig 1A), and agar degradation was not evidenced once the external (non-penetrating) cells were removed. Importantly, we identified this semisolid agar-penetration trait in several clinical isolates as well as in the commensal-like strain E. faecalis OG1RF (S1A Fig). Kinetic analyses revealed that E. faecalis agar entrance was macroscopically evident 48 h post-inoculation, and that it progressed concomitantly with colony expansion (S1B Fig). Of note, the agar penetration ability of all E. faecalis strains tested in this study was not lost upon laboratory domestication. This phenomenon was not only evidenced when E. faecalis colonies were grown on culture media solidified with agarose, but also with the copolymer poloxamer-407 (MOLP-407) (S1C and S1D Fig; top), demonstrating that this process is not specific to the nature of the gelling agent.
To quantify the number of penetrating and non-penetrating bacteria, colony forming units (CFUs) were determined from both inside and outside E. faecalis populations grown on semisolid MOLP. Some of the penetrating bacteria were able to form colonies, but the total number of CFUs obtained for this population (~5x108) was reproducibly one order of magnitude lower than that of the external cells (~6x109; Fig 1B). Similarly, the total CFUs obtained from the penetrating population grown on MOLP-407 were lower (~2x108) than the CFUs obtained for external cells (~1x109; S1D Fig, bottom). Flow cytometric analyses of penetrating and non-penetrating bacteria grown on MOLP were performed to determine the viability of these cells after staining with the live/dead dye Brilliant Violet-570 (BV-570). In contrast to the heat-killed control showing less than 5% viable cells, ~90% of the penetrating and non-penetrating population were viable under our culture conditions (Fig 1C). Together, these data indicate that E. faecalis can pass into semisolid surfaces and that the majority of the penetrating cells remain viable during this process.
To further understand the E. faecalis penetration process, we analyzed agar side sections of approximately 2 mm-wide obtained from six-day-old colony-prints produced by penetrating bacteria grown in MOLP. Interestingly, isolated aggregates with varying morphologies and sizes readily formed inside the agar (Fig 1D; top). Penetration was found to decrease proportionally from the center (S1E Fig; depth of ~128 μm) to the edge (S1F Fig; depth of ~6 μm) of the colony, with several aggregates penetrating deeper at the center (Fig 1D; top and S1E Fig). DAPI staining (Fig 1D; middle) and epifluorescence analysis of penetrating bacterial cells constitutively expressing m-Cherry (Fig 1D; bottom) confirmed that these internal clumps contained viable and metabolically active E. faecalis cells.
Scanning electron microscopy (SEM) analyses of E. faecalis colonies were performed to determine the morphological status of external and MOLP-penetrating cells. No major changes in cell morphology were found and only normal diplococcal, clumped or isolated, bacterial cells were observed. Strikingly, however, cells within the aggregates were covered with and connected by an extracellular matrix that appeared to be more abundantly produced by invading cells than surface cells (Fig 1E). Indeed, automated SEM image analysis (see Material and Methods) determined that internal cells exhibited significantly higher matrix coverage than external cells (Fig 1F). These data indicate that E. faecalis penetrates the agar surface and that this process is accompanied by the generation of microcolonies formed by matrix-covered cells.
Cellular structures, such as pili, have been shown to be involved in mediating Gram-positive bacterial motility [25]. To determine whether pili expression could mediate the penetration process observed, we tested two previously generated pili-deficient E. faecalis mutants (ΔebpABC and ΔebpA), and their parental OG1RF strain [26], under our conditions. After 6 days of growth on MOLP, we found that Ebp mutants and the wild-type (WT) strain exhibited similar penetration capacities, and only a slight change on the shape of the colony-print was evidenced in the absence of pili (S2A Fig). These data suggest that the bacterial migration process is mediated by an Ebp pilus-independent mechanism.
To further understand the E. faecalis semisolid surface penetration process, we performed genetic screening of a Mariner transposon insertion library. We sought to identify mutants that developed normal colonies above the agar, but were impaired in their semisolid surface-penetration capacity. Out of approximately 6,000 mutants screened, seven were found to be defective in penetration. Five of the seven mutants identified exhibited substantial growth defects and were thus excluded from subsequent studies. Of the remaining two mutants that were unable to generate WT-like colony-prints inside the agar (Fig 2A), we determined they had transposon inserts in either the glnA (glutamine synthetase) or rpiA (ribose-5-phosphate isomerase) genes. The glnA::TnM, but not the rpiA::TnM strain, exhibited a slight growth defect in liquid MOLP (S2B Fig), but both strains formed external colonies similar in size to those of WT (Fig 2A). The external and internal numbers of bacterial cells were next determined by differential CFU analysis. No significant differences were found in the CFU counts of the glnA::TnM and rpiA::TnM mutants on the agar surface in comparison with their paternal strain (S2C Fig and S1 Table). However, inside the semisolid surface, the rpiA::TnM and glnA::TnM mutants exhibited a significant reduction in the CFU counts in comparison with their parental strain (6x109; Fig 2B), consistent with the visible decrease in the colony-print generated by these two transposon mutants (Fig 2A). Genetic complementation with plasmids expressing either RpiA or GlnA correspondingly restored the invading phenotype of these mutant strains (Fig 2A and 2B). SEM analysis of external cells of six-day-old colonies revealed that while all strains exhibited diplococcal morphology, rpiA::TnM cells were bigger than both WT and glnA::TnM cells (Fig 2C). Most importantly, the extracellular matrix normally covering and connecting WT E. faecalis cells (Figs 1E and 2C) was either decreased or almost absent in glnA::TnM or rpiA::TnM mutants, respectively (Fig 2C and 2D). Together, these results indicate that GlnA and RpiA are necessary for efficient penetration of E. faecalis into semisolid surfaces, and that mutants lacking these genes also failed to produce the extracellular matrix that naturally covers the WT cells.
We next determined the molecular mechanisms by which GlnA and RpiA promote E. faecalis semisolid surface penetration. Since both enterococcal mutants unable to pass into agar failed to produce the extracellular matrix evident in the colony-prints of the parental strain (Figs 1E, 1F, 2C and 2D), we hypothesized that extracellular factors produced by WT cells could restore penetration by strains lacking GlnA and RpiA. To test this idea, WT cells expressing m-Cherry were independently mixed with GFP-labeled glnA::TnM or rpiA::TnM mutants, and colony-prints were analyzed after 6 days via fluorescence stereomicroscopy. WT invading cells formed a bright red fluorescent colony-print, whereas monocultures of glnA::TnM or rpiA::TnM showed decreased invasion capacity as evidenced by minimal GFP-derived fluorescence. However, a remarkable increase in GFP-positive invading cells was found when either mutant was co-cultured with WT cells (Fig 2E). These observations were consistent with the higher relative fluorescent intensity (RFI) observed with the colony-prints from mutants co-cultured with WT than those obtained from the monoculture area (Fig 2F). Hence, extracellular factors produced by the WT strain can restore the agar penetration defects intrinsic to the glnA::TnM and rpiA::TnM mutant cells.
GlnA and RpiA participate in key metabolic pathways (Fig 2G). GlnA plays an essential role in the metabolism of nitrogen by catalyzing the condensation of glutamate and ammonia (NH3) to generate glutamine [27]. RpiA catalyzes the reversible conversion of ribose-5-phosphate to ribulose-5-phosphate, a central enzymatic reaction in the pentose phosphate pathway [28]. We hypothesized that the metabolic functions of GlnA and RpiA may converge in the hexosamine biosynthetic pathway, where the glutamine produced by GlnA, together with fructose-6-phosphate generated from metabolites of the pentose phosphate pathway, could promote the formation of glucosamine-6-phosphate [29, 30]. We further postulated that decreased ability to penetrate MOLP by glnA::TnM and rpiA::TnM mutants could be a consequence of alterations in the levels of intracellular hexosamine biosynthetic pathway metabolites (Fig 2G). To test these hypotheses, we performed a metabolic complementation assay by supplementing exogenous substrates related to this pathway. Semisolid surface penetration by WT cells was not altered by addition of any of the substrates to the medium (Fig 2H). However, exogenous glutamine, glucosamine and GlcNAc, but not glutamate or fructose, rescued the defective penetration phenotype of glnA::TnM cells. In addition, penetration of rpiA::TnM mutants was restored by fructose, glucosamine or GlcNAc supplementation (Fig 2H). These data suggest that low availability of cellular fructose-6-phosphate (in rpiA::TnM) or glutamine (in glnA::TnM) compromises the hexosamine biosynthetic pathway and consequently, decreases the cellular levels of products of this pathway such as glucosamine-6P and UDP-GlcNAc that are required for enterococcal migration into semisolid MOLP.
We postulated that products derived from UDP-GlcNAc could mediate agar penetration by E. faecalis. Interestingly, we observed that E. faecalis readily produced extracellular GlcNAc-containing products, as evidenced by staining with wheat germ agglutinin (WGA; Fig 3A) that binds to GlcNAc residues [31]. S. aureus MN8, a bacterial strain that has been shown to secrete GlcNAc-derived polymers[8, 32], was used in this assay as a positive control (Fig 3A).
Since UDP-GlcNAc is a common precursor for the synthesis of bacterial cell walls and some polyGlcNAc exopolysaccharides, such as PNAG (PIA) [9], we hypothesized that polyGlcNAc-containing exopolysaccharides could mediate enterococcal semisolid surface penetration. To test this, we examined the surface of enterococcal colonies grown on MOLP by immunofluorescence using the monoclonal antibody (mAb) F598, which specifically recognizes β-1,6-linked polyGlcNAc polymers [8, 11]. We detected polyGlcNAc-containing polymers on both the WT enterococcal cell surface and the positive control S. aureus (Fig 3A). The staining specificity was confirmed with an S. aureus strain (Δica) unable to produce the polyGlcNAc polymer, PNAG (PIA) [32] (S3A Fig). Further validating these results, the presence of these polymers was decreased and severely mislocalized in both WT E. faecalis and the positive control S. aureus MN8 upon treatment with Dispersin B (DspB; Fig 3A), an enzyme that specifically cleaves the β-1, 6 linkage of glucosamine and depolymerizes PNAG (PIA) [33]. DspB treatment did not affect the binding of WGA to any WT strain (Fig 3A), suggesting that this lectin may react with additional cellular components different from β-1,6-linked GlcNAc polymers. Indeed, WGA has been shown to detect not only GlcNAc residues, but also β-1,4-linked GlcNAc oligomers [31] such as the exposed-GlcNAc residues of the peptidoglycan layer of Gram (+) bacteria [34]. This observation was further confirmed by the remaining positive WGA signal found in the PNAG (PIA)-deficient S. aureus mutant Δica (S3A Fig). In contrast to WT E. faecalis, polyGlcNAc-containing polymers were not detected in glnA::TnM or rpiA::TnM mutants stained with mAb F598 (Fig 3B). Strikingly, metabolic complementation using GlcNAc-supplemented media restored polyGlcNAc-derived polysaccharide production (Fig 3B) and semisolid surface penetration (Fig 2H) in these two mutant strains.
We performed colony immunoblot assays to define the localization of polyGlcNAc-containing polysaccharides. Non-lysed E. faecalis cells from colonies grown on MOLP were transferred onto nitrocellulose membranes and incubated with mAb F598 to detect polymer production. Consistent with our microscopy results, we only observed a strong signal in the WT strain but not in glnA::TnM and rpiA::TnM mutants. Complementation by either addition of exogenous GlcNAc to the media (S3B and S3C Fig) or with plasmids expressing either RpiA or GlnA correspondingly (Fig 3C) restored extracellular polyGlcNAc-derived polysaccharide production. Similarly, the control S. aureus MN8 also demonstrated positive detection in these analysis (Fig 3C; and S3C Fig).
To further characterize the exopolysaccharides produced by E. faecalis grown on semisolid surfaces, we used calcofluor white (CFW), a fluorescent dye known to bind surface fibrillar exopolysaccharides harboring either β-1,3 or β-1,4 linkages such as cellulose and chitin [35–37]. In contrast to the CFW positive binding observed with Candida albicans colonies, a strain known to synthesize chitin (a β-1,4-linked oligosaccharide) [38], E. faecalis colonies did not produce CFW-reactive exopolysaccharide under our conditions. Similarly, the negative control Escherichia coli DH5α [39], did not exhibit any fluorescence with CFW in the culture media (S3D Fig). Taken together, these data suggest that E. faecalis produces β-1,6-linked GlcNAc-containing polysaccharides that are extracellularly localized and necessary for agar penetration capacity.
In S. aureus, the synthesis of polyGlcNAc polymers, such as PNAG (PIA), depends on the expression of biosynthetic enzymes encoded by the icaADBC operon [40–42]. IcaA is a glycosyltransferase that uses UDP-GlcNAc as a substrate [40], and IcaB is responsible for the deacetylation of PNAG (PIA) [43]. Since in silico analyses revealed that E. faecalis does not have homologs of these genes, we used Nanostring technology [44, 45] to identify potential glycosyltransferase genes that could be involved in synthesizing polyGlcNAc-containing polysaccharides that mediate E. faecalis agar penetration. Transcript levels of genes encoding putative glycosyltransferases were determined in cell lysates from E. faecalis colonies undergoing agar penetration, normalizing gene expression to multiple independent housekeeping genes (S2 Table). Several genes demonstrated significant expression changes during the semisolid surface entering process (S4A Fig). We focused on EF2170 (epaX) because its transcript levels were markedly elevated in cells that entered the agar, compared with non-penetrating cells on the surface (S4A Fig). To determine the role of EpaX in enterococcal semisolid surface penetration, we tested a strain harboring mutations in the epaX gene (EF2170) [46]. This mutant had no apparent growth defects under our conditions (Fig 4A and S4B Fig), but demonstrated a substantial defect in agar penetration, which was corrected upon genetic complementation with a plasmid expressing EpaX (Fig 4A). Further confirming these results, deletion of epaX in a closely-related E. faecalis strain (MMH594) also resulted in attenuated semisolid surface invasion (S4C Fig). Similarly to glnA::TnM and rpiA::TnM strains, SEM analysis of external E. faecalis cells in six-day-old colonies revealed that ΔepaX showed a profound reduction in the extracellular matrix normally covering and connecting its WT counterpart (Fig 4B and 4C).
To further characterize the role of EpaX in the synthesis of polysaccharides under our penetration conditions, we used polyacrylamide gel electrophoresis with subsequent alcian blue and silver nitrate staining to analyze the polysaccharide content of WT and ΔepaX extracted from colonies grown on MOLP. Consistent with previous results [46], a band disappeared in the ΔepaX strain, which was restored upon genetic complementation (S4D Fig), suggesting drastic changes in the oligosaccharide composition between these strains. Indeed, further analysis using acid methanolysis combined with gas chromatography-mass spectrometry (GC-MS) determined that loss of EpaX severely altered the glycosyl composition of E. faecalis (Fig 4D). Specifically, ΔepaX stains demonstrated increased glucose content that was accompanied by a profound reduction in rhamnose, GalNAc and GlcNAc, compared with their parental strain (Fig 4D). These data suggested that EpaX might be required for the synthesis of GlcNAc-containing exopolymers that are necessary for optimal E. faecalis semisolid surface penetration. To test this idea, we performed immunofluorescence analyses using mAb F598. Notably, E. faecalis ΔepaX was not recognized by the antibody (Fig 4E and S4E Fig), but this defect could be corrected upon genetic complementation with a plasmid encoding EpaX (Fig 4E). The binding of WGA to E. faecalis was unaltered in the absence of EpaX (S4E Fig), suggesting that this putative glycosyl transferase is necessary to generate β-1,6-polyGlcNAc-containing polymers, but not to synthesize other GlcNAc-containing cellular components or polysaccharides detected by the lectin. Moreover, colony immunoblot assays further demonstrated that polyGlcNAc-containing exopolysaccharides were detected only in colonies from strains with a functional EpaX (Fig 4F). To define whether EpaX operates upstream or downstream of GlnA and RpiA, we tested if exogenous fructose, glucosamine or GlcNAc could rescue the defective invasive phenotype of ΔepaX, as previously observed in glnA::TnM and rpiA::TnM mutants (Fig 2H). The WT parental strain formed bigger colonies and penetrated more efficiently when fructose, glucosamine and GlcNAc were supplemented. However, none of these substrates rescued penetration in ΔepaX strains (Fig 4A and S4C Fig). Supplementation of exogenous GlcNAc also failed to restore polyGlcNAc-containing polymer production by the ΔepaX strain (Fig 4E). Strikingly, however, exogenous addition of exogenous purified PNAG (PIA) from S. aureus MN8 fully restored invasion by ΔepaX strains (Fig 4G). CFUs quantification of invading cells confirmed that S. aureus MN8-derived PNAG (PIA) rescued the attenuated invasive phenotype observed in ΔepaX (Fig 4H). No major bacterial growth defects were observed upon PNAG (PIA) addition to semisolid media (S4F and S4G Fig). Together, these results demonstrate that EpaX functions downstream of GlnA and RpiA to drive β-1,6-linked polyGlcNAc polymer-mediated surface penetration in E. faecalis.
E. faecalis has the potential to translocate from the gastrointestinal tract to the blood stream [47], most likely via a paracellular mechanism that allows the bacterium to move through epithelial cell monolayers [48]. We hypothesized that E. faecalis mutants defective in biosynthesis of polyGlcNAc-containing polymers and semisolid surface penetration would also be altered for translocation through intestinal epithelial barriers. To this end, we used a previously described two-chamber transcytosis system [49, 50] where translocation is evaluated by determining the bacterial number (as CFUs) capable of passing from the apical side through T84 human intestinal epithelial cell monolayers, to the basolateral side of the chamber (Fig 5A). We found that 8 hours post-infection, the integrity of inoculated and non-inoculated monolayers exhibited transepithelial resistance values similar (~8,900 Ω/cm2) to those obtained prior to bacterial inoculation (~8,300 Ω/cm2), thus indicating that the T84 cell monolayers remained mostly intact throughout the experiment. All strains evaluated reached approximately 108−109 CFUs/mL in the apical side of all the wells tested (Fig 5B, S5A and S5B Fig). Consistent with previous reports [49, 50], the negative control E. coli DH5α was not detected in the lower chamber of any of the inserts analyzed (Fig 5B, S5A and S5B Fig), but WT E. faecalis, in sharp contrast, demonstrated a remarkable capacity to translocate in this assay. When E. faecalis mutant strains were evaluated, reduced numbers of rpiA::TnM cells (~5x102 CFUs/mL) were detected in the basolateral section in comparison with its parental WT strain (~6x107 CFUs/mL; S5A Fig). Interestingly, the glnA::TnM mutant did not show a significant decrease in translocation, likely due to metabolic complementation by exogenous glutamine present in the translocation culture medium (S5A Fig). Indeed, removing glutamine from the system drastically attenuated the translocation capacity of glnA::TnM but not WT cells (~4x102 vs. ~1x108 CFUs/mL, respectively; S5B Fig). Similarly, while WT E. faecalis moved efficiently through epithelial cell monolayers, ΔepaX demonstrated a significant decrease in translocation (~2x109 and ~1x103 CFUs/mL for WT and mutant, respectively; Fig 5B). Of note, genetic complementation of rpiA::TnM, glnA::TnM or ΔepaX cells restored the ability of each corresponding mutant strain to translocate (Fig 5B, S5A and S5B Fig). As control in our assays, we used a ΔepaB deletion strain unable to produce the glycosyl transferase EpaB (Orfde4), a protein previously shown to be necessary for efficient E. faecalis translocation through human epithelial cell monolayers [6, 49]. Surprisingly, under our conditions, the translocation ability of the ΔepaB mutant was similar to its parental strain (S6A Fig). In addition, ΔepaB was capable of producing polyGlcNAc-containing polymers at similar levels as the WT strain (S6B and S6C Fig), and exhibited normal capacity to penetrate semisolid agar (S6D Fig). These data suggest that, under our experimental conditions, the synthesis of polyGlcNAc-containing polymers is sufficient to enable surface penetration by strains devoid of EpaB.
To further characterize E.
faecalis translocation, T84 human intestinal epithelial cell monolayers were infected with GFP-labeled E. faecalis parental and mutant strains. Immunofluorescence analyses were performed by reacting each sample with phalloidin, DAPI and mAb F598 to visualize the actin cytoskeleton, nuclei (and bacterial DNA) and polyGlcNAc-containing polymers, respectively. Laser scanning confocal microscopic analyses of stained samples were carried out to localize bacteria within the T84 monolayers. After two hours of infection, we observed that enterococci were frequently co-localized with actin-rich areas (Fig 5C and S5C and S5D Fig). Moreover, the orthogonal views showed that bacterial aggregates concentrated on the top of the monolayers where parental strains (WT) formed surface invaginations, in comparison with the smooth surface of intact T84 human intestinal epithelial monolayers (Fig 5C, S5C and S5D Fig), hence suggesting that WT strains alter the actin cytoskeleton during the translocation process. These surface perturbations were observed to a lesser extent in monolayers infected with either ΔepaX (Fig 5C), rpiA::TnM (S5C Fig) or glnA::TnM mutants (S5D Fig). Importantly, polyGlcNAc-containing polysaccharides were detected around WT cell aggregates, but not in any of the mutants tested (Fig 5D, S5C and S5D Fig). These polysaccharides were frequently found to cover (or be adjacent to) bacterial cells, and their presence was visualized even after 6 hours post-infection (Fig 5D). Interestingly, we only observed surface openings of the epithelial barrier upon incubation with WT strains, suggesting that host cell lysis is caused by E. faecalis during the translocation process (Fig 5C and 5D and S5C and S5D Fig).
Taken together, our data reveal that E. faecalis utilizes the RpiA-GlnA-EpaX axis to generate β-1,6-linked polyGlcNAc-containing exopolysaccharides necessary for optimal migration into semisolid surfaces and for efficient paracellular translocation through human epithelial cell monolayers.
In this study, we uncovered molecular pathways and metabolic mediators that endow E. faecalis with the capacity to move into semisolid surfaces and translocate through human epithelial cell barriers (proposed model; Fig 6).
Exopolysaccharides have been well characterized as prominent components of the extracellular matrices of surface-associated multicellular communities termed biofilms [51–53]. Our study reveals a new role for polyGlcNAc-containing extracellular polysaccharides as key mediators of E. faecalis migration traits. These exopolysaccharides may operate as a “glue” that holds cells together [52, 54, 55] while promoting the formation of matrix-encased multicellular aggregates during enterococcal migratory behavior. Indeed, it has been proposed that polyGlcNAc polymers facilitate intercellular adhesion by bridging electrostatic interactions between cells surfaces [56]. Alternatively, or in addition, polyGlcNAc-containing hydrated exopolysaccharides may help E. faecalis to spread in a manner similar to that found in Proteus mirabilis, which secretes polysaccharides that create a fluidic environment promoting movement on surfaces with low moisture [57]. Furthermore, in B. subtilis, a PNAG (PIA)-defective strain was shown to lose its hydrophobic or nonwetting surface characteristics [58], indicating that polyGlcNAc polymers provide a means to shape the external environment in a manner amenable to bacterial movement or penetration. Additional biophysical and chemical analyses are thus warranted to comprehensively understand how these glycopolymers promote surface penetration by E. faecalis.
Our study unearths new metabolic factors mediating enterococcal surface penetration. The first one is GlnA, which plays an essential function in the generation of glutamine [27] that is used as a constituent of proteins and a nitrogen donor for many biosynthetic reactions [59, 60]. The second factor is RpiA, which catalyzes a central enzymatic reaction in the pentose phosphate pathway that is a major route of intermediary carbohydrate metabolism. RpiA is also involved in the generation of lipopolysaccharide components in Gram-negative bacteria [28]. Our results indicate that the metabolic functions of GlnA and RpiA converge in the hexosamine biosynthetic pathway to generate the UDP-GlcNAc necessary to produce polyGlcNAc-containing polymers such as PNAG (PIA) [61]. Based on our genetic and metabolic supplementation experiments, we propose that the hexosamine biosynthetic pathway and the pentose phosphate pathway supply metabolic substrates that serve as precursors for synthesizing enterococcal polyGlcNAc-containing polymers (see proposed model, Fig 6). A link between the pentose phosphate pathway and polysaccharide synthesis was previously described in bacteria. Somerville and colleagues reported that the S. aureus transcriptional regulator RpiR, which is known to control rpiA expression, also acts as a sugar-responsive regulator that modulates polysaccharide synthesis in response to metabolite concentrations [62].
The biosynthesis of polysaccharides first occurs in the cytoplasm, and the repeating units are then assembled and exported to the surface. This process involves several key enzymes including glycosyltransferases that mobilize sugar units [63]. Our study emphasizes a major role for the putative glycosyltransferase EpaX in E. faecalis physiology, as this enzyme was pivotal for semisolid surface penetration and paracellular translocation. Supporting this concept, Rigottier-Gois and colleagues had demonstrated that EpaX is a major determinant of E. faecalis intestinal colonization in mice [46]. Of note, the penetration-defective phenotype of ΔepaX could not be complemented by exogenous GlcNAc, indicating that EpaX is required for synthesis of the polyGlcNAc structure needed for optimal surface migration. Our data also show that EpaX acts downstream of RpiA and GlnA, which explains why only the addition of exogenous PNAG (a polyGlcNAc polymer) could rescue the penetration-defective phenotype of ΔepaX strains. Consistent with the notion that GlcNAc-derived polysaccharides are necessary for semisolid surface penetration, we demonstrated that epaX mutants do not produce detectable amounts polyGlcNAc-containing exopolymers. Interestingly, bioinformatic analysis using the Protein Homology/AnalogY Recognition Engine (Phyre2) [64] indicated that EpaX has 100% similarity across its predicted secondary structure to glycosyltransferases such as N-acetylgalactosamyltransferases. Furthermore, analysis using the conserved domain architecture retrieval tool (CDART) revealed that EpaX also has similar domain architecture to YdaM, a putative glycosyltransferase shown to be required for exopolysaccharide synthesis in Bacillis subtilis [65]. However, the function of EpaX remains elusive, and its activity might have an epistatic relationship with other factors required for the production or cell surface display of polyGlcNAc-containing polymers. Recent studies proposed that epaX deletion alters the synthesis of the rhamnopolysaccharide EPA in E. faecalis by compromising the decoration of these polymers with galactose and/or GalNAc. Therefore, it was suggested that EpaX functions as a GalNAc transferase [46]. Our findings indicate that, under our conditions, the absence of EpaX in E. faecalis not only dramatically decreases the levels of GalNAc-, but also of rhamnose- and GlcNAc-containing oligosaccharides. Though rhamnose was not detected in the polysaccharides produced by ΔepaB using GC-MS analysis [6], we found that E. faecalis lacking EpaB was still able to penetrate and generate polyGlcNAc-containing polymers, suggesting that the presence of GlcNAc, but not rhamnose, in E. faecalis exopolymers is necessary for optimal penetration into semisolid surfaces. While the structure of E. faecalis EPA has not been elucidated, similar polysaccharides with branching structures composed by other oligosaccharides bound to GlcNAc or terminal β-linked GlcNAc side chains have been evidenced in other Gram-positive bacteria [66]. Indeed, our results using DspB demonstrated the presence of β-1,6 glycosidic bonds within the structure of E. faecalis polyGlcNAc-containing exopolysaccharides. However, the precise nature of the polymer involved in E. faecalis semisolid surface and epithelial barrier penetration has not yet been elucidated by purification and chemical analyses. Indeed, either EPA or another polysaccharide yet to be identified might mediate the penetration process. Future analyses to determine the structure of E. faecalis polyGlcNAc-containing exopolymers, and their link with EPA, will hence be of significant interest.
E. faecalis is a leading cause of nosocomial infections world-wide [67]. It has been shown that E. faecalis can translocate across mouse and rat intestinal tracts to reach other body sites [68, 69]. Most recently, Krueger et al. reported that after feeding mice with antibiotics, E. faecalis could be found in the liver, spleen, and mesenteric lymph nodes [70, 71]. PolyGlcNAc-like polysaccharides might mediate these processes by promoting enterococcal translocation across the intestinal epithelium. Interestingly, E. faecalis has been shown to form microcolonies surrounded by an extracellular matrix that not only covers the bacterial cells, but also extends into the intestinal space between cell clusters [72]. In line with our observations during semisolid surface invasion and epithelial barrier assays, Peng and collaborators described that E. faecalis formed cellular aggregates that localized with the actin cytoskeleton during the process of translocation [48]. Our findings therefore uncover that production of polyGlcNAc-containing exopolysaccharides is a mechanism that enables non-motile E. faecalis to penetrate semisolid surfaces and cross human intestinal epithelial cell monolayers.
S3 Table describes all strains and plasmids used in this study [6, 10, 26, 32, 46, 73–83] E. faecalis was cultured overnight at 37°C in Tryptic Soy Broth (TSB) with 0.25% Glucose (Becton Dickinson) under shaking conditions, unless indicated otherwise. E. coli strains were cultured in Lysogeny Broth (LB). Antibiotics were added to the medium when appropriate as follows: Chloramphenicol 10 μg/mL, spectinomycin 150 μg/mL or ampicillin 100 μg/mL for E. coli. Either tetracycline 15 μg/ml, chloramphenicol 10–15 μg/mL or spectinomycin 750 μg/mL for E. faecalis strains when specified. All chemicals were purchased from Sigma-Aldrich unless stated otherwise.
2 μL of TSB-grown E. faecalis overnight cultures were inoculated onto modified medium optimal for lipopeptide production (MOLP) [24], containing 30 g/L peptone, 7 g/L yeast extract, 1.0 mM MgSO4, 25 μM MnSO4, 25 μM FeCl2, 0.001 mg/L CuSO4, 0.004 mg/L Na2MoO4, 0.002 mg/L KI, 5 μM ZnSO4.7H2O, 0.001 mg/L H3BO3, 25 mM potassium phosphate buffer (pH 7), 125 mM MOPS (morpholinepropanesulfonic acid; pH 7) and 10 g/L agar (Becton Dickinson). Saline solutions were filter-sterilized independently before mixing the MOLP components. Semisolid MOLP agar was prepared the day before and air-dried (opened plates inside the biological hood) for at least 30 minutes prior to bacterial inoculation. E. faecalis MOLP-inoculated plates were incubated upside down in a highly humid environment to avoid dryness at 37°C for 6 days, unless indicated. After this period, semisolid media penetration was determined by removing all cells above the agar with 3 to 4 washes with ~10 mL distilled water and then observing bacterial growth within the agar.
When stated, MOLP media was solidified with poloxamer-407 (Sigma, Aldrich; MOLP-407), a fully autoclavable copolymer based on polyoxyethylene and polypropylene previously used for bacterial media growth development [84]. At low temperature, a poloxamer-407 solution is liquid, but becomes solid upon reaching room temperature (RT). MOLP-407 was prepared by the addition of 10 g of the polymer powder each day into 50 mL distilled water held at 4°C until a concentration of 60% [w/v] was achieved. This solution was then kept at 4°C for an extra 24 hours to ensure complete dissolution, prior to autoclaving. Next, it was cooled to RT and chilled to 4°C to liquefy. Once at low temperature, the poloxamer-407 solution was mixed 1:1 with a cold solution of 2X MOLP to a final volume of 100 mL. Subsequently, 1.5 mL of this chilled media were added into each well of 24-well plates (Falcon, Corning) and allowed to solidify at RT prior to inoculation with 1 μL of each bacterial strain grown overnight in TSB.
Colony forming units (CFU) of penetrating and non-penetrating E. faecalis cells grown on MOLP (for 6 days) solidified with either 1% (w/v) agar or 30% [w/v] poloxamer-407 were measured using two distinct strategies: To determine the number of external/internal cells grown on MOLP-407, the external cells from the colonies were collected and suspended in 500 μL of saline solution (0.89% NaCl). Subsequently, the surface of the plates was washed 3–4 times with 10 mL of sterile distilled water at RT, and invading bacteria were recovered by transferring the growth from each well into media previously chilled at 4°C to sterile Eppendorf tubes. All bacterial suspensions (internal and external cells) were centrifuged at 4°C and washed 3 times with ice-cold saline solution prior to making serial dilutions and plating on TSB agar plates. After 24 hours of incubation at 37°C, the final CFU number was calculated.
To quantify invasion of E. faecalis colonies grown on MOLP with 1% agar, the colonies were grown on top of 3.0 μm filters (Whatman) to separate the external from internal cells. The first ones (external) were collected by suspending each filter and suspended them in 500 μL of 1X Dulbecco’s Phosphate Buffered Saline solution (DPBS; Corning-Cellgro) and the remaining non-penetrating bacteria was removed by three washes with 10 mL of sterile distilled water and two washes with 70% ethanol (10 mL). The internal cells were recovered by removing an area of ~1 cm2 from the top layer, that was then suspended in 500 μL of DPBS as previously described above. Penetrating and non-penetrating bacterial suspensions were homogenized with a mortar and pestle followed a passage through a needle (27-gauge) syringe and filtered with a 40 μm nylon filter (BD Falcon). Final saline suspensions (1 mL) were sonicated for 2 minutes (30 seconds ON and OFF cycles) at 30% amplitude (Sonics Vibra Cell) to separate cellular clumps and then they were serially diluted and CFUs were determined as described above. Only when mutants exhibited growth differences to their parental strain, the final CFUs/mL was normalized to the absorbance (OD600) of each saline suspension from which serial dilutions were performed (Normalized CFUs/mL).
Internal and external cells of MOLP-grown colonies were recovered and processed as described above (see agar-penetration quantification section). Each penetrating and non-penetrating population was then diluted down to 0.5 OD600 prior to be stained with Brilliant Violet-570 (BV-570; LIVE/DEAD staining kit—Life technologies) for 30 min at RT in the dark, following the manufacturer’s instructions. Samples were washed twice with 1 mL DPBS and subsequently fixed with 4% paraformaldehyde (BioWorld) overnight at 4°C. Heat killed (100°C for 24 hours) TSB-grown E. faecalis was used as dead control. Live and dead bacteria were analyzed using a BD LSRII Flow cytometer.
E. faecalis colonies were grown on MOLP as described above. SEM samples were prepared as previously described [85], with some minor modifications: External cells grown on MOLP were carefully transferred to ∼10 mm diameter pieces of 0.1% poly-L-lysine (Sigma, Aldrich) pre-treated Silicon wafers (Ted Pella). Samples were then fixed in a solution with 2.5% glutaraldehyde, 0.1% DMSO (dimethyl sulfoxide), 0.15% alcian blue and 0.15% safranin O [86], at RT for 18 hours. When stated, a 90 min post-fixation step with 1% Osmium tetroxide was performed. After three 15 min washes with distilled water and dehydration through a graded series of ethanol, the samples, unless specified were infiltrated with hexamethyldisilazane (HMDS; Sigma, Aldrich), through one incubation in 50% HMDS (in 100% ethanol) at RT for 1 hour and then two in 100% HMDS for 30 min. At this point the PDMS-bound samples were mounted on pins, dried under vacuum overnight, sputter-coated with gold–palladium alloy, and examined by SEM. For analyzing invading bacteria, small agar sections were placed on the silicon chips after removing the external cells with water; and treated as described above.
In order to objectively quantify the fraction of cells covered in matrix, the SEM images were analyzed using automatic image analysis software, Ilastik 1.3.0 [87]. The software was first trained to recognize different image structures, including background, matrix and cells, based on a single SEM image only. In this training stage, we manually identified regions in the image corresponding to background, matrix and cells, which the software uses to update a machine learning algorithm. After training, the algorithm was used to automatically analyze all remaining SEM images. In those images, pixels corresponding to either matrix or cells are automatically detected, thereby providing an estimate for fraction of cell surface that is covered in matrix. In all cases, we analyzed the SEM images at the same 20,000 X magnification. This magnification was chosen such that we could examine as much surface as possible, without compromising on the resolution needed for automatic image analysis.
A Mariner transposon insertion library in the multidrug resistant clinical isolate MMH594 [88] was constructed. E. faecalis was transformed with mariner delivery system pLB02 (kind gift of Dr. Lynn E. Hancock), which is identical to progenitor pCAM45 [89], except that erythromycin and kanamycin resistance markers were swapped for tetracycline and chloramphenicol resistance, respectively. Essentially as described in previous studies [89],, transformants were initially selected at 30°C and the cure of the delivery vector was done at elevated temperature with selection for only the chloramphenicol resistance harbored by the transposon. A total of >300,000 MMH594 colonies possessing mariner insertions were collected as a pool.
To find targets necessary for enterococcal semisolid surface penetration, a replica-plating method was used. Approximately 6,000 mariner transposon mutants grown on TSB in 96-well plates for 24 hours were replica plated onto MOLP plates to screen for agar penetration capacity. A non-penetrating phenotype was designated as the ability to form WT-like colonies without growth inside the semisolid surface. To determine the genome site insertion of the mariner transposon, we used a previously described modified arbitrary PCR method with few modifications [90]. Amplification of short DNA-fragments was performed by using Platinum PCR High Fidelity SuperMix (Thermo Fisher Scientific) as described by the manufacturer. External and internal oligonucleotides specific for the Tn-Mariner (TnMextF1,and TnMxtF2; S4 Table) and the arbitrary primers (STAPHarb1, STAPHarb2, and STAPHarb3)[90] were utilized for the PCR reactions. The first round was performed using the arbitrary primers STAPHarb1 and SATPHarb2 (0.6 μM) paired with TnMexF1 (0.3 μM). 5.0 μL of a lysate of each mutant colonies obtained as previously described [91] was used for the PCR reaction: 95°C for 3 minutes; five cycles of 94°C for 30 seconds, 30°C for 30 seconds and 72°C for 1 min; then 25 cycles of 94°C for 30 seconds, 52°C for 30 seconds and 72°C for 1 minute; and finally 72°C for 5 minutes. Samples were kept at 4°C. The second PCR round was performed with primers TnMextF2 (0.3 μM) and STAPHarb3 (0.6 μM) as follow: 3 minutes at 94°C, 30 cycles of 94°C for 30 seconds, 55°C for 30 seconds and 72°C for 1 minute, followed by 72°C for 5 minutes. The samples were then kept at 4°C. The PCR products were visualized by agarose gel electrophoresis, and the second round PCRs containing at least one distinct visible fragment were used for further characterization. Nucleotide sequence analysis was performed with TnMextF2 primer. To identify of the Tn-Mariner insertion sites, a basic local alignment search tool (BLAST) was used.
The mariner transposon mutants, glnA::TnM and rpiA::TnM, were complemented in-trans by inserting the corresponding WT gene in the pAT28 plasmid [82]. To this end, PCR-amplified gene products with their corresponding promoters were generated for rpiA and glnRA from purified MMH594 genomic DNA. For glnRA, we used primers JD15 and JD30 (for sequences see S4 Table) to amplify a fragment of 2248 bp, which included a region 388 bp upstream of glnR, as well as glnR and glnA open reading frames (ORF). The PCR product was digested with EcoRI and BamHI (NEB) and ligated into pAT28 to generate the complementation vector pJR01. Likewise, rpiA amplification was done using primers HV172 and HV173 (S4 Table). The amplified product (1291 bp) was digested with BamHI and XbaI and ligated into pAT28 to generate the complementation vector pAH01. Plasmids, pJR01 and pAH01, were electroporated in E. coli and after sequencing several colonies; one was selected for complementation of each transposon mutant. The complementation vectors were transformed by electroporation into the corresponding E. faecalis strains and recovered on TSB plates (750 μg/mL spectinomycin) as previously described [78].
The fluorescence reporter strains were constructed by conjugation of the vector pV158-GFP between the donor E. faecalis OG1RF and the recipient MMH594 rpiA::TnM and glnA::TnM strains, as previously reported [92]. Briefly, TSB-grown overnight cultures of donor (15 μg/mL tetracycline) and recipients (10 μg/mL chloramphenicol) were diluted down to 0.05 OD600 and allowed to reach and absorbance of O.5 OD600. Then, the donor and recipient were mixed 1:10, 10:1 and 1:1 prior to concentrating these solutions to a final volume of 50 μL. Each one was finally placed onto a 0.2 μm-pore-size polycarbonate membrane (13 mm; Nucleopore) previously placed on TSB agar plates. After 24 hours at 37°C, filters were removed from the plates and placed in 1 ml 1X PBS (Dulbecco’s phosphate buffer saline; Sigma, Aldrich). Cells suspensions were then diluted and 10−3, 10−7 and 10−9 dilutions were plated on TSB agar with 15 μg/mL tetracycline, 10 μg/mL chloramphenicol and 250 μg/mL gentamycin. After 24 hours of incubation at 37°C, GFP fluorescent colonies were selected by microscopic analysis. The vector pV158-GFP was electroporated into electrocompetent cells of E. faecalis WT and ΔepaX strains prepared as previously described [93]. Cells were allowed to recover for 2 hours in 1 mL of SGM17MC recovery medium [93] before being plated and selected on TSB agar as described above.
E. faecalis MMH594 was used for the generation of the EF2170 (V583 epaX homolog) deletion mutant by allelic exchange (ef2170::spcR) using the pMINIMAD thermosensitive plasmid [81]. Briefly, fragments upstream and downstream of the EF2170 gene were PCR amplified with Phusion polymerase (NEB) using JD1, JD3; and JD6, JD8 primers, respectively. The spectinomycin resistance gene spcR was amplified from vector pIC333 [79] with primers JD4 and JD5. These three purified PCR fragments were assembled with Gibson Assembly Mix 2X (NEB) following the protocol suggested by the manufacturer. The final reaction was then used as template for amplifying a ~3 kb product (using primers JD2 and JD7) that was inserted into the BamHI site of pMINIMAD generating the vector pJR02. This plasmid was then modified by inserting in the SalI site, the chloramphenicol resistance gene cat, amplified from pLT06 [10] with primers JD44 and JD45 (S4 Table) generating pJR03. This last vector was then transformed into E. coli Top10 and transformants were selected after growth overnight in LB broth with 10 μg/mL chloramphenicol at 37°C. The plasmid pJR03 was purified and electroporated into WT MMH594 as previously described [78]. Transformed bacteria were grown on Todd-Hewitt agar plates with 15 μg/mL chloramphenicol at RT. Next, positive colonies were grown overnight on TSB with chloramphenicol at RT. Cells were centrifuged, suspended in 200 μl of fresh media, and plated in TSB agar with 400 μg/mL of spectinomycin. After 48 hours at 42°C, candidate colonies were grown on TSB agar with either 400 μg/mL of spectinomycin or 15 μg/mL chloramphenicol at RT. Allelic exchange was confirmed by PCR for the spectinomycin resistant and chloramphenicol sensitive colonies.
Bacterial strains were cultured in 500 mL of either MOLP broth under static conditions to an OD600 of 0.6 or MOLP agar for 6 days at 37°C. Polysaccharides were extracted as previously described [46, 94] with minor modifications. Briefly, cells from either liquid cultures (for glycosyl composition analysis) or colonies were centrifuged 20 minutes at 4000 rpm and washed with 10 mL of sucrose-buffer (25% sucrose, 10 mM Tris-HCl; pH 8). Pellets were then suspended in 15 mL sucrose-buffer supplemented with 1 mg/mL lysozyme (Thermo Scientific) and 10 U/mL mutanolysin and incubated at 37°C overnight with gentle agitation. Following this incubation, the cellular fraction was removed by centrifugation for 20 minutes at 4500 rpm. The supernatants were treated with 200 μg/mL RNase A, 200 μg/mL DNase, 5 mM MgCl2, and 1 mM CaCl2 at 37°C for 8h to remove nucleic acids. Protein impurities were removed by adding proteinase K (50 μg/mL) to each supernatant and incubating them at 42°C overnight. Remaining contaminants were extracted using 1 mL of chloroform. The aqueous phase was transferred to a new tube following centrifugation (4500 rpm) for 15 minutes. Polysaccharides were precipitated by adding ethanol to a final concentration of 75% and incubation at − 80°C for 30 minutes, followed by a centrifugation (4500 rpm for 1 hour) at 4°C. Precipitated pellets were washed using 75% ethanol and allowed to air dry. Cell wall polysaccharide samples were submitted for glycosyl composition analysis to the Complex Carbohydrate Research Center (University of Georgia). Glycosyl composition analyses were performed using GC-MS of the per-O-trimethylsilyl (TMS) derivatives of the monosaccharide methyl glycosides. The TMS derivatives were produced from the sample by acidic methanolysis [95]. GC-MS analysis of TMS methyl glycosides was done on an Agilent 7890A GC interfaced to a 5975C MSD, using an Supelco Equity-1 fused silica capillary column (30 m x 0.25 mm ID). A total of 3 independent biological samples per strain were analyzed.
To visually analyze the composition of polysaccharides extracted from non-penetrating cells of E. faecalis VE14089 WT, ΔepaX and ΔepaX p-epaX, dry samples were suspended in 100 mL of Tris–NaCl (50 mM Tris-HCl, 150 mM NaCl; pH 8.0) and mixed with 16% glycerol prior to be run (25 μL) on a 10% polyacrylamide gel (acrylamide to bisacrylamide, 29:1; Fisher Scientific) in Tris-borate-EDTA buffer (89 mM Tris base, 89 mM boric acid, 2 mM EDTA; pH 8.0) for 90 minutes at 130 volts. Detection of polysaccharides was made with silver staining as previously described [46] with minor modifications. Briefly, the polyacrylamide gel was washed once with distilled water and incubated 45 minutes with 1 mg/mL of alcian blue in 3% acetic acid. Later, after three washes with distilled water, the gel was incubated in a solution with 3.4 mM K2Cr2O7 and 3.2 mM HNO3 for 7 minutes and then washed with water as described above. Following these steps, the gel was then treated with 12mM AgNO3 and exposed to intense light for 30 min. It was later washed with water, soaked in 50 mL of 0.28 M Na2CO3 and 6mM formaldehyde until signal was visually detected and transferred to a solution of 0.1 M acetic acid for storage.
Bacterial samples from either external or penetrating E. faecalis grown on MOLP for 6 days were collected and processed as previously described [8], with the following modification: Cells were suspended in 1000 μL of DPBS and then spotted onto microscope slides. After samples air-dried, they were fixed by methanol: acetone 1:1 for 10 min at -20°C. Then, they were treated with 100% ice-cold methanol for ~1 min, followed by 3 washes with PBS-NaCl buffer (20 mM PBS and 150 mM NaCl). Samples were blocked with 2.5% Normal Horse serum (Vectors Lab) for 45 min at RT. PBS-1% BSA (bovine serum albumin; Sigma, Aldrich) was then added after removing the blocking serum. After 1 min incubation at RT, slides were reacted overnight at 4°C with 20 μg/mL of human IgG mAb F598, which specifically binds to β-1,6-linked GlcNAc polysaccharides [8, 52]. After 3 washes with PBS-NaCl, samples were reacted with the secondary antibody, anti-human IgG labeled with Alexa Fluor-488 (15 μg/mL; Invitrogen) and DAPI (2.0 μg/mL), for 2 hours at RT. To visualize GlcNAc residues, samples were incubated with 5 μg/mL of the lectin WGA (wheat germ agglutinin) directly conjugated to Texas Red (Thermo Fisher Scientific) for 30 min at RT. Slides were then washed and cover-slipped using Fluoromount-G media (SouthernBiotech). Images were captured at 63x magnification at 1000 ms for DAPI, and 600–1500 ms for FITC and Rhodamine. Imaging was performed with a Zeiss AxioObserver inverted wide field/fluorescence microscope and processed using MetaMorph software. All images were adjusted to reduce background fluorescence.
For enzymatic treatments, cell samples obtained as described above were diluted 1:10 in PBS. 10 μL of these cell solutions were suspended in 90 μL of Tris-buffered saline (TBS; pH 7.4) containing 300 μg/mL DspB [96]. Samples were incubated for 24 hours at 37°C with constant shaking, and then centrifuged to suspend the pelleted cells into 50 μL of fresh TBS. Each suspension was subsequently placed onto glass slides to then be treated as described above.
E. faecalis colonies were analyzed for extracellular production of polyGlcNAc-polymers using a protocol previously described [97] with some modifications: Briefly, 0.45 μm nitrocellulose membranes (Bio-Rad) were placed on 1-day-old colonies grown on MOLP until they became completely wet. The plates/membranes were incubated at 37°C for 10 minutes prior to be carefully removed and transfer colony side up to a glass petri dish for air-drying (10 minutes at 37°C). Following this step, the air-dried membranes were immersed in chloroform at RT for ~15 minutes or until the chloroform completely evaporated. Each nitrocellulose membrane was incubated colony side down for 60 minutes in the blocking buffer (25mM Tris-base, 0.15M NaCl, 0.1% Tween-20, 5% non-fat milk). After 3 washes, 5 minutes each with TBS-T (25mM Tris-base, 0.15M NaCl, 0.1% Tween-20), the membranes were incubated overnight at 4°C with gentle agitation in TBS-T with 5% bovine serum albumin (BSA) and 200 μg/mL of the primary antibody mAb F598. Membranes were washed with TBS-T as described above and then incubated for 60 minutes in TBS-T containing a 1/10000 dilution of peroxidase-conjugated goat anti-human IgG polyclonal antiserum (Thermo Fisher Scientific). Membranes were then washed 3 times for 5 min each with TSB-T and were developed using the SuperSignal West Pico Chemioluminescent Substrate Kit as directed by manufacturer (Thermo Fisher Scientific).
External and penetrating cells, from 2-day-old colonies grown on MOLP were collected and suspended in 2 mL of cold RNAlater. Samples were pelleted and supernatant was discharged prior to storage at -80°C. For cell lysis, these pellets were suspended in RLT buffer (500 μL; Qiagen) and completely disrupted with a beat beater in one volume 0.5 mm zilica/sirconia beads for ~4 minutes (4 times × 60 seconds). Cellular debris were removed and supernatants were then subjected to probe hybridization and processing with the Nanostring nCounter Prep Station and Digital Analyzer according to the manufacturer’s instructions. Raw code counts were analyzed according to manufacturer’s guidelines; briefly, total transcript counts were normalized using internal controls with background subtraction. Transcript counts for 5 genes (gyrB, def, sigA, aqpZ and folB) were used for geometric mean normalization to correct for differences in total mRNA concentration. All data were collected from 2 biological replicates and gene expression was considered significantly altered if the transcript number changed >2-fold. Total counts were expressed as log2-change relative to the counts of non-penetrating cells at day 1, a time point where invasion was not evidenced in MOLP.
T84 human intestinal epithelial cells (Sigma, Aldrich) were grown and maintained as previously described [49] with some modifications. Briefly, cell monolayers were grown on plastic in a 1:1 Dubelcco’s Modified Eagle’s medium and nutrient mixture F-12 (DMEM/F12; Corning Inc.) supplemented with 10% heat inactivated fetal bovine serum (FBS; Atlanta biologicals), 2 mM glutamine, 1 mM sodium piruvate, 10 mM HEPES buffer (pH 7), 1X non-essential amino acids, 50 units/mL penicillin and 50 μg/mL streptomycin (Corning Inc.), 5 μg/mL prophylactic plasmocin (InvivoGen), and 0.007% β-mecaptoethanol (Sigma, Aldrich). When monolayers reached confluence or near-confluence, cells were detached and split as previously described [98]. Translocation was performed by seeding 105 T84 human epithelial cells from previous passages into a 24-well Transwell system with 3.0-μm-pore-size polycarbonate membranes (Corning Costar Corp). This pore size allows bacteria, but not T84 cells, to penetrate the membrane. A volume of 300 and 1000 μL of the tissue culture medium described above was added to the apical and basolateral chambers, respectively; and this medium was changed every 2–3 days. The developing progress of T84 tight junctions was monitored by Millicell-ERS-2 measurement (Millipore). Translocation experiments were performed after 8 days of culture, when the trans-epithelial electrical resistance (TER) of T84 monolayers reached ~8000 Ω/cm2 or higher. To prepare bacteria for translocation, 12-hours-bacterial cultures (with appropriate media and antibiotics) were diluted down in HBSS (Hanks balanced salt solution without Ca2+ and Mg2+; Corning Inc.) to an absorbance of 0.25 OD600 (~108 CFU/mL). Bacterial solutions were then washed twice with 1 mL of HBSS and finally suspended in Translocation Media (TM; Gibco): Advanced DMEM/F-12 mixture supplemented with 5% FBS, 10 mM HEPES buffer (pH 7), 0.007% β-mecaptoethanol and when specified, 2 mM GluN. Prior to bacterial inoculation, the filters were washed twice with TM. After this step, 1000 μL of fresh medium were added to the basolateral chamber, and 300 μL of each TM-suspended bacterial culture prepared as described above, were inoculated to the apical side of the chamber; this inoculum is consistent with that used by others [49] and with the density of intestinal enterococci in some settings. TER was monitored at the beginning and after 8 hours post-infection. The TER values remained similar to those obtained for the pre-infected monolayers, indicating that the integrity of cell barriers was conserved throughout the experiments. CFUs of viable bacteria in both chambers were counted at 0, and 8 hours by removing 20 μl aliquots, serially diluting and plating on TSB agar plates. For each strain, 8–9 independent transwells were used and the experiments were repeated at least three times.
To visualize translocating bacteria, filters seeded with polarized human enterocyte-like T84 cells as described above were infected for 2-hours with E. faecalis constitutively expressing GFP, and then samples (infected and uninfected) were stained and observed by laser scanning confocal microscopy. Bacteria and epithelial cell translocation assays were done in TM supplemented with 15 μg/mL tetracycline. For immunofluorescence staining, medium on each transwell was removed and filters were washed two times with pre-warmed (37°C) PBS. Cells were then fixed by 4% paraformaldehyde for 40 min. Following fixation, cells were washed with DPBS for 30 seconds, and permeabilized by incubating them with PBT (PBS and 0.5% TritonX-100) solution for 15 minutes. The solution was removed and the cells (transwells) were washed twice with DPBS for 30 seconds. After this, samples were blocked with PBS-1% BSA for 1 hour at RT, and then washed once with PBS as previously described. Cells were reacted overnight at 4°C with 20 μg/mL of MAb F598 [8, 52]. After three washes with PBS (5 minutes) samples were incubated with the secondary antibody, goat anti-human IgG labeled with Alexa Fluor-647 (15 μg/mL; Invitrogen) for 2 hours at RT. Thereafter, cells were washed three times with PBS (5 minutes), followed by incubation with cellular dyes (200nM of both Alexa Fluor 594-coupled phalloidin and DAPI; Invitrogen) in PBS containing for 30 min in the dark at RT. The solution was removed and samples were washed three times with PBS for 30 seconds. Filters were cut and transferred into ~10 μL of ProLong Diamond Antifade Mountant (Thermo Fisher), followed by sealing on glass slides and storing in the dark at 4°C until microscopy. Imaging was performed with a LSM880 confocal microscope and processed using Image J software. All images were adjusted to reduce background fluorescence.
Unless noted otherwise, all experiments were repeated at least three times and results were similar between repeats. All statistical analyses were determined using GraphPad Prism 7.0. Differences between the means of experimental groups were calculated using either a two-tailed unpaired Student’s t-test or one-way analysis of variance (ANOVA). Error bars represent SEM from independent samples assayed within the represented experiments. P<0.05 was considered to be statistically significant.
|
10.1371/journal.pbio.0050275 | A Novel Snf2 Protein Maintains trans-Generational Regulatory States Established by Paramutation in Maize | Paramutations represent heritable epigenetic alterations that cause departures from Mendelian inheritance. While the mechanism responsible is largely unknown, recent results in both mouse and maize suggest paramutations are correlated with RNA molecules capable of affecting changes in gene expression patterns. In maize, multiple required to maintain repression (rmr) loci stabilize these paramutant states. Here we show rmr1 encodes a novel Snf2 protein that affects both small RNA accumulation and cytosine methylation of a proximal transposon fragment at the Pl1-Rhoades allele. However, these cytosine methylation differences do not define the various epigenetic states associated with paramutations. Pedigree analyses also show RMR1 does not mediate the allelic interactions that typically establish paramutations. Strikingly, our mutant analyses show that Pl1-Rhoades RNA transcript levels are altered independently of transcription rates, implicating a post-transcriptional level of RMR1 action. These results suggest the RNA component of maize paramutation maintains small heterochromatic-like domains that can affect, via the activity of a Snf2 protein, the stability of nascent transcripts from adjacent genes by way of a cotranscriptional repression process. These findings highlight a mechanism by which alleles of endogenous loci can acquire novel expression patterns that are meiotically transmissible.
| Genetics is founded on the principle that heritable changes in genes are caused by mutations and that the regulatory state of gene pairs (alleles) is passed on to progeny unchanged. An exception to this rule, paramutations—which reflect the outcome of interactions between alleles—produce changes in gene control that are stably inherited without altering the DNA sequence. It is currently thought that these allelic interactions cause structural alterations to the chromatin surrounding the gene. Recent work in both maize and mice suggests that RNA molecules may be responsible for paramutations. Several genes are required to maintain the repressed paramutant state of a maize purple plant1 (pl1) allele, and here we report that one of these genes encodes a protein (RMR1) with similarity to a protein previously implicated in facilitating genomic DNA modifications via small RNA molecules. Genetic and molecular experiments support a similar role for RMR1 acting at a repeated sequence found adjacent to this pl1 gene. Although loss of these DNA modifications leads to heritable changes in gene regulation, the data indicate these changes do not represent the heritable feature responsible for paramutation. These findings highlight an unusual but dynamic role for repeated genomic features and small RNA molecules in affecting heritable genetic changes independent of the DNA template.
| The term “paramutation” describes a genetic behavior in which the regulatory state of specific alleles is heritably altered through interactions with their homologous partners in trans [1,2]. This behavior presents an exception to the Mendelian principle that alleles segregate from a heterozygous state unchanged [3]. Paramutations have been best characterized at loci encoding transcriptional regulators of pigment biosynthesis in maize, but similar behaviors have been described in other plant and animal systems, most recently in mice [4,5]. While the broader roles of paramutation in genome-wide regulation and evolution remain to be seen, the Pl1-Rhoades allele of the maize purple plant1 (pl1) locus presents a tractable system to study the paramutation process.
The pl1 locus encodes a Myb-like protein that acts as a transcriptional activator of genes required for anthocyanin pigment production [6]. Inheritance patterns illustrate that the Pl1-Rhoades allele can exist in quantitatively distinct regulatory states, reflected by differences in plant color. When individuals with a highly expressed reference state of Pl1-Rhoades, termed Pl-Rh, are crossed with plants having a repressed state, referred to as Pl′, only progeny with weak pigmentation are produced [7,8]. Pl-Rh states invariably change to Pl′ in Pl-Rh/Pl′ heterozygotes [7]; this is a typical hallmark of paramutation. Relative to Pl-Rh, the Pl′ state displays reductions in both Pl1-Rhoades RNA levels (∼10-fold) and transcription rate (∼3-fold) that are associated with a reduction in plant pigment [8]. This repressed Pl′ state is meiotically stable when maintained in a Pl1-Rhoades homozygote, with no reversion to Pl-Rh seen to date. Pl′ can, however, revert to Pl-Rh when heterozygous with some pl1 alleles other than Pl1-Rhoades, when maintained in a hemizygous condition, or in the presence of specific recessive mutations [9–12].
Genetic screens for ethyl methanesulfonate (EMS)–induced recessive mutations identify at least ten loci, including required to maintain repression1 (rmr1), rmr2, rmr6, and mediator of paramutation1 (mop1), whose normal functions maintain the repressed Pl′ state ([10,11,13]; J. B. H., unpublished data). These rmr mutations specifically affect the expression of Pl1-Rhoades and not other pl1 alleles [10,11], indicating that the Pl1-Rhoades allele is a direct and specific target of paramutation-based epigenetic changes. mop1 was recently identified [14,15] as encoding the putative ortholog of the Arabidopsis protein RDR2, a presumed RNA-dependent RNA polymerase involved in siRNA-based maintenance of de novo cytosine methylation [16]. Recessive mutations defining rmr1, rmr2, and rmr6 destabilize the repressed Pl′ state, resulting in darkly pigmented plant tissues, an increase in pl1 RNA levels, and meiotic transmission of Pl-Rh revertant states [10,11]. To date, the molecular identity of these rmr factors remains unknown.
In this report we identify rmr1 as encoding a novel Snf2 protein that represents a founding member of a subgroup of factors similar to proteins involved in plant small RNA metabolism. Our analyses show that RMR1 affects both pl1 RNA transcript stability as well as small interfering RNA (siRNA) accumulation and DNA methylation patterns at Pl1-Rhoades. These results support a model in which maintenance of paramutant states is dependent on a repression mechanism similar to the recently proposed cotranscriptional gene silencing mechanism in fission yeast [17,18]. To our knowledge, RMR1 is the first protein identified that maintains trans-generationally repressed states established by paramutation.
The rmr1 locus is defined by four recessive mutations (Protocol S1) characterized by a darkly pigmented plant phenotype that results from loss of Pl′ repression. Previous RNase protection experiments showed a 26-fold increase in pl1 RNA in floret tissue between rmr1–1 mutant plants and heterozygous siblings [10]. However, these experiments did not address if changes in pl1 transcript abundance correlated with changes in actual transcription at the pl1 locus.
In vitro transcription assays using nuclei isolated from husk leaf tissue revealed there was no statistically significant change in relative transcription rates of the Pl1-Rhoades allele between rmr1–1 mutants and heterozygous siblings (Figure S1). However, transcription rates of anthocyaninless1 (a1), a direct target of the PL1 transcriptional activator [7,19], were ∼4-fold greater in rmr1–1 mutants (Figure S1), reflecting significantly increased PL1 activity. Transcription rates from colored plant1 (b1)—a locus encoding a basic helix-loop-helix factor genetically required for a1 transcription— remained unchanged. These results were recapitulated in comparisons between nuclei isolated from rmr1–3 mutants and heterozygous siblings in which in vitro transcription assays revealed no significant change in transcription rate of Pl1-Rhoades (Figures 1A and S1; n = 4, two-tailed two-sample t-test, t = 0.8, p = 0.5) while RNase protection experiments showed a 5.7-fold increase in pl1 RNA for rmr1–3 mutants (Figure 1B and 1C; n = 2, two-tailed two-sample t-test, t = 10.8, p < 0.01) using RNA isolated from the same tissues of the same individuals. Similar comparisons from identical tissues but in a different genetic background again showed that transcription rates at pl1 remained unchanged while pl1 RNA levels increased 7.52-fold in rmr1–3 mutants compared to heterozygous siblings (n = 1; see Protocol S1).
These RNA expression results sharply contrast those of previous reports using identical in vitro transcription assays that detected significant differences in Pl1-Rhoades transcription rates between Pl′ and Pl-Rh states and between rmr6 mutants and non-mutants [8,11]. This indicates our in vitro results represent an accurate assessment of transcription rates and not a limitation of the assay to detect rate differences at the pl1 locus. Combined, these results imply an increase of pl1 RNA abundance disproportionate to insignificant changes in transcription rate in rmr1 mutants, the most direct interpretation being that RMR1 functions at a post-transcriptional level to stabilize Pl1-Rhoades RNA.
To better understand Rmr1 function and the paramutation mechanism, we used a map-based approach to identify the rmr1 gene. Using a polymorphic F2 population we looked for genetic linkage between the mutant phenotype and previously mapped chromosome markers [20]. The dark-color phenotype of rmr1–1 homozygotes showed invariant cosegregation with the mutant parent polymorphism of SSLP markers bnlg1174a (680 chromosomes tested; <0.15 cM) and npi252 (60 chromosomes tested; <1.7 cM), indicating rmr1 was tightly linked to those markers in bin 6.05 on Chromosome 6. We used the high degree of synteny between this region and rice Chromosome 5 to identify candidate rmr1 orthologs (Figure 2A and 2B).
Within the syntenic rice region we identified a gene model, Os05g32610 (http://rice.tigr.org/), predicted to encode a Snf2 protein. The Snf2 protein family is composed of members similar to Saccharomyces cerevisiae Snf2p with a bipartite helicase domain containing Pfam SNF2_N and Helicase_C profiles, and includes many proteins involved in ATP-dependent chromatin remodeling [21,22]. While there was no public maize expressed sequence tag for this candidate, we used BLAST searches to identify genomic survey sequence similar to Os05g32610. Oligonucleotide primers were designed from these sequences and used to generate PCR amplicons spanning the maize Os05g32610 ortholog, which were sequenced from individuals homozygous for Rmr1 progenitor alleles and mutant derivatives (see Materials and Methods and Dataset S1). The maize sequence generated from each of the homozygous mutants revealed single unique transition-type base pair changes consistent with EMS mutagenesis relative to the progenitor (Figure 2C). The amino acid change associated with the rmr1–1 allele is predicted to prevent proper folding of the helicase domain [23], while the non-conservative amino acid substitutions associated with the rmr1–2 and rmr1–4 alleles occur at highly conserved residues in the SNF2_N profile (Figure 2D). The rmr1–3 allele is associated with a nonsense mutation predicted to truncate the peptide before the conserved helicase domain. CAPS markers were designed to the potential rmr1–1 and rmr1–3 lesions and used to show that the base pair polymorphisms at each of the probable lesions invariably cosegregate with the mutant phenotype (see Materials and Methods). These results support these polymorphisms as bona fide molecular lesions in the rmr1 gene. Based upon molecular genetic mapping data, DNA sequencing results, and the relevance of the fact that Snf2 proteins affect chromatin environments, we conclude the rmr1 locus encodes a protein containing a Snf2 helicase domain.
Os05g32610 gene models and our cDNA sequencing analysis (see Materials and Methods) indicate rmr1 encodes a 1,435-amino-acid protein. In addition to having the conserved Snf2 helicase domain, the protein has a large N-terminal region with no significant identity to any known or predicted proteins. Phylogenetic comparison with other known Snf2 proteins in maize, rice, Arabidopsis, and budding yeast shows RMR1 is a member of a Rad54-like subfamily defined by DRD1 (Figure 3). Arabidopsis DRD1 is a putative chromatin remodeling factor affecting RNA-directed DNA methylation (RdDM) patterns [24–26]. In the emerging RdDM pathway model, DNA sequences are targeted for de novo cytosine methylation by complementary siRNA molecules generated from “aberrant” RNA transcripts. The putative MOP1 ortholog in Arabidopsis, RDR2, is required in this pathway to presumably generate double-stranded RNA from these transcripts and provide a substrate for siRNA biogenesis through activity of a Dicer-like enzyme [27]. DRD1 is thought to be a downstream effector protein that facilitates de novo methylation of targeted DNA sequences, possibly by modulating chromatin architecture to provide access to de novo methyltransferases [24–26,28]. The DRD1 subfamily also includes the recently identified CLSY1 protein implicated in the systemic spreading of siRNA-mediated silencing in Arabidopsis [29].
Multiple sequence alignments (Figure S2) indicate RMR1 is not the structural ortholog of either DRD1 or CLSY1. The DRD1 subfamily can be divided into three distinct monophyletic groups, with RMR1, DRD1, and CLSY1 defining different groups (Figure 3). The presumed maize ortholog of DRD1 is likely one of two proteins in the DRD1 subgroup, Chromatin remodeling complex subunit R 127 (CHR127) (http://chromdb.org/), a partial protein predicted from maize expressed sequence tag sequences, or CHR156, a full-length protein predicted from maize genomic sequence (see Materials and Methods). RMR1 is more similar to Arabidopsis proteins predicted from At1g05490 and At3g24340. RNA interference knockdowns of these putative Arabidopsis orthologs are known to have little to no effect in response to DNA damage [30].
Taking into account the phylogenetic analysis of the predicted coding sequence, it is possible RMR1 function may be similar to, but distinct from, that of DRD1 and CLSY1. The three proteins may fulfill a similar role in RdDM, but perhaps function under different conditions or in distinct genomic contexts. Alternatively, they could perform different roles within an RdDM pathway, or function in separate epigenetic mechanisms altogether. Given the results of our pl1 RNA expression analyses, it is possible that RMR1 represents a Snf2 protein that links chromatin organization to RNA transcript stability.
In the described Arabidopsis RdDM pathway, DRD1 maintains cytosine methylation at nonsymmetrical CNN sequences represented by siRNAs [24–26]. Many endogenous genomic targets of DRD1 appear to be repetitive elements [31]. At Pl1-Rhoades there is a 402-bp terminal fragment of a CACTA-like type II DNA transposon, similar to doppia, 129 bp upstream of the translational start site [8,32,33]. Assuming analogous functional roles of RMR1 and DRD1 we compared DNA methylation patterns at this upstream repetitive element in rmr1 mutants and non-mutant siblings.
Previous restriction-enzyme-based comparisons of DNA methylation status between Pl-Rh and Pl′ states found no differences, although few 5′ proximal sites were evaluated [8]. Using Southern blot hybridization analysis following digestion of genomic DNA with methylation-sensitive restriction enzymes, we found that the doppia fragment is hypomethylated at specific sites in plants homozygous for the rmr1–1 mutation compared to heterozygous wild-type siblings (Figures 4A, 4B, and S3). Consistent with findings in Arabidopsis RdDM mutants [16,34–36], the sites hypomethylated in rmr1 mutants were of the CNN context. A relative hypomethylation pattern in 5′ sequences is also present in plants homozygous for mutations at either rmr6 or mop1 (Figures S4 and S5). In rmr6 mutants the extent of hypomethylation was greater than that of either rmr1 or mop1 mutants and encompassed CG methylation sites as well as non-CG targets, suggesting Rmr6 has a broader effect in cytosine methylation maintenance. The presence of these methylation differences in multiple mutant backgrounds indicates that this hypomethylation pattern reflects the chromatin status at doppia in plants where maintenance of repressed paramutant states is compromised.
Consistent with the Arabidopsis RdDM model, small RNAs (∼26 nt) with sequence similarity to the doppia element are detected in wild-type Pl′ plants in both sense and antisense orientations (Figures 4D and S6). These small RNAs are undetectable in rmr1 mutants, unlike in wild-type siblings. This result contrasts those in Arabidopsis showing that DRD1 deficiencies do not affect the abundance of endogenous siRNAs representing repetitive elements [31]. However, it has been reported that the abundance of endogenous siRNA and trans-acting siRNA populations are highly reduced in CLSY1 mutants [29].
To test if the doppia fragment hypomethylation was indicative of genome-wide changes we assayed the cytosine methylation status at centromeres and 45S repeat sequences. Cytosine methylation patterns were unaffected in either of these regions in rmr1 mutants as compared to non-mutant siblings (Figure S7). Additionally, we examined the methylation status of doppia-like loci genome-wide (Figure 4E) and found no obvious differences between rmr1 mutants and non-mutant siblings. These results indicate that while RMR1 acts on the doppia sequence upstream of Pl1-Rhoades, doppia elements appear unaffected throughout the genome. This specificity of RMR1 function may be due to its intimate and exclusive involvement with alleles that undergo paramutation, or may be indicative of differential regulation of repetitive elements depending on their genomic and epigenetic context.
If RMR1 is involved in maintaining cytosine methylation patterns characteristic of repressed paramutant states then a prediction would be that the methylation differences seen between mutants and non-mutants would reflect the Pl′ and Pl-Rh regulatory states. Surprisingly, there are no methylation differences at the doppia fragment between Pl-Rh and Pl′ states (Figures 4C and S8). These results suggest that while the upstream doppia element of Pl1-Rhoades is a target of multiple factors involved in maintaining the epigenetic repression associated with paramutation, the actual process of paramutation does not result in similar changes of DNA methylation at this element.
Based on a reverse transcriptase PCR (RT-PCR) expression profile (Figure S9) rmr1 appears to be expressed in all rapidly dividing somatic tissues, consistent with a role in maintaining paramutant states throughout development. However, since the methylation patterns maintained by RMR1 appear unrelated to the paramutant state of Pl1-Rhoades, we questioned whether RMR1 is directly required for paramutation to occur. This process results in the invariable establishment of the Pl′ state in Pl′/Pl-Rh plants, as evidenced by the observation that only Pl′/Pl′ progeny are found when Pl′/Pl-Rh plants are crossed to Pl-Rh/Pl-Rh testers [7,8]. If RMR1 were directly involved in this process we would expect that an rmr1 deficiency might interfere with the Pl′ establishment event. To test this, we tracked the behavior of individual Pl1-Rhoades alleles in test crosses to assess the ability of the Pl′ state to facilitate paramutations in Pl′/Pl-Rh; rmr1–1/rmr1–2 plants. The Pl1-Rhoades allele in a Pl-Rh state was genetically linked (∼1.5 cM) to a T6–9 translocation breakpoint (T6–9). The T6–9 interchange can act as a dominant semi-sterility marker, allowing us to trace specific Pl1-Rhoades alleles through genetic crosses [11]. rmr1 mutants heterozygous for the T6–9 interchange (T6–9 Pl-Rh/Pl′) were crossed to a Pl-Rh/Pl-Rh tester (Figure 5; Table S1). If establishment of the Pl′ state was prevented in rmr1 mutants, we would expect all progeny receiving the interchange to display a Pl-Rh/Pl-Rh phenotype (dark anther pigmentation). We observed that over half the progeny inheriting the interchange displayed a Pl′/Pl′-like phenotype (light anther pigmentation), indicating that paramutation was established in the rmr1 mutant parent. It should also be noted that Pl-Rh/Pl-Rh plants, and those of an intermediate phenotype of partial pigmentation [7], were present in both progeny inheriting the interchange and those inheriting a normal chromosome. These results are consistent with previous work showing Pl′ can revert to a Pl-Rh state in rmr1 mutants [10].
Corresponding analysis of the establishment of paramutant states at the b1 locus generated similar results (Table S2). The repressed B′ state of the B1-Intense allele [37] was established in B′/B-I rmr1 mutants greater than 95% of the time. While it is possible that rmr1 defects affect establishment efficiency, it will be difficult to differentiate any such effects from its clear role in maintenance [11]. These results point to an interesting duality in RMR1 function in which the wild-type protein is necessary for meiotic heritability of repressed epigenetic states, but is not required to establish these states. This duality is markedly different from results generated in the analysis of DRD1, which was shown to be necessary for the maintenance, establishment, and removal of repressive epigenetic marks [24,25].
RMR1 is the first protein identified whose function acts to maintain trans-generationally repressed states associated with paramutation, a genetic behavior that affects meiotically heritable epigenetic variation through allelic interactions at endogenous loci. The identification of RMR1 as a Snf2 protein highlights an emerging role of these proteins in establishing and maintaining epigenetic marks. In Arabidopsis the Snf2 proteins DRD1 and DDM1 [38,39] are known to maintain cytosine methylation patterns. Lsh1, the mammalian protein most closely related to DDM1, is also required for normal DNA methylation patterns [40–42]. There are some 42 Snf2 proteins in Arabidopsis and at least as many in maize (http://chromdb.org/). This diversity likely represents great functional specialization amongst these proteins. We have placed RMR1 in an RdDM pathway based on its helicase domain similarity to DRD1 and the recent identification of MOP1 as an RDR2 ortholog [14,15]. Consistent with this proposed pathway, the rmr1 mRNA expression profile (Figure S9) closely matches that of mop1 [15]. Additionally, both RMR1 and MOP1 are necessary to maintain cytosine methylation patterns at silenced transgenes [43], the Pl1-Rhoades doppia sequences, and certain Mutator transposable elements ([15,44]; J. B. H. and D. Lisch, unpublished data). DRD1 is also known to target repetitive elements found in euchromatic contexts through an RdDM pathway [31]. However, the role RMR1 plays to maintain the repressed paramutant states at Pl1-Rhoades appears different than the function of DRD1 in the Arabidopsis RdDM pathway, as RMR1 has, in addition to its requirement for CNN methylation at doppia, a role in the normal accumulation of small RNAs with similarity to that element.
It is unclear how RMR1 mediates the post-transcriptional regulation of pl1 transcripts as suggested by the in vitro transcription and RNase protection assays reported here. It is possible that pl1 transcripts resulting from Pl1-Rhoades in the Pl′ state are less stable than those produced from the Pl-Rh state because of differences in the chromatin environment of Pl1-Rhoades. However, there do not appear to be any Pl′-specific small RNAs produced from the pl1 coding region [12]. In S. pombe it has been shown that the chromatin environment of a locus can affect RNA transcript levels without altering RNA polymerase II occupancy of that locus, leading to the proposal of a cotranscriptional gene silencing mechanism whereby nascent transcripts initiating in a heterochromatic environment are degraded by complexes targeted via heterochromatic small RNAs [17,18]. Chromatin differences in the upstream region of Pl1-Rhoades may favor recruitment of alternative RNA-processing factors or RNA polymerases, which in turn influence the stability of pl1 transcripts. In plants, localization of the large subunit 1a of RNA polymerase IV to loci targeted for RdDM appears necessary for the biogenesis of siRNAs from these loci [28]. When Pl′ repression is disrupted in rmr1 mutants, this alternate genesis or processing of the pl1 transcript may also be lost. Alternatively, our results may highlight a novel role for RMR1-like Snf2 proteins in directly interacting with nascent RNA transcripts via a helicase domain, or in recruiting factors that directly destabilize these transcripts.
Importantly, our analysis of rmr1 mutants calls into question the relationship between RMR1 function and the mechanism of paramutation at Pl1-Rhoades. The mutational screens identifying rmr1, rmr6, and mop1 were designed to discover genetic components necessary to maintain the repressed state of Pl′, not necessarily factors needed to establish this repressed state [10,13]. Therefore, it is possible that loci thus far identified may be indirectly related to the paramutation mechanism. Our results are consistent with a model wherein RMR1 functions in an RdDM pathway, along with an RDR2-like enzyme, MOP1, to maintain a persistent heterochromatic-like chromatin structure at the repetitive element found directly upstream of the pl1 coding region. While it is not clear where RMR1 acts in this pathway it presumably acts coordinately with the maize orthologs of known RdDM components identified in Arabidopsis, namely DCL3 [16,45], the DRM methyltransferases [36], AGO4 [46,47], the RNA polymerase IV subunits, and the maize DRD1 ortholog (Figure 6A). In this model, doppia transcripts, perhaps because of the repetitive nature of the doppia genomic elements and/or the numerous internal subterminal repeats that are present in these elements [32,48], are the source of aberrant RNA that is processed via MOP1 and a DCL3 enzyme into siRNAs. This small RNA production is carried out in a manner that is dependent on RMR1 activity, possibly via direct interaction with a small RNA processing complex or by making the DNA accessible to factors necessary for siRNA precursor generation such as polymerase IVa. These siRNAs, through the activity of AGO4, DRM enzymes, and polymerase IVb, then establish a heterochromatic state at the Pl1-Rhoades doppia-like element that is present in both Pl-Rh and Pl′ states. The methylation effects seen in rmr1 mutants might indicate that this heterochromatization machinery depends on the activity of RMR1 to feed back on the doppia element, or loss of RMR1 may short circuit this pathway and thus affect methylation activity indirectly. An RMR1 defect then affects stability of paramutant states at pl1 because of the chromatin context of the Pl1-Rhoades allele, and not through direct disruption of components required for paramutations to occur. This is in line with a report that MOP1-dependent small RNAs produced at the b1 locus are insufficient to mediate paramutation [49].
The relationship between RMR1 action, the chromatin organization of Pl1-Rhoades, and the repressed Pl′ state is not clearly understood at this time. It is possible that derepression of the upstream repetitive element makes the region more accessible to general transcription factors whose actions could destabilize repressive Pl′ chromatin states that are independent of those maintained at doppia (Figure 6A). Indeed, RNA polymerase processivity can lead to changes in the chromatin environment through histone modifications or histone replacement [50,51]. Alternatively, Pl′ chromatin states may represent a spreading of the heterochromatic domain at doppia into a euchromatic region defined by the Pl1-Rhoades gene space (Figure 6B). In fission yeast, heterochromatic domains nucleated by small RNAs have the ability to spread in cis through successive H3 K9 methylation [52]. In this situation, loss of RMR1 function would alleviate Pl′ repression by disrupting maintenance of this expanded heterochromatic domain. In either of these situations RMR1 affects Pl1-Rhoades paramutations by virtue of its role in maintaining heterochromatic states at a proximal repetitive element.
McClintock was the first to describe derivative alleles in which transposons acted to control the expression patterns of attendant genes [53]. It is now clear that epigenetic modulations of the transposons themselves—what McClintock referred to as “changes in state”—can alter the regulatory properties of individual genes both somatically [54] and trans-generationally [55,56]. Our results indicate that even transient changes in state of the Pl1-Rhoades doppia fragment can have trans-generational effects on pl1 gene expression patterns. These experimental examples, in the context of McClintock's thesis [53], point to a dynamic source of regulatory, and potentially adaptive, variation adjunct to the DNA itself. Precisely how this epi-variation relates to existing genome structure and function, as well as its evolutionary potential, remains a largely unexplored area of investigation.
Currently, well-characterized examples of paramutation are limited to loci where expression states have a clear phenotypic read-out, such as pigment synthesis. cis-Elements required to facilitate paramutation have been functionally identified at specific alleles of b1 and colored1 (r1) [57–59]. To date, there is no evidence that the chromatin status of these cis-elements is affected by mutations at trans-acting loci required for maintenance of repressed paramutant states. It appears that paramutations represent a type of emergent system wherein genomic context and maintenance of chromatin states interact to facilitate meiotically heritable epigenetic variation. In this view, it is possible that cis- and trans-elements necessary for maintenance of such variation might not interact in a direct and predictable manner. What remains to be seen is the extent to which this type of system acts throughout the genome. Genome-wide screens for paramutation-like behavior, in which expression states are affected by allele history, remain technologically and conceptually challenging. Recent work by Kasschau et al. [60] suggests that in Arabidopsis, few endogenous genes are regulated by proximal presumed RdDM targets. However, it is tempting to speculate that examples of paramutation represent an exception to this trend, representing a mechanism by which populations can quickly, and heritably, change their transcriptome profile and regulation.
Plants were scored as carrying Pl-Rh or Pl′ states through visual inspection of anther pigmentation and assignment of an anther color score as previously described [7]. Pl′/Pl′ (anther color score 1 to 4) anthers show little to no pigmentation while Pl-Rh/Pl-Rh (anther color score 7) anthers are dark red to purple. Mutants were scored in the same way, with rmr and mop mutants showing a Pl-Rh/Pl-Rh-like phenotype, except in the case of the F2 rmr1 mapping populations, in which mutants were chosen on the basis of a dark seedling leaf phenotype [10].
Elite inbred lines (B73, A619, and A632) were provided by the North Central Regional Plant Introduction Station (http://www.ars.usda.gov/main/site_main.htm?modecode=36-25-12–00). Color-converted versions of A619 and A632 inbred lines were created by introgressing the Pl1-Rhoades allele into each [11]. The rmr1–1, rmr1–2, mop1–1, and rmr6–1 alleles have been previously described [8,10,13]. The rmr1–3 allele was derived from identical materials used to isolate rmr1–1 and rmr1–2; rmr1–4 was derived from EMS-treated pollen from an A619 color-converted line applied to a color-converted A632 line [11] (see Protocol S1 and Table S3 for complementation tests). The T6–9 translocation line carrying the Pl1-Rhoades allele used in Pl′ establishment tests has been described previously [11].
In vitro transcription assays (rmr1–1 and rmr1–3; Figures 1 and S1) and RNase protection assays (rmr1–3 only; Figure 1) were carried out as described [8] with husk nuclei and RNA isolated from single ears of the same genetic stocks used to measure pl1 RNA differences in rmr1–1 anthers [10]. The b1 and pl1 genotypes of these plants are as follows: B1-Intense (B-I)/B-I; Pl1-Rhoades (Pl′) Rmr1/Pl′ rmr1–1 and B-I/B-I; Pl′ rmr1–1/Pl′ rmr1–1, or B-I/B-I; Pl′ Rmr1/Pl′ rmr1–3 and B-I/B-I; Pl′ rmr1–3/Pl′ rmr1–3. Identical procedures were applied to single ears from plants homozygous for Pl′ and either homozygous or heterozygous for rmr1–3 following a single backcross into the KYS inbred line [12]. Additional details regarding stock syntheses are available upon request.
A F2 mapping population was created from inbred (S9) rmr1–1/rmr1–1, Pl′/Pl′, and color-converted A632 inbred (Pl′/Pl′, >93% A632) parents. DNA was isolated using the DNeasy 96 plant kit (Qiagen, http://www1.qiagen.com/) from F2 mutant seedlings, mapping parents, and F1 hybrid leaf tissue. These DNA samples were screened with SSLP markers developed from the Maize Mapping Project (http://www.maizemap.org/; US National Science Foundation award number 9872655; primer sequences and protocol available at http://maizegdb.org/). Initial marker choice was restricted to Chromosomes 6 and 9 because of linkage of rmr1 to a T6–9 breakpoint. In addition to the rmr1–1 mapping population, a second F2 mapping population created with inbred (S7) rmr1–3/rmr1–3, Pl′/Pl′, and color-converted A632 parents showed similar cosegregation with marker bnlg1174a (178 chromosomes tested; <0.56 cM). CAPS [61] markers were designed to test cosegregation of the rmr1–1- and rmr1–3-associated lesions with the rmr1 mutant phenotype (see Protocol S1 for details). No recombinant chromosomes (876 chromosomes tested for rmr1–1, 268 chromosomes tested for rmr1–3) were found using either marker.
A BLAST search using the rice Os05g32610 ORF as a query identified maize GSS and sorghum expressed sequence tag sequences that were used to generate a contig representing the putative maize gene (see Protocol S1 for sequence identifiers). Oligonucleotide primers (Sigma-Genosys, http://www.sigmaaldrich.com/Brands/Sigma_Genosys.html) were designed from these sequences and used in PCR amplification of genomic DNA from three separate individuals homozygous for each rmr1 mutant allele as well as functional reference alleles Rmr1-B73, Rmr1-A632, and Rmr1-A619. PCR amplicons were purified using QIAquick gel extraction kit (Qiagen) and dideoxy sequenced (UC Berkeley DNA Sequencing Facility, http://mcb.berkeley.edu/barker/dnaseq/). To verify the intron/exon structure of rmr1, cDNA was generated from rmr1–1 mutants as well as non-mutant B73 plants as described [15], and rmr1 was amplified via RT-PCR. The resulting products, which were the predicted size for spliced rmr1 transcript, were sequenced to validate the intron/exon structure shown in Figure 2C. See Protocol S1 and Table S4 for all oligonucleotide primer sequences used.
Sequencing reads from genomic and cDNA were aligned and edited with Sequencher (Gene Codes, http://www.genecodes.com/) to create a contig representing rmr1. The N-terminal prediction is based on alignment of RMR1 with the protein model for Os05g32610. A search of the Pfam database (http://www.sanger.ac.uk/Software/Pfam/) with the predicted RMR1 protein sequence was used to identify the conserved SNF2_N and Helicase_C protein profiles of the Snf2 helicase domain. MUSCLE [62] was used to generate an alignment between RMR1 and proteins from Arabidopsis, rice, maize (CHR127 and CHR156), and budding yeast over the helicase domain (Figure S2). Sequences for CHR127 and CHR156 were retrieved from ChromDB (http://www.chromdb.org/). Additional sequence information for CHR156 was identified from BAC CH201-3L17 (GenBank accession AC194602), and gene model prediction was performed using FGENESH+ (Softberry, http://www.softberry.com/) with RMR1 as similar protein support. A distance tree was created and bootstrap values were calculated using PAUP* 4.0 from the above alignment (Sinauer Associates, http://www.sinauer.com/).
Genomic DNA was isolated as described [13] from the terminal flag leaves of adult plants segregating for rmr1, rmr6, and mop1 mutants and heterozygous siblings as well as Pl′ and Pl-Rh plants as assayed by anther pigmentation [7,8,10,13]. Restriction digest and subsequent Southern blots were carried out as previously described [13], using the restriction enzymes listed in Figure 4 (New England Biolabs, http://www.neb.com/). The probes specific to pl1 are shown in Figure 4; the 45S and centromere probes are as described [13].
Small RNAs were prepared from 10-mm immature ear tissue and used to generate small RNA northern blots as previously described [63]. In Figure 4D the small RNAs were run with a 27-bp DNA oligonucleotide containing doppia sequence that hybridized with the riboprobe used to identify the small RNAs. The riboprobe was synthesized as described [63] from a plasmid containing the region denoted probe B in Figure 4A linearized at an AseI site so as to contain only doppia sequence.
Establishment of the Pl′ state in rmr1 mutants was assayed essentially as described previously [11]. When the T6–9 interchange pair is heterozygous with structurally normal chromosomes, the plants display ∼50% pollen sterility due to meiotic-segregation-induced aneuploidy in the resulting gametes. Pollen sterility was assayed in the field using a pocket microscope. rmr1 mutants were crossed to Pl-Rh/Pl-Rh A619 or A632 inbreds (Table S1), and the resultant progeny were scored with respect to Pl1-Rhoades expression state.
|
10.1371/journal.ppat.1001130 | The SR-BI Partner PDZK1 Facilitates Hepatitis C Virus Entry | Entry of hepatitis C virus (HCV) into hepatocytes is a multi-step process that involves a number of different host cell factors. Following initial engagement with glycosaminoglycans and the low-density lipoprotein receptor, it is thought that HCV entry proceeds via interactions with the tetraspanin CD81, scavenger receptor class B type I (SR-BI), and the tight-junction proteins claudin-1 (CLDN1) and occludin (OCLN), culminating in clathrin-dependent endocytosis of HCV particles and their pH-dependent fusion with endosomal membranes. Physiologically, SR-BI is the major receptor for high-density lipoproteins (HDL) in the liver, where its expression is primarily controlled at the post-transcriptional level by its interaction with the scaffold protein PDZK1. However, the importance of interaction with PDZK1 to the involvement of SR-BI in HCV entry is unclear. Here we demonstrate that stable shRNA-knockdown of PDZK1 expression in human hepatoma cells significantly reduces their susceptibility to HCV infection, and that this effect can be reversed by overexpression of full length PDZK1 but not the first PDZ domain of PDZK1 alone. Furthermore, we found that overexpression of a green fluorescent protein chimera of the cytoplasmic carboxy-terminus of SR-BI (amino acids 479–509) in Huh-7 cells resulted in its interaction with PDZK1 and a reduced susceptibility to HCV infection. In contrast a similar chimera lacking the final amino acid of SR-BI (amino acids 479–508) failed to interact with PDZK1 and did not inhibit HCV infection. Taken together these results indicate an indirect involvement of PDZK1 in HCV entry via its ability to interact with SR-BI and enhance its activity as an HCV entry factor.
| Hepatitis C virus (HCV) infection is a major cause of serious liver disease, with approximately 170 million people infected worldwide. Although significant advances have been made in the characterization and development of novel therapeutics to combat HCV infection, there is still a great need for an improved understanding of the HCV lifecycle and potential future targets of antiviral therapy. HCV entry into hepatocytes involves numerous plasma membrane proteins including CD81, scavenger receptor class B type I (SR-BI), claudin-1 and occludin. Although these proteins may comprise the complete set of essential HCV entry factors, the secondary factors that influence the co-ordinated involvement of these proteins in HCV entry remain to be determined. Here we identify the SR-BI partner protein PDZK1 as an indirect regulator of HCV entry. Our results indicate that binding of PDZK1 to the cytoplasmic carboxy-terminus of SR-BI influences the receptor's involvement in HCV entry such that disruption of this interaction may represent a future target of therapeutic intervention.
| It is estimated that approximately 170 million people worldwide are infected with hepatitis C virus (HCV); a major cause of serious liver disease. At present there is no preventative vaccine available and the widely preferred treatment regime of pegylated interferon alpha (IFN-α) and ribavirin in combination is expensive, causes adverse side effects and is only effective for a fraction of individuals. Despite significant advances in identification of novel antiviral agents that inhibit HCV replication and polyprotein processing, concerns remain regarding the toxicity of these compounds and the likelihood of development of antiviral resistance [1]. The rapidly increasing understanding of the HCV entry process and significant advances in the development and application of HIV entry inhibitors (for review see [2]) have lead to a growing appreciation that HCV entry is another promising target for future antiviral therapies.
The recent development of the retroviral HCV pseudoparticle system (HCVpp), in which HCV E1E2 glycoproteins are assembled onto retroviral cores [3], [4], [5], and the infectious HCV cell culture (HCVcc) system, in which the full viral lifecycle is recapitulated in cell culture [6], [7], [8], have allowed in-depth analysis of the HCV entry process. At present there is strong evidence to suggest that the essential HCV entry factors include the tetraspanin CD81 [5], [9], [10], [11], the class B scavenger receptor SR-BI [9], [12], [13], [14], and the tight-junction proteins claudin-1 and occludin [15], [16], [17], [18], [19], [20]. Considering that these proteins may comprise the complete set of essential HCV entry factors [18], it still remains to be determined what the relative involvement of each of these entry factors is and, beyond expression, what secondary factors influence the contribution of these proteins to HCV entry.
SR-BI is the major receptor for high-density lipoproteins (HDL) and mediates both bi-directional flux of free cholesterol between cells and lipoproteins and selective uptake of cholesteryl esters into cells from HDL (reviewed in [21]). The latter function is of greatest significance in the liver and steroidogenic tissues [22], [23], where SR-BI is most highly expressed [24]. Studies using rodents have revealed that hepatic SR-BI expression is subject to little transcriptional regulation but instead is largely regulated at the post-transcriptional level by its interaction with the cytoplasmic adaptor molecule PDZK1 (reviewed in [25]).
PDZK1, which is also known as NHERF3, CAP70, CLAMP and NaPi-Cap1, is a four PDZ domain-containing adaptor protein that is predominantly expressed in the liver, kidney and small intestines [26]. Since the demonstration that the extreme C-terminus of SR-BI interacts with the first N-terminal PDZ domain of PDZK1 [26], [27], a number of in vivo and in vitro studies have demonstrated the importance of this interaction to the plasma membrane content of SR-BI and its activity as an HDL receptor in hepatocytes. Strikingly, in PDZK1 knockout (KO) mice, hepatic levels of SR-BI protein are reduced by greater than 95% and plasma HDL-cholesterol is elevated [28] to a level that approaches those levels seen in SR-BI KO mice [29]. Although these effects suggested the involvement of PDZK1 in regulating the stability of SR-BI and its effective targeting to the plasma membrane in hepatocytes, further studies have since revealed that hepatic overexpression of SR-BI in SR-BI/PDZK1 double knockout mice results in effective targeting of SR-BI to the hepatocyte plasma membrane and restoration of apparently normal lipoprotein metabolism [30]. The authors of this work therefore concluded that PDZK1 is not essential for the correct localization and function of SR-BI in the liver but instead is required for maintenance of ‘steady state levels’ of hepatic SR-BI protein [30].
Recently the importance of PDZK1/SR-BI interaction to the total hepatic abundance, plasma membrane content and activity in lipoprotein metabolism of SR-BI has been further dissected by studies involving the hepatic overexpression of C-terminally truncated PDZK1 mutants in transgenic mice [31], [32]. While overexpression of the first PDZ domain (PDZ1) of PDZK1 could not rescue normal SR-BI expression and activity in PDZK1 KO mice, overexpression of PDZ1 in wildtype mice resulted in cytoplasmic retention of endogenous SR-BI and a commensurate effect on lipoprotein metabolism [32]. Further studies revealed that transgenic expression of all four PDZ domains of PDZK1 is required to restore normal SR-BI protein abundance, plasma membrane content and activity in lipoprotein metabolism in PDZK1 KO mice [31]. Collectively these results indicate that interaction of the C-terminus of SR-BI with PDZK1, in itself, is not sufficient to enhance total SR-BI expression, plasma membrane localization and function. Instead other features of PDZK1 appear to be necessary for its impact on SR-BI expression and function. These may include phosphorylation of Ser-509 in the C-terminus of PDZK1, which is associated with increased SR-BI abundance in rat hepatoma cell culture [33] and/or association with other proteins in a macromoleular complex.
The importance of the cytoplasmic carboxy-terminus of SR-BI to its involvement in HCV entry has been examined in two recent studies [12], [13]. Firstly, Grove et al reported that soluble E2 binding and HCVcc infection levels are enhanced by overexpression of SR-BI or SR-BII, a splice-variant of SR-BI which features an alternative cytoplasmic carboxy terminus [13]. However, overexpression of SR-BI was associated with increases in HCVcc infection levels that were several-fold higher than those observed for overexpression of SR-BII [13], suggesting that features of the cytoplasmic tail dictate the molecule's efficient involvement in HCV entry. Secondly, it has recently been reported that complementation of SR-BI expression in rodent hepatoma cells co-expressing human CD81 and claudin-1 rendered these cells susceptible to infection with HCVpp and that this effect was limited when SR-BI constructs bearing various mutations in the cytoplasmic carboxy-terminus were substituted [12]. In that study, however, the extreme carboxy-terminus of SR-BI and expression of PDZK1 were not found to be significant determinants of HCV entry levels.
In the present study we have examined the association of human SR-BI and PDZK1 and how this impacts upon HCV entry and replication. Specifically we show that the association between SR-BI and PDZK1 is important for efficient entry of HCV into hepatoma cells and suggest that disruption of this interaction may be a future target of anti-HCV therapy.
To date studies of the interaction of SR-BI and PDZK1 and the functional significance of this interaction have been performed using rodents and rodent-derived cell lines. To confirm the interaction of human SR-BI and PDZK1 and to examine the requirement of certain domains of each protein for the interaction, co-immunoprecipitation (co-IP) experiments were performed. For this, 293T cells were co-transfected with expression plasmids encoding wildtype or mutant variants of SR-BI and PDZK1 (Figure 1A) prior to immunoprecipitation of FLAG-tagged PDZK1 proteins and immunoblot detection of Myc-tagged SR-BI proteins (Figure 1B; lower panel). In these experiments 293T cells were used as they can be transiently transfected at high efficiency, therefore allowing ready detection of overexpressed proteins in whole cell lysates and immunoprecipitates. As expected, wildtype SR-BI readily co-immunoprecipitated with wildtype PDZK1, while a truncated SR-BI variant that lacks the final carboxy-terminal lysine residue (mycSR-BIΔ509) was not detectable in immunoprecipitates of co-transfected wildtype PDZK1. Furthermore we confirmed that the first amino-terminal PDZ domain of PDZK1 (PDZ1) was sufficient to co-IP wildtype SR-BI, with the NYGF motif present at amino acid residues 19–22 of PDZ1 being a predicted site of interaction with the carboxyl terminus of SR-BI. Next we examined whether phosphorylation of PDZK1 at Ser-505 impacted upon PDZK1/SR-BI association. For the sake of consistency with the previous report of phosphorylation of the corresponding site in rodent PDZK1 [33], this site will henceworth be referred to as Ser-509. While mutation of this site (S509A and S509D) resulted loss of serine phosphorylation in 293T cells (Figure 1C), both phosphodefective (S509A) and phosphomimetic (S509D) mutants of PDZK1 associated with wildtype SR-BI with no apparent difference in the relative amount of SR-BI that was co-immunoprecipitated with these PDZK1 mutants (Figure 1B). Next, given that the C-terminus of SR-BI has been implicated in its dimerization and multimerization [34], we investigated whether the PDZK1-interacting domain at the C-terminus of SR-BI was involved in receptor dimerization. Co-IP studies in transfected 293T cells revealed that mycSR-BIΔ509 readily interacts with FLAG-tagged full-length SR-BI and FLAG-SR-BIΔ509, suggesting that PDZK1-interaction is not involved in homodimerization of SR-BI (Figure S1, A). Similarly, co-IP studies in Huh-7 cells confirmed dimerization of FLAG-SR-BI with mycSR-BIΔ509 and the ability of mycSR-BI to interact with PDZK1-FLAG in this cell type (Figure S1, B). Accordingly confocal analysis of the localization of full length SR-BI and SR-BIΔ509 revealed no appreciable difference in the localization of these proteins in transfected Huh-7 cells (Figure S1, C).
Finally, confocal analysis revealed extensive co-localization of non-tagged wildtype SR-BI with FLAG-tagged PDZK1 in the cytoplasm of transfected Huh-7 hepatoma cells (Figure 1D), with no apparent effect of mutation of PDZK1 on its localization or co-localization with SR-BI for each of the mutants S509A and S509D. However, overexpressed PDZ1 displayed additional nuclear localization, despite substantial co-localization with cytoplasmic SR-BI. To reveal the localization of surface-displayed endogenous SR-BI with respect to overexpressed PDZK1-FLAG, transfected Huh-7 cells were fixed with formalin and SR-BI was labelled by indirect immunofluorescence with an antibody that recognizes the extracellular loop of SR-BI (and SR-BII) prior to subsequent fixation, permeabilization and labelling of PDZK1-FLAG (Figure 1E). These studies revealed a small degree of overlap in the localization of surface SR-BI and PDZK1-FLAG that was most evident at sites of cell-cell contact. Taken together these experiments demonstrate that the interaction of SR-BI with PDZK1 is not dependent on Ser-509 phosphorylation of PDZK1 and that the disruption of this interaction by truncation of SR-BI cannot be attributed to an appreciable alteration in its localization.
Having confirmed that human SR-BI and PDZK1 interact we next investigated the involvement of PDZK1 in entry of HCV using stable shRNA-knockdown of endogenous expression of PDZK1 in Huh-7 cells. Despite effective knockdown of PDZK1 expression, total SR-BI protein levels were not significantly altered (Figure 2A). Furthermore, cell-surface biotinylation experiments demonstrated that SR-BI remained readily detectable in streptavidin precipitates of plasma membrane proteins in PDZK1-knockdown cell lines (Figure 2B). Interestingly PDZK1 was co-precipitated with biotinylated plasma membrane proteins purified from control shRNA-expressing cells, but not from PDZK1 knockdown cells indicating that residual endogenous PDZK1 in the latter cells is not preferentially plasma membrane-associated (Figure 2B). Since PDZK1 has no transmembrane domains it is likely that enrichment of PDZK1 in streptavidin precipitates in these studies is the result of its association with the cytoplasmic tail of SR-BI at the inner leaflet of the plasma membrane and possibly other transmembrane partner proteins present in these cells. Furthermore, flow cytometric analysis of surface levels of SR-BI and splice variant SR-BII revealed no impact of PDZK1 knockdown on surface levels of these proteins (Figure 2C). Likewise surface levels of CD81 were not significantly altered by PDZK1 knockdown (Figure 2C). Interestingly, confocal analysis of surface-labelled cells revealed extensive colocalization of SR-BI and CD81 which was not appreciably altered by PDZK1 knockdown (Figure S2, A). Similarly we did not note any changes in the staining pattern of occludin in PDZK1 knockdown cells compared to control cells (Figure S2, B). Given that the extracellular domain of SR-BI is shared by splice-variant SR-BII [35], experiments involving confocal analysis of the localization of intracellular SR-BI and surface SR-BI/II in the same samples were undertaken to confirm that the majority of surface SR-BI/II staining is also co-stained with an antibody directed against the SR-BI-specific cytoplasmic C-terminus (Figure S3). From these images it was also apparent that PDZK1 knockdown does not cause an appreciable shift in the proportion of intracellular and surface SR-BI staining. Together these results indicate that near-complete loss of PDZK1 expression in Huh-7 cells does not significantly alter total or surface levels of SR-BI or its localization with respect to other requisite entry factors CD81 and occludin at the cell surface.
Next we investigated the influence of stable knockdown of PDZK1 expression in Huh-7 cells on HCV entry and replication. Following infection of these cells with cell culture propagated HCV (HCVcc; JFH-1), HCV RNA levels were approximately 40% lower in the PDZK1-knockdown cell lines compared to cells that expressed a non-targeting shRNA control (Figure 3A). Likewise, the relative susceptibility of these cells to HCVcc infection was approximately 50% lower than control cells, as determined by enumeration of HCV-positive foci (focus forming units/ml; FFU/ml) following indirect immunofluorescent labelling (Figure 3B). Similarly infection of these cells with Jc1/GFP or Jc1/RFP infectious HCV chimeras, that bear the GFP- or RFP-coding sequences inserted in-frame into domain III of NS5A of the infectious genotype 2a chimera Jc1 [36], and flow cytometric analysis of GFP- or RFP-associated epifluorescence revealed that PDZK1-knockdown cells were approximately 80% (Jc1/GFP) and 60% (Jc1/RFP) less susceptible to infection with these chimeras than their control counterparts (Figures 3C and 3D). To confirm that the observed effects were attributable to an impact of PDZK1-knockdown on HCV entry, and not an effect on HCV replication or spread, HCVpp (H77s) entry into these cells was measured. As for HCVcc infections, HCVpp entry was significantly reduced by PDZK1 knockdown (Figure 3E). In these experiments VSVG-pseudoparticle infection levels were unchanged by PDZK1 knockdown, indicating that PDZK1 knockdown causes HCV-specific inhibition of viral entry. Further evidence that HCV entry, and not HCV replication, is limited by PDZK1 knockdown was provided by the demonstration that stable knockdown of PDZK1 expression in Huh-7 cells that harbour the genome-length NNeo/C-5B(RG) replicon did not significantly impact upon HCV RNA levels or HCV-associated immunofluorescence (Figure S4). From these experiments we conclude that expression of PDZK1 contributes to HCV entry into Huh-7 cells most probably via an influence on its binding partner SR-BI despite no dramatic effects on total or cell surface levels of SR-BI protein. Given our observations that PDZK1 knockdown is associated with decreased HCV entry levels for both genotype 1a (HCVpp) and 2a (HCVcc) infection systems and the high degree of divergence between E1/E2 sequences for these genotypes, it is likely that the observed involvement of PDZK1 will hold true for all HCV genotypes. However, it has been reported that the relative involvement of SR-BI in HCV entry can differ between HCV genotypes and subtypes [37] and thus the relative influence of PDZK1 on SR-BI-dependent HCV entry may differ between HCV genotypes/subtypes accordingly.
Given the contrasting effects on total and cell surface SR-BI protein levels of PDZK1 knockdown in Huh-7 cells and PDZK1 gene knockout in mice [28] and the past demonstration that ectopically expressed SR-BIdel509 functions similarly to wildtype SR-BI in transfected cell lines but is unstable and mislocalized in transgenic mouse liver [27], we set about generating a polarized HepG2 cell culture model which, in some respects, would more accurately reflect the in vivo liver situation than non-polarized Huh-7 cells. To this end we stably overexpressed CD81 in the HepG2(N6) cell line, which displays simple columnar polarity [38], to render these cells permissive to HCVcc and HCVpp infection [9], [11], [39], [40]. Following confirmation of high-level cell surface expression of CD81 (not shown), these cells were stably transduced with PDZK1-specific or non-target control lentiviral shRNA vectors. As for Huh-7 cells, stable knockdown of endogenous PDZK1 expression in HepG2(N6)+CD81 cells that had been grown under polarizing conditions (5 days post-confluency) did not significantly alter total levels of SR-BI protein (Figure 4A). Likewise there was no discernable impact of PDZK1 knockdown on cell surface levels of SR-BI or CD81, as determined by Western analysis following cell-surface biotinylation and streptavidin precipitation of plasma membrane associated proteins (Figure 4B). However, we found that PDZK1 knockdown was associated with a moderate yet significant reduction in the susceptibility of HepG2(N6)+CD81 cells to HCVcc (Jc1/Myc) infection (not shown). Moreover HCVpp infection was substantially reduced in confluent HepG2(N6)+CD81 cells that expressed PDZK1-specific shRNA's compared to control shRNA-expressing counterparts (Figure 4C). Taken together these results demonstrate that knockdown of endogenous PDZK1 expression in HepG2(N6)+CD81 cells results in their reduced susceptibility to HCV infection.
To further confirm the involvement of PDZK1 in HCV infection we generated a lentiviral PDZK1-FLAG expression construct bearing silent mutations (Stca261Sagc) to render the encoded transcripts refractory to shRNA silencing by PDZK1 shRNA#5. This construct was then overexpressed in both non-target shRNA control and PDZK1 shRNA#5 Huh-7 cell lines, to examine whether overexpression of the shRNA-refractory PDZK1 construct would impact upon HCVcc infection and whether ‘normal’ levels of HCVcc infection could be restored. In these experiments a PDZ1-FLAG lentiviral expression construct served as an additional control that was expected to interact with the C-terminus of SR-BI but not restore PDZK1 function in PDZK1 knockdown cells. Following generation of stable cell lines using these vectors Western analysis showed strong expression of PDZK1-FLAG and PDZ1-FLAG that was comparable between each of the cell lines and did not alter total endogenous SR-BI protein levels (Figure 5A), while parallel immunofluorescent labelling of the overexpressed FLAG-tagged proteins indicated that over 70% of cells expressed these constructs for each of the cell lines (not shown).
Infection of these cells with HCVcc (Jc1/Myc) and quantification of HCV-positive foci three days later revealed that overexpression of full-length PDZK1 significantly increased HCVcc infection levels in shRNA control cell lines and restored levels of HCVcc infection in PDZK1 shRNA#5-expressing cells to levels approaching those of the parental non-target shRNA control cells (Figure 5B). In contrast to the effects of overexpression of full-length PDZK1-FLAG, overexpression of PDZ1-FLAG did not restore HCVcc infection levels in PDZK1 shRNA#5 cells to those of non-target shRNA control cells. Given that hepatic expression of a similar murine PDZ1 construct is reported to cause mislocalization of endogenous SR-BI in transgenic mice and a commensurate effect on HDL metabolism [32], we anticipated that overexpression of PDZ1-FLAG would have a similar dominant-negative effect on HCVcc infection. In contrast we did not observe any significant impact of PDZ1-FLAG expression on the susceptibility of either non-target shRNA- or PDZK1 shRNA-expressing cells to HCVcc infection. However, we have observed that the PDZ1 polypeptide used in this study appears somewhat instable compared to full-length PDZK1 (for example see Figure 1B) and it is possible that high-level overexpression of an alternative construct that encodes a more stable PDZ1 variant may cause the inhibitory effects on HCV entry that were predicted. Taken together these data indicate that the ability of PDZK1 to enhance HCV infection requires regions of the protein that lie outside the SR-BI-interacting domain of the molecule.
To further investigate the domains of each protein involved in SR-BI/PDZK1 association and to investigate the validity of a putative dominant-negative inhibitor of the interaction, the C-terminal 30 amino acids of the cytoplasmic carboxy-terminus of wildtype SR-BI (WT-ctt) were appended in-frame to the carboxy-terminus of enhanced green fluorescent protein (EGFP). As for full-length SR-BI, the EGFP-WT-ctt chimera was readily detectable in co-immunprecipitates of co-transfected wildtype PDZK1 (Figure 6A). Importantly, a control chimera that lacked the final C-terminal lysine residue (EGFP-Δ509-ctt) did not co-IP with wildtype PDZK1. As for full-length SR-BI, EGFP-WT-ctt was co-immunoprecipitated with wildtype PDZK1, PDZ1, PDZK1 S509A and PDZK1 S509D.
Next, Huh-7 cell lines that stably express EGFP, EGFP-WT-ctt or EGFP-Δ509-ctt were generated. Confocal analysis of the localization of these fluorescent protein chimeras revealed that EGFP-WT-ctt localized to distinct cytoplasmic punctae that partially co-localized with LAMP1, a resident protein of late endosomes and lysosomes (Figure 6B). In contrast, EGFP-Δ509-ctt, which does not interact with PDZK1, was indistinguishable in localization to unmodified EGFP (Figure S5, B), suggesting that the distinctive punctate cytoplasmic localization of EGFP-WT-ctt could be attributed to its ability to interact with endogenous PDZK1. Western analysis of these cells revealed that stable expression of EGFP-WT-ctt had no significant effect on the total endogenous PDZK1 and SR-BI content in Huh-7 cells compared to cells that stably expressed unmodified EGFP or EGFP-Δ509-ctt (Figure 6C). Similarly, cell surface levels of SR-BI were not appreciably altered by expression of EGFP-WT-ctt, compared to those of cell lines expressing EGFP or EGFP-Δ509-ctt (Figure S5, A). Nevertheless we reasoned that EGFP-WT-ctt would interact with endogenous PDZK1 and, at high expression levels, out-compete endogenous SR-BI for PDZK1 binding. In agreement with this theory Huh-7 cells that stably expressed EGFP-WT-ctt were nearly 50% less susceptible to HCVcc infection than native EGFP-expressing control counterparts and nearly 80% less susceptible to HCVcc infection than Huh-7 cells stably expressing EGFP-Δ509-ctt (Figure 6D). Altogether these results indicate that expression of the cytoplasmic carboxy terminus of SR-BI, as a fusion to a soluble reporter, causes an inhibition of HCVcc infection that can be attributed to its ability to bind PDZK1.
Recent studies have indicated that the intracellular domains of SR-BI, particularly the cytoplasmic C-terminus, contribute to the activity of the receptor in HCV entry [12], [13], indicating that the binding of HCV to the extracellular domain of SR-BI, in itself, is not sufficient for the efficient involvement of SR-BI in the HCV entry process. Instead the cytoplasmic regions of the molecule may dictate dynamic changes in receptor localization and/or interactions with other HCV entry factors that facilitate HCV entry. In the context of studies of the involvement of SR-BI in HDL metabolism in mice, tissue-specific interaction of the cytoplasmic C-terminus of SR-BI with the adaptor protein PDZK1 has emerged as an important determinant of the receptor's effective enrichment at the hepatocyte plasma membrane and activity in HDL-CE transport (reviewed in [25]).
In the present study we have examined the association of human variants of SR-BI and PDZK1 and the importance of this interaction to HCV entry. We have confirmed the interaction of SR-BI with PDZK1 and provide evidence that the extreme C-terminus of SR-BI and the first PDZ domain of PDZK1 are critical sites of the interaction. Accordingly these proteins were found to overlap extensively in subcellular localization in the cytoplasm of transfected Huh-7 cells, with no discernable impact of mutation of the sites of interaction in both proteins or the major site of phosphorylation of PDZK1 on the colocalization of these proteins. While it has been reported that the protein kinase A (PKA)-dependent phosphorylation of the corresponding serine residue in rat PDZK1 is necessary for the ability of overexpressed PDZK1 to increase endogenous SR-BI protein levels in Fao hepatoma cells [33], we found no discernable impact of phosphodefective or phosphomimetic mutations at this site on the interaction or localization of these proteins, suggesting that phosphorylation at this site causes more subtle effects on SR-BI biology in human hepatocytes. It will be interesting to examine the relative importance of phosphorylation of PDZK1 to the involvement of SR-BI in the HCV entry process.
Effective knockdown of PDZK1 expression in Huh-7 cells was associated with a decreased susceptibility of these cells to infection with HCVcc and HCVpp, despite no discernable impact on total or surface levels of SR-BI protein. These effects contrast the more dramatic effects of PDZK1-knockout on the SR-BI protein content in mouse liver [28] and suggest that the plasma membrane-associated SR-BI in these cells is impaired in its ability to facilitate HCV entry. It is possible that loss of interaction with PDZK1 alters the localization of SR-BI within the plasma membrane. For example, it has been reported that SR-BI localizes to caveolae, a specialized subset of lipid raft domains of the plasma membrane, and that this localization may be important to the receptor's activity [41]. Consistent with this theory, removal of membrane raft cholesterol with ΜβCD inhibits HCV entry [42]. However, it has also been reported that neither SR-BI, nor other requisite entry factors CD81 and claudin-1, are strongly associated with plasma membrane lipid raft-enriched detergent-resistant membranes (DRMs) of Huh-7 cells [43]. Our analysis of the localization of SR-BI revealed that SR-BI staining, which was largely cytoplasmic, frequently colocalized CD81 at the cell surface and this colocalization was not appreciably altered by PDZK1 knockdown. Similarly, we did not observed any marked effects of PDZK1 knockdown on the staining pattern of occludin.
Based on what is known of the kinetics of antibody-mediated inhibition of HCV entry [15], [44], [45] and the differences in the localization of the four major HCV entry factors in human liver and hepatoma cells that support HCV entry [39], [46], [47], it has been suggested that SR-BI and CD81 are involved in post-binding steps of HCV entry, after which lateral migration of virus/receptor complexes towards tight junctions occurs and interaction with CLDN1 and occludin takes place.
Tight junctions are likely to be largely inaccessible to HCV under normal physiological conditions. Indeed, recent studies involving the use of the colorectal adenocarcinoma cell line Caco-2 that develops simple columnar polarity and HepG2 hepatoma cells that develop complex hepatic polarity have demonstrated that the formation of genuine tight junctions between cells as they develop polarity results in restricted access to viral receptors and inhibition of HCV entry [39], [48]. Interestingly, however, disruption of the integrity of tight junctions of polarized HepG2-CD81 cells with inflammatory cytokines does not perturb cell polarity or enhance HCV entry, whereas phorbol ester-induced activation of protein kinase C (PKC) results in both disruption of junctional integrity and cellular polarity and enhancement of HCV entry [39]. Similarly cAMP-dependent activation of protein kinase A (PKA) has also been shown to be important to HCV entry, with inhibition of PKA activity causing redistribution of CLDN1 to intracellular sites, disruption of CLDN1 association with CD81 and inhibition of HCV entry [49]. Intriguingly PKA activation is also associated with phosphorylation of PDZK1 [33], suggesting that the effects of PKA activity on HCV entry may also coincide with PDZK1-dependent changes in the involvement of SR-BI in the entry process. This possibility warrants further investigation. The involvement of HCV itself in the disruption of tight junctions and enhancement of HCV entry is also gaining attention. For example, HCV infection has been associated with disruption of the localization of claudin-1 and occludin to sites of cell-cell contact [50] and modulation of CD81 homodimerization and CD81 heterodimerization with claudin-1 [51]. Furthermore, a recent study has revealed that vascular endothelial growth factor (VEGF), which is upregulated in HCV-infected hepatocytes, promotes disruption of hepatocyte polarity and tight junctions between cells, thereby increasing their susceptibility and the susceptibility of nearby cells to HCV infection [52]. Further studies are required to determine whether disruption of cellular polarity and HCV infection have any impact upon the subcellular localization of SR-BI and PDZK1.
In argument against a rate-limiting role for PDZK1 in HCV entry the degree of inhibition of viral entry observed in PDZK1-knockdown Huh-7 cells (30-80%) did not closely reflect the degree of knockdown of PDZK1 protein levels (>90%). Although it could be argued that residual amounts of PDZK1 protein are sufficient to partially compensate for the knockdown of the majority of endogenous PDZK1 expression, we could not detect surface-associated PDZK1 in streptavidin precipitates of biotinylated plasma membrane proteins prepared from these cells indicating that residual amounts of PDZK1 were not preferentially associated with SR-BI at the cell-surface. Previously it was shown that a truncated mutant of SR-BI (SR-BIdel509), that does interact with PDZK1, was not readily detectable at the plasma membrane of hepatocytes prepared from SR-BIdel509 transgenic mice, indicating that in polarized hepatocytes interaction with PDZK1 is required for the correct localization of SR-BI [27]. Somewhat paradoxically, hepatic overexpression of full-length SR-BI in PDZK1-knockout mice can restore wildtype levels of total and surface SR-BI protein and reverse the effects of PDZK1-knockout on HDL metabolism [30], suggesting that alternative and likely inefficient PDZK1-independent means for the effective enrichment of full-length SR-BI at the hepatocyte plasma membrane exist. Further studies involving the use of polarized hepatoma cells or primary hepatocytes in which PDZK1 expression has been ablated are required to accurately predict the relative importance of PDZK1 to HCV infection in the human liver in vivo. Given recent major advances in the identification of factors that determine human hepatotropism of HCV entry [18] and adaptation of HCVcc to murine CD81 [53], studies of HCV entry into primary hepatocytes isolated from PDZK1-KO mice and derivatives may be possible in the foreseeable future.
It has recently been reported that PDZK1-knockdown does not have a substantial effect on HCVpp infection or HDL-mediated enhancement of HCVpp infection [12]. Although the reasons for the differences between the findings of that study and ours are not obvious, it is possible that near-complete knockdown of endogenous PDZK1 is required to observe inhibition of HCV entry. It is also possible that, between Huh-7 cell derivatives, variations in the relative levels of each of the HCV entry factors influences the relative dependence on that protein for efficient HCV entry, such that high levels of CD81, for example, could result in decreased dependence on SR-BI for initial capture of cell-free virus. Nevertheless, we employed additional overexpression and dominant-negative strategies to further investigate the importance of SR-BI/PDZK1 interaction to HCV entry. These studies revealed that overexpression of full-length PDZK1 significantly increased HCVcc infection levels in Huh-7 cells and restored baseline HCVcc infection levels in Huh-7 cells displaying knockdown of endogenous PDZK1. In contrast the SR-BI-interacting domain (PDZ1) of PDZK1 alone was not sufficient to restore normal HCVcc infection levels in PDZK1 knockdown cells indicating that regions outside this domain are required. Relevant to this, all four PDZ domains of PDZK1 are required to restore wildtype SR-BI protein levels and function in lipoprotein metabolism in PDZK1 knockout mice [31]. The reasons that PDZ1 overexpression did not have an inhibitory impact on HCVcc infection, as might be expected given the dominant-negative effects of hepatic PDZ1 (amino acids 1–116) expression on SR-BI expression and function in transgenic mice [32], are not clear. However the additional nuclear localization of our PDZ1 construct (amino acids 1–130) that was not observed in PDZ1-transgenic mice indicates the presence of substantive differences between the constructs that were not anticipated. Nevertheless, expression of the PDZK1-interacting domain of SR-BI, as a GFP fusion protein, resulted in reduced susceptibility of Huh-7 cells to HCVcc infection and a cytoplasmic organellar localization of the fusion protein. We therefore propose that pharmaceutical mimicry of this PDZK1-interacting domain of SR-BI may represent a future target of anti-HCV therapy which is relatively liver-specific.
Interestingly, like SR-BI, occludin and claudin-1 also interact with PDZ domain-containing cytoplasmic adaptor molecules (ZO-1 and ZO-2, respectively) via their cytoplasmic carboxy-termini (reviewed in [54]). While knockdown of ZO-1 expression has been associated with a decreased susceptibility of Huh-7 cells to HCV entry [16], this effect was not observed in another similar study [46]. Similarly, although C-terminally truncated claudin-1 can still support HCV entry, this mutation was associated with an approximately 3-fold reduction in HCV entry [15], suggesting that the PDZ-interacting domain of claudin-1 contributes to its efficient involvement in HCV entry. Interactions of SR-BI, occludin and claudin-1 with their respective PDZ domain-containing adaptor molecules at the inner leaflet of the plasma membrane may be particularly important to their respective localizations and activities in polarized hepatocytes in vivo.
Taken together our results indicate that the interaction of PDZK1 with SR-BI contributes to the efficient infection of hepatoma cells by HCVcc. Since the identification of what may represent the complete set of essential HCV entry factors, it remains to be determined when and how each of these proteins participates in viral entry and the secondary factors that influence dynamic changes in the localization and activity of each of the HCV entry factors. Future studies of this nature will improve our understanding of HCV infection and tropism and may reveal new targets of antiviral therapy.
A Huh-7 cell line that is permissive to HCV infection and replication was kindly provided by Eric Gowans (Children's Health Research Institute, Adelaide). Huh-7 cells harbouring the genotype 1b NNeo/C-5B(RG) genomic HCV replicon [55] were kindly provided by Stanley Lemon (University of Texas Medical Branch, Galveston). The HepG2(N6) clone was kindly provided by David Anderson (Macfarlane Burnet Institute, Melbourne). 293T cells were obtained from the American Type Culture Collection. With the exception of HepG2(N6) cells and their derivatives which were culture as previously described [38], all cells were maintained in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 2 mM L-glutamine, non-essential amino acids, penicillin (100 U/ml), streptomycin (100 µg/ml) and 10% foetal bovine serum (FBS). Where appropriate, selective antibiotics G418 (800 µg/ml), puromycin (3 µg/ml) and/or blasticidin (3 µg/ml) were added to culture media. All cells were cultured at 37°C in 5% CO2.
Rabbit polyclonal antibodies against PDZK1 (Zymed Laboratories), SR-BI (Novus Biologicals), C-myc (Santa Cruz) and FLAG (Sigma) were purchased. Mouse monoclonal antibodies against the following antigens were purchased: FLAG (M2; Sigma), β-actin (clone AC-15; Sigma), C-myc (clone 9E10; Roche Applied Science), phosphoserine (clone 7F12; Invitrogen), CD81 (clone JS81; Pharmingen), and occludin (clone OC-3F10; Zymed Laboratories). Biotinylated goat anti-GFP antibody was purchased from Rockland Immunochemicals. Human anti-SR-BI monoclonal antibody C-167 [56] was kindly provided by Alessandra Vitelli (Istituto di Ricerche di Biologia Molecolare P. Angeletti, Rome). Mouse monoclonal anti-NS5A antibody 9E10 was kindly provided by Charles Rice (The Rockefeller University, New York). Alexa Fluor-488 and -555 conjugated secondary antibodies were purchased from Invitrogen. HRP-conjugated anti-mouse IgG and anti-rabbit IgG antibodies and streptavidin-conjugated HRP were from Pierce.
The following plasmids were generous gifts: pJFH-1 [7] (Takaji Wakita; National Institute of Infectious Diseases, Shinjuku-ku), pJc1/GFP and pJc1/RFP [36] (Ralf Bartenschlager; University of Heidelberg, Heidelberg), pER-RFP and pGolgi-RFP [57] (Erik Snapp; Albert Einstein College of Medicine, New York) and pLenti6/V5-D-TOPO-tdTomato (Yuka Harata-Lee; University of Adelaide, Adelaide). Plasmid pJc1/Myc was generated by in-frame replacement (XbaI/RsrII) of the GFP coding region of pJc1/GFP with an oligonucleotide adaptor duplex encoding the C-myc epitope (EQKLISEEDL). PDZK1 cDNA (NM_002614) expression constructs with C-terminal FLAG (DYKDDDDK) epitope tags, PDZK1-FLAG and PDZ1-FLAG (encoding aa 1–130), were cloned into pcDNA3 (HindIII/XbaI) (Invitrogen) and pLenti6/V5-D-TOPO (BamHI/XhoI) (Invitrogen). Where indicated mutations encoding the amino acid substitutions S505A or S505D (referred to as S509A and S509D for consistency with the original description of Ser-509 phosphorylation in rat PDZK1 [33]), and the silent mutation Stca261Sagc were introduced by QuickChange site-directed mutagenesis (Stratagene). Full-length and C-terminally truncated (Δ509) SR-BI cDNA (NM_005505) expression constructs, with or without N-terminal Myc or FLAG epitope tags, were cloned into pcDNA6/V5-HisB (HindIII/XbaI) (Invitrogen). EGFP-WT-ctt and EGFP-Δ509-ctt expression plasmids were generated by cloning the cDNA region encoding the cytoplasmic C-terminus of SR-BI (aa 479–509) with a Myc epitope tag at the N-terminus into pcDNA3 (XhoI/XbaI). The EGFP coding sequence, with the stop codon removed, was then cloned in-frame into upstream restriction sites (EcoRI/XhoI). To generate pcDNA3-EGFP the EGFP coding sequence, including the stop codon, was cloned into pcDNA3 (EcoRI/XhoI). Plasmid pcDNA3-CD81myc was generated by cloning CD81 cDNA (NM 004356), with a C-terminal C-myc epitope tag, into pcDNA3 (HindIII/BamHI). All constructs were confirmed by automated DNA sequencing. Exact cloning details are available upon request. To generate HepG2(N6) cell lines expressing CD81, HepG2(N6) cells were transfected with pcDNA3-CD81myc using FuGene6 (Roche) as per manufacturer's instructions and, 48 h post-transfection, G418 (800 µg/ml) selection was applied. G418-resistant clones were later individually expanded and screened for CD81 expression by flow cytometry (not shown). One clone displaying uniform, high-level expression of CD81 was chosen for further analysis. Huh-7 cells stably expressing EGFP, EGFP-WT-ctt or EGFP-Δ509-ctt were generated by transfection with the corresponding pcDNA3 expression plasmid, selection with G418 and enrichment of GFP-positive cells using a FACSAria flow cytometer (BD Biosciences).
Cell culture propagated HCV particles (HCVcc; JFH-1, Jc1/GFP, Jc1/RFP and Jc1/Myc) were generated as described previously [8]. Briefly, in vitro transcribed HCV RNA was generated from XbaI-linearized plasmid pJFH-1 or MluI-linearized plasmids pJc1/GFP, pJc1/RFP or pJc1/Myc using a MEGAscript T7 kit (Ambion). Huh-7 cells were electroporated with 10 µg of RNA and at 5 d post-transfection (3 d post-transfection for Jc1/GFP and Jc1/RFP preparations) virus containing supernatants were collected, filtered (0.45 µm) and used undiluted to infect naïve cells for 3 hours. Cell monolayers were then washed twice with PBS and returned to culture for 48–72 h, as indicated, prior to analysis of HCV infection. HCVcc infectious titres (focus forming units [FFU]/ml) were determined as described previously [8].
Pseudoviruses encoding Luciferase were generated by co-transfection of 293T cells with the lentiviral packaging plasmids psPAX2 (Addgene plasmid 12260), pRSV-Rev (Addgene plasmid 12253), pLenti6/V5-D-TOPO-Luciferase and either VSV-G envelope expression plasmid pMD2.G (Addgene plasmid 12259) for generation of VSVpp, the expression plasmid pE1E2H77c [4] for generation of HCVpp or empty plasmid pcDNA3 for generation of Env-pp. Supernatants were harvested at 48 and 72 h post-transfection, pooled and filtered (0.45 µm). Virus-containing cell culture supernatants were then diluted in normal culture media (1∶500 for VSVpp, 1∶2 for HCVpp and Env-pp) and incubated with target cells, seeded at 2×104 cells/cm2 the day before (unless otherwise specified), for 7 h before washing with PBS and return to culture. Pseudoparticle infections were performed in the presence of 4 µg/ml polybrene. At 72 h post-infection luciferase activity was measured using a Luciferase Assay System (Promega) and a GloMax 96 microplate luminometer (Promega). Specific HCVpp and VSVpp infectivity levels were determined by subtraction of the luciferase signals associated with the use of non-enveloped pseudoparticles (Env-pp).
Lentiviral pLKO.1 shRNA constructs specific for human PDZK1 were purchased from Open Biosystems (target set for NM_002614). PDZK1-specific shRNA clones (PDZK1 shRNA #4, TRCN0000059671; PDZK1 shRNA#5, TRCN0000059672) and a control non-target shRNA clone (Sigma; SHC002) were packaged into lentiviral vectors by co-transfection of 293T cells along with psPAX2 and pMD2.G. Lentiviral overexpression vectors were generated by co-transfection of 293T cells with psPAX2, pMD2.G, pRSV-Rev and either pLenti6/V5-D-TOPO-PDZK1-FLAG or pLenti6/V5-D-TOPO-PDZ1-FLAG. Lentiviral supernatants were collected at 48 and 72 h post-transfection, pooled, filtered (0.45 µm) and stored at −80°C. Target cells were infected for 4 h with lentiviral supernatants diluted 1∶5 in normal culture medium containing 8 µg/ml polybrene (Sigma). Antibiotic selection with puromycin ([3 µg/ml] for pLKO.1 constructs) or blasticidin ([3 µg/ml] for pLenti6/V5-D-TOPO constructs) was applied 48 h later to generate antibiotic-resistant polyclonal cell lines.
Huh-7 cells were grown on 0.2% gelatin-coated coverslips overnight, transfected where applicable and fixed the following day in ice-cold acetone:methanol (FLAG, Myc, NS5A) or ice-cold 5% buffered formalin (GFP/RFP epifluorescence, SR-BI, CD81, occludin, LAMP1). Where appropriate, cells were permeabilized with 0.05% saponin in PBS, prior to blocking in 5% BSA in PBS and incubation with primary antibody diluted in 1% BSA in PBS for 1 h at room temperature. After washing twice with PBS, Alexa Fluor-488 and/or Alexa Fluor-555 conjugated secondary antibodies were applied for 1 h at 4°C. Cell monolayers were then washed twice with PBS and coverslips were mounted using ProLong Gold antifade reagent (Invitrogen). Where combinations of antibodies were used, control experiments using isotype-matched irrelevant antibodies or secondary antibody alone (SR-BI surface labelling) were performed to ensure minimal non-specific binding of antibodies. Cells were viewed using an Olympus IX70 inverted microscope linked to a Bio-Rad Radiance 2100 confocal microscope. Samples were visualized using a ×60/1.4NA water immersion lens (2–3× zoom) and images were acquired sequentially for each fluorophore and processed using Adobe Photoshop 6.0 software (Adobe Systems Inc).
Extraction of total cellular RNA, first-strand cDNA synthesis and real-time RT-PCR was performed as described [58].
Immunoprecipitation of FLAG-tagged proteins was performed as follows. Near-confluent cells in 35 mm dishes were washed with PBS before lysis in 0.5 ml of NP-40 lysis buffer (1% NP-40, 150 mM NaCl, 50 mM Tris [pH 8.0]) containing protease inhibitors (Sigma) on ice for 30 min. Samples were then passed through a 26 gauge needle 10 times and cleared of nuclear debris by centrifugation (10,000×g, 10 min). All incubations were performed at 4°C with end-over-end rotation. Lysates were pre-cleared by incubation with 20 µl of protein A/G PLUS agarose (Santa Cruz Biotechnology) for 1 h. Samples were then centrifuged (1000×g, 5 min, 4°C) and supernatants were collected and incubated with 1 µg of anti-FLAG mAb overnight. 20 µl of protein A/G PLUS agarose was then added to each sample prior to incubation for 1 h. Beads were then pelleted by centrifugation (1000×g, 5 min, 4°C) and washed 5 times with 1 ml of NP-40 lysis buffer before resuspension in 40 µl of SDS-PAGE loading buffer and Western analysis.
Biotinylation of plasma membrane proteins with membrane impermeable sulfo-NHS-biotin (Pierce) and streptavidin-precipitation of biotinylated proteins was performed as described previously [59].
Cells were lysed in RIPA buffer (150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, 50 mM Tris, pH 8.0) containing protease inhibitors (Sigma) on ice for 30 min, homogenized and cleared by centrifugation (10,000×g, 10 min). For each sample approximately 30 µg of protein was separated by 12% SDS-PAGE and transferred to HyBond-ECL nitrocellulose membrane (Amersham Biosciences) for immunoblotting using anti-FLAG, anti-Myc, anti-GFP, anti-PDZK1, anti-SR-BI or anti-β actin. HRP-conjugated secondary antibodies or HRP-conjugated streptavidin were detected by chemiluminescence using SuperSignal West Femto (Pierce). Where indicated densitometric analysis of bands was performed using NIH Image software.
Cells were harvested by trypsinization, washed (PBS/1%FBS) and incubated at 4°C for 1 h with appropriately diluted anti-CD81, anti-SR-BI (C-167), irrelevant isotype-matched control MAb (CD81 labelling) or no primary antibody (SR-BI labelling). Following washing, cells were incubated for 1 h at 4°C with Alexa Fluor-555 conjugated goat anti-mouse IgG or Alexa Fluor-555 conjugated goat anti-human IgG. Cells were then washed, fixed (1% formalin in PBS containing 111 mM D-glucose and 10 mM NaN3) and analysed using a FACS-Canto flow cytometer (BD Biosciences).
Data are expressed as means + the standard error of the mean (SEM). Statistical analysis was performed by Student's t test, with P<0.05 considered to be statistically significant.
PDZK1, NM_002614; SR-BI, NM_005505; CD81, NM_004356; Claudin-1, NM_021101; Occludin, NM_002538; HCV (JFH-1), AB047639; HCV (HC-J6), HPCPOLP; HCV (H77), NC_004102.
|
10.1371/journal.ppat.1004999 | THY-1 Cell Surface Antigen (CD90) Has an Important Role in the Initial Stage of Human Cytomegalovirus Infection | Human cytomegalovirus (HCMV) infects about 50% of the US population, is the leading infectious cause of birth defects, and is considered the most important infectious agent in transplant recipients. The virus infects many cell types in vivo and in vitro. While previous studies have identified several cellular proteins that may function at early steps of infection in a cell type dependent manner, the mechanism of virus entry is still poorly understood. Using a computational biology approach, correlating gene expression with virus infectivity in 54 cell lines, we identified THY-1 as a putative host determinant for HCMV infection in these cells. With a series of loss-of-function, gain-of-function and protein-protein interaction analyses, we found that THY-1 mediates HCMV infection at the entry step and is important for infection that occurs at a low m.o.i. THY-1 antibody that bound to the cell surface blocked HCMV during the initial 60 minutes of infection in a dose-dependent manner. Down-regulation of THY-1 with siRNA impaired infectivity occurred during the initial 60 minutes of inoculation. Both THY-1 antibody and siRNA inhibited HCMV-induced activation of the PI3-K/Akt pathway required for entry. Soluble THY-1 protein blocked HCMV infection during, but not after, virus internalization. Expression of exogenous THY-1 enhanced entry in cells expressing low levels of the protein. THY-1 interacted with HCMV gB and gH and may form a complex important for entry. However, since gB and gH have previously been shown to interact, it is uncertain if THY-1 directly binds to both of these proteins. Prior observations that THY-1 (a) interacts with αVβ3 integrin and recruits paxillin (implicated in HCMV entry), (b) regulates leukocyte extravasation (critical for HCMV viremia), and (c) is expressed on many cells targeted for HCMV infection including epithelial and endothelial cells, fibroblast, and CD34+/CD38- stem cells, all support a role for THY-1 as an HCMV entry mediator in a cell type dependent manner. THY-1 may function through a complex setting, that would include viral gB and gH, and other cellular factors, thus links virus entry with signaling in host cells that ultimately leads to virus infection.
| Human cytomegalovirus (HCMV) is an important human pathogen that infects about half the US population and is a major cause of birth defects and morbidity in transplant recipients. Despite extensive research, much is still unknown regarding how the virus enters cells. We identified THY-1, a protein on the surface of many different cell types susceptible to CMV infection, as having an important role for facilitating virus infection. We found that antibody to THY-1 or soluble THY-1 protein blocked HCMV infection in multiple cell types, suggesting that THY-1 might serve as a potential therapeutic target to reduce infection. Expression of exogenous THY-1 increased susceptibility of cells to HCMV infection. We showed that THY-1 has an important role in a host signaling pathway that is initiated when HCMV infects cells. Furthermore, we found that THY-1 interacted with HCMV glycoproteins that initiate entry of virus into the cell. THY-1 is known to interact with several host cell proteins important for infection and is expressed on numerous types of cells that can be infected by HCMV. Thus, we have identified THY-1 as a molecule that has an important role in the initial stage of HCMV infection.
| Human cytomegalovirus (HCMV) infects about 50% of the US population and is the leading infectious cause of birth defects and the most important infectious agent in transplant recipients. In vivo, HCMV predominantly infects epithelial, endothelial, fibroblast, smooth muscle, and mononuclear cells including myeloid progenitors and dendritic cells [1,2]. Primary infection typically begins with virus replication in mucosal epithelium followed by leukocyte-associated viremia. Among more than 50 putative glycoproteins encoded by HCMV, gH/gL and gB are conserved in the herpesvirus family, and are required for HCMV entry into cells [3]. gH and gL interact with UL128-UL131 proteins to form a pentameric complex or with gO to form a trimer, that are important for infection of different cell types [4–6]. gB has been reported to bind to gH/gL, and functions as a fusogen [7,8]. In addition, gB binds to heparan sulfate proteoglycans [9–11].
HCMV initiates infection by attachment to cell surface heparan sulfate proteoglycans [12,13] followed by engagement of cellular receptors or entry mediators. Previous studies have identified several cellular proteins that may function at early steps of infection, including platelet-derived growth factor receptor-α (PDGFR-α) [14,15], epidermal growth factor receptor (EGFR) [16], DC-SIGN [17], αVβ3 and β1 integrins [18,19], and paxillin [20]. HCMV, like many other viruses, utilizes host molecules to facilitate entry in a cell type dependent manner. gB and gH interact with these cellular molecules [14,16–18,21,22]; however, it is not clear whether the interactions are direct or indirect through protein complexes that may include various viral and cellular components. Virus entry is not only limited to virion internalization and cell signaling is an integral part of the entry process [23]. Previous work has shown that HCMV induced activation of the Akt signaling pathway is required at an early step in virus entry [22]. HCMV utilizes PDGFR- α to facilitate entry and simultaneously induces phosphorylation of PDGFR- α when the virus infects fibroblasts, endothelial and epithelial cells. The virus activates EGFR when it infects monocytes, and employs integrins and paxillin at the beginning of infection. The activation of either PDGFR- α or EGFR in turn leads to activation of downstream cellular phosphatidylinositol 3-kinase (PI3K), Src kinase and focal adhesion kinase (FAK) signaling pathways, and induces cytoskeletal rearrangements to create an intracellular environment to facilitate infection [14,20].
HCMV infects a broad spectrum of human cells ranging from epithelial and endothelial cells to hepatocytes and neuronal cells. This may reflect the capability of the virus to utilize multiple cellular molecules to gain entry depending on the type of cell. The observation that cells expressing neither PDGFR- α nor EGFR are still permissive for HCMV infection implies that the virus exploits additional host factors at an early step of infection [3,24]. In an attempt to identify other cellular proteins important for infection, we utilized 54 human cell lines from the NCI-60 panel of diverse tissue origins whose gene expression profiles have been extensively analyzed across multiple platforms [25–27]. A previous study showed that transcript-protein correlation in these cell lines is highly statistically significant [28]. In conjunction of bioinformatics analysis, this panel of cell lines has been a valuable screening tool for identifying host factors important for viral infection [29–33]. We investigated the susceptibility of these cell lines for HCMV and correlated infectivity with gene expression profiles for each of the cell lines using bioinformatics analysis. This approach allowed us to evaluate the contribution of individual host molecules to infection in the context of overall gene expression in the cells. We focused on membrane associated proteins since they are likely to be involved in the very early steps of virus infection. Using a series of loss-of-function, gain-of-function and ligand interaction analysis, and additional non-transformed cells, the biological function of one candidate protein was further validated. Here, we report that THY-1 has an important role in the early stages of HCMV infection in a diverse group of cell lines.
Prior studies to identify entry mediators for HCMV have been limited by the types of viruses and cell lines used. High passage strains of HCMV, deleted for the UL128-131 region, which are restricted to efficient growth in fibroblasts, have been predominately used to identify HCMV entry mediators [14,16,22]. In addition, previous studies defining HCMV entry used relatively few cell lines, and most studies focused on fibroblasts. Since HCMV utilizes different host molecules to infect specific types of target cells (mainly endothelial, epithelial, and mononuclear cells in vivo), these more traditional approaches with one or only a few cell lines have limitations. To address the issue, we utilized a panel of 54 adherent cell lines of diverse origins from the NCI-60 panel whose molecular profiles have been extensively characterized at the DNA, RNA and protein levels, and integrated with each other by integromic analyses [26,34] (S1 Table). We infected the cells with both fibroblast (Towne-GFP) (a gift from Dr. H. Zhu, UMDNJ-New Jersey Medical School) and epithelial/endothelial tropic (BADrUl131-GFP and TB40E-GFP) HCMV which express GFP (gift from Dr. T. Shenk, Princeton University) [4,35–37]. Two or three days post infection, susceptibility to HCMV was determined based on GFP positivity of the cells. For bioinformatics analyses, infection of each cell line with each virus was performed in at least three independent experiments and each time in triplicate wells. Infectivity was then determined by FACS analysis of GFP positive cells and the mean infectivity score was calculated by normalization using epithelial (ARPE-19) cells for epithelial/endothelial tropic virus and fibroblasts (MRC-5 cells) for fibroblast tropic virus (S2 Table). Correlations between HCMV infectivity and expression of each cellular gene were calculated using the COMPARE algorithm [38] and further detailed using MAPP software [30]. COMPARE utilizes gene expression profiling as determined by microarray analysis across multiple microarray platforms to identify genes that correlate (based on the Pearson Correlation Coefficient) with the experimentally determined HCMV infection profile [30,38]. The mean infectivity score and the expression level of each gene were computed and the Pearson Correlation Coefficient was determined.
The highest rated membrane associated protein whose expression correlated positively with virus susceptibility was PDGFR-α, which has been shown to function in HCMV entry [14,15]. Transfection of MRC-5 cells with PDGFR-α specific siRNAs reduced HCMV infection (S1A Fig). THY-1 was implicated as the next highest scoring membrane associated protein whose expression correlated positively with HCMV infectivity. Infectivity of both epithelial/endothelial and fibroblast tropic HCMV strains showed a positive correlation with THY-1 expression at a level similar to or higher than that of PDGFR-α (Fig 1). The correlation of THY-1 expression was statistically significant for both Towne-GFP (P = 0.0002, Pearson Correlation Coefficient 0.46) and TB40E-GFP HCMV (P = 0.0004, Pearson Correlation Coefficient 0.44). Likewise, expression of PDGFR-α correlated with infection for Towne-GFP (p<0.00001, Pearson Correlation Coefficient 0.53) and TB40E-GFP HCMV (p = 0.016, Pearson Correlation Coefficient 0.29). Similar correlations for THY-1 and PDGFR-α expression with infectivity were also observed for epithelial/endothelial tropic strain BADrUl131-GFP HCMV.
To determine whether THY-1 is important for HCMV infection, we performed a series of loss-of-function experiments. First, we determined if soluble THY-1 (a.a. 20–130) can block HCMV infection. Wild-type THY-1 is initially synthesized as a 161 amino acid peptide. Upon maturation, the signal peptide (a.a.1-19) is cleaved and the C-terminal a.a. 132–162 is replaced with a GPI anchor. A soluble form of THY-1 (a.a. 20–130) exists in vivo and the recombinant form of THY-1 retains its biological function in binding integrins [39,40]. HCMV or control virus (HSV-2-GFP or adenovirus-GFP) was premixed with soluble THY-1-His protein or a control His protein (soluble varicella-zoster virus gE-His) at room temperature for 10 min, added to HS-578T cells for virus binding on ice for 60 min. Internalization was initiated by raising the temperature to 37°C for 60 min, and then non-absorbed virus was inactivated at low pH, and infectivity was quantified using GFP 3 days later. Compared with the control protein at each dose, soluble THY-1 protein reduced HCMV infectivity in a dose-dependent manner (Figs 2A and S3) in adenocarcinoma cells, and inhibited infection in MRC-5 fibroblasts (Figs 2B and S4 top). In contrast, it did not reduce HSV-2 infectivity (Figs 2C and S4 bottom) or adenovirus infectivity (Figs 2D and S4 Bottom). Soluble THY-1 protein was required during the initial viral entry step to block HCMV infectivity, since addition of the protein after virus binding and internalization did not inhibit infectivity (Fig 2A, last bar). In natural hosts HCMV infection likely occurs at a relatively low m.o.i. A review of studies of virus shedding from saliva of infants, children, and adults, often the source of transmitted virus, showed that the titer of virus in saliva ranged from 103 to 2 x 104 pfu/ml [41]. Therefore we infected the cells with titers ranging from 4 x 104 (HS-578T) to 1 x 105 pfu/ml (MRC-5), which corresponds to a relatively low m.o.i. (0.05 to 1) to try to replicate what may occur during natural infection. Furthermore, we used acid inactivation to limit the infection within the first 60 min to focus on the initial stages of virus infection and the most efficient pathways for viral entry (Fig 2). During the first 60 minutes after infection (m.o.i. 0.05–1 with acid inactivation), about 2–10% of the cells were infected, which corresponds to about 20–35% of the cells if the same infection is allowed to continue for a prolonged time, i.e. without acid inactivation (S5 Fig). Soluble THY-1 protein blocked over 90% of the infection that occurred within the first 60 min (m.o.i. 0.05–1) at a dose of 0.5 μg/ml (Fig 2A). In contrast, with a high m.o.i (4, based on titration in MRC-5 cells) 10-fold more soluble protein was required to block >90% of the infectivity (during entry over 60 min with acid inactivation), and soluble THY-1 blocked infection less efficiently for the virus that enters with slower kinetics (75% reduction in infectivity without acid inactivation, S6 Fig).
Next, we examined whether specific antibody 5E10 binds to cell surface THY-1 protein. NCI-60 cell lines SNB-19 (glioblastoma) and HS-578T (adenocarcinoma), as well as primary human diploid (MRC-5) fibroblasts all express THY-1 mRNA [34] (Fig 3A), and THY-1 protein was detected on the surface of these cells (Figs 3B and S1B). Both HS-578T and SNB-19 cells support productive HCMV infection and produce progeny virus (S2 Fig), although HCMV cell-to-cell spread in SNB-19 cells is limited, especially with TB40E-GFP HCMV. To ascertain whether THY-1 specific antibody blocks HCMV infection, THY-1 or isotype control antibody was allowed to bind to the surface of HS-578T cells on ice for 60 min, the antibody mixture was removed from the cells, and HCMV was added on ice for 60 min to synchronize virus binding. To focus on the early steps of virus entry, the temperature was raised to 37°C for 60 min to allow virus entry, followed by low pH treatment to inactivate any virions that still remained on the cell surface or in the medium. After washing, the cells were then cultured for 6 hr before RNA extraction to quantify combined HCMV UL123 (encodes IE1) and UL55 (encodes gB) RNA expression by RT-qPCR [42] or for 3 days to measure infectivity by FACS for GFP. Although UL55 is a late gene, UL55 transcripts start to appear at 4 hrs post-infection, and expression is not strictly dependent on new viral DNA synthesis [43,44]. In 4 independent experiments, quantitative RT-PCR showed that THY-1 specific antibody blocked expression of HCMV UL123 and UL55 genes, compared with isotype control antibody (Fig 3C, P = 0.0002 for 4 independent experiments). Similar blocking result with THY-1 antibody was also seen when infectivity was assayed at 3 days post-infection by virus-encoded GFP (Fig 3D, P = 0.0004, 3 independent experiments). THY-1 specific antibody, but not isotype control, blocked HCMV infectivity in a dose-dependent manner (Figs 3D and S7). THY-1 antibody also blocked HCMV infection in primary MRC-5 cells (Fig 3E).
To confirm the loss-of-function findings observed with THY-1 specific antibody, we used THY-1 specific siRNAs to knockdown THY-1 expression in permissive cells, and analyzed the effect on HCMV infection. Nucleofection of cells with THY-1 specific siRNAs reduced THY-1 expression by over 90% compared with control siRNAs at the time of infection both at the mRNA (Fig 4A) and protein level (S10A Fig and see section on THY-1 and Akt activation below). HCMV infectivity was reduced by 30–50% at 3 days post-infection based on FACS analysis for GFP (Fig 4B) (P value <0.0001, 12 independent experiments). However, THY-1 siRNAs knocked down cell surface THY-1 protein (S10B Fig) less effectively than total THY-1 protein (S10A Fig). This might be due to increased stability of surface THY-1 protein when it is anchored into lipid rafts, and could contribute to the lower level of inhibition of HCMV infection with THY-1 siRNAs than with antibody or soluble protein (see above). The impairment of HCMV infectivity following knockdown of THY-1 was observed in glioblastoma (SNB-19), adenocarcinoma (HS-578T) and MRC-5 cells infected with either epithelial/endothelial or fibroblast tropic HCMV. Since the infection protocol allowed only 60 min for virus entry before virus was inactivated by low pH, the reduction of infectivity occurred during initiation of HCMV infection.
In contrast with SNB-19 and HS-578T cells which support HCMV infection and express THY-1 on their surface (used above in loss-of-function experiments), SF-539 (gliosarcoma) cells express negligible levels of THY-1 mRNA or THY-1 protein on the cell surface (Figs 3A and S1B), and are refractory to HCMV Towne infection. Molecular profiling of NCI-60 cells showed that SF-539 cells express comparable levels of PDGFR-α, EGFR, αVβ3 and β1 integrins as SNB-19 and HS-578T cells. Therefore, we used SF-539 cells for gain-of-function studies. pCMV-THY-1 or empty vector was transfected into SF-539 cells by nucleofection and 48 hr later the cells were incubated with Towne-GFP for 1 hr on ice, then at 37°C for 1 hr, followed by low pH to inactivate virus that had not entered the cells. RNA was extracted from the cells at the time of infection to monitor THY-1 expression and at 6 hrs post-infection to detect HCMV UL123 and UL55 expression. Quantitative RT-PCR showed that SF-539 cells transfected with control vector expressed very low levels of THY-1 mRNA, while cells transfected with pCMV-THY-1 expressed high levels of THY-1 mRNA (Fig 4C). Expression of THY-1 from the pCMV-THY-1 plasmid enhanced HCMV infectivity of the cells (Fig 4D, P <0.0001, 7 independent experiments). Since the infection was restricted to the initial 60 min of viral inoculation, we conclude that exogenous expression of THY-1 enhances the initial stage of HCMV infection of cells.
gB and gH/gL are essential for HCMV infection [3]. HCMV gB has a furin cleavage site that results in covalently bound N-terminal and C terminal fragments of about 55 kD each. gB has been reported to bind to gH and may form glycoprotein complexes with other components, including gO or UL128-131 [9,10,45]. We postulated that since THY-1 is important for HCMV infection, it might interact with one or more of these glycoproteins, either directly or as part of a complex. We incubatedanti-THY-1 or isotype control antibody with HCMV-infected and uninfected cell lysates, and separated the immune complexes by gel electrophoresis. Several protein bands were found in lysates from HCMV Towne infected MRC-5 cells immunoprecipitated with antibody to THY-1, but not in lysates immunoprecipitated with isotype control antibody or in uninfected cells. Mass spectrometry of these unique bands identified gB and gH with a Mascot score of 1141 and 281 (a score of 45 represents the significance threshold for individual peptide matches P <0.05), with multiple peptide sequence coverage for both glycoproteins. In contrast, gM and gO were each identified by only a single peptide (S3 Table).
Since co-immunoprecipitation followed by Western blotting was inefficient for detecting proteins that interact with gB in infected cells, we constructed protein columns by binding THY-1-His protein, or control VZV gE-His protein to Talon beads, added lysates from HCMV-infected cells to the columns, eluted proteins bound to the columns, and immunoblotted the proteins with antibody to HCMV ICP8 or gB. Two different cell lysis buffers were used, PBS with 0.1% NP-40 [30] and 25 mM Tris, 15 mM NaCl and 0.1% NP-40 [46]. HCMV gB was detected in the infected cell lysate and in eluates from THY-1 protein columns, but not the control VZV gE protein column (Fig 5A). Interestingly, THY-1 complexed with full length gB (160 kD), as well as its proteolytic cleavage products of 55 kD [47]. Purified THY-1 protein pulled down more 55 kD gB than full length gB. A previous study has shown the cleaved form of gB was more abundant than full length gB in infected cell lysate and in purified virions [48]. In contrast, the 135 kD HCMV ICP8 was detected in infected cell lysate, but not in eluates from THY-1 or control VZV gE protein columns (Fig 5B). Similarly, gH was co-precipitated from infected cell lysate by purified THY-1 protein (Fig 5C). These results suggest that THY-1 may form a complex with HCMV gB and gH in infected cells. Since gB and gH have been shown to form a complex, it is possible that THY-1 interacts directly with gB and that the interaction of THY-1 with gH is indirect and solely due to gH interacting with gB. Alternatively, THY-1 has been shown to bind to integrins, and gB and gH from several herpesviruses interact with integrins; thus, the interaction between THY-1 and gB and gH may be indirect and mediated through integrins.
To further study the possibility of an interaction between THY-1 and the HCMV gB and gH glycoproteins [10], MRC-5 cells were infected with HCMV AD169 (which does not express GFP) and live cell staining was performed with goat anti-THY-1 antibody and mouse monoclonal anti-gB, anti-gH, or isotype control antibody followed by anti-goat and anti-mouse fluorescent antibodies and confocal microscopy. THY-1 colocalized with gB (Pearson Correlation Coefficient 0.88 where 1.0 is 100% colocalization [49] (Fig 6A, row 1) and gH (Pearson Correlation Coefficient 0.84, Fig 6A row 2). Incubation of MRC-5 cells with secondary antibody alone did not give background staining, goat anti-THY-1 did not cross react with secondary anti-mouse fluorescent antibody, and mouse anti-glycoprotein antibodies did not cross react with secondary anti-goat fluorescent antibody (Fig 6A, row 3). In HCMV- infected adenocarcinoma HS-578T cells, gB also colocalized with THY-1 (Figs 6B and S11). As a control, gB did not colocalize with cell surface protein ZO-1 (Fig 6B). Interestingly, confocal microscopy with 3-D reconstruction of the cell surface showed that gB appeared to bind predominantly on top of THY-1 molecules on the plasma membrane (Fig 7). Although gB is conserved among human herpesviruses, HCMV gB (AD169 strain) and VZV gB (Dumas strain) share only 20% amino acid identity and 31% similarity. As an additional control, we co-transfected THY-1 with either HCMV gB or VZV gB, and performed confocal microscopy. HCMV gB colocalized with THY-1 at levels similar to that in infected cells, but VZV gB did not colocalize with THY-1 (S11 Fig).
These results suggest that THY-1 may form a complex with HCMV gB and gH in infected cells. Since glycoproteins gB and gH have been shown to form a complex, it is possibly that THY-1 interacts directly with gB and that the interaction of THY-1 with gH is indirect and solely due to gH interacting with gB. Alternatively, THY-1 has been shown to bind to integrins, and gB and gH from several herpesviruses interact with integrins, thus, the interaction between THY-1 and gB and gH may be indirect and mediated through integrins,
Previous studies have shown THY-1 modulates the phosphatidylinositol 3-kinase (PI3K) signaling pathway [50]. Activation of the PI3K pathway is required for HCMV infection at the entry step [14,20,22]. Therefore, we analyzed the effect of THY-1 on the ability of HCMV to phosphorylate Akt, a downstream molecule in the PI3K pathway. Knock-down of THY-1 expression with specific siRNAs blocked HCMV-induced phosphorylation of Akt at 15 min post-infection and reduced HCMV infectivity within the first 60 min of infection (Figs 8A and 8B and S12) compared with control siRNAs (p = 0.01, 6 independent experiments). These data suggest that HCMV engagement of THY-1 during the initial 15 min of infection contributes to HCMV signaling through the PI3K/Akt pathway.
We then tested whether THY-1 antibody could block HCMV mediated Akt activation during entry. Binding of THY-1 antibody, but not isotype control antibody, to the cell surface inhibited Akt activation within 45 min after the incubation temperature was raised to 37°C to allow for HCMV internalization (S9 Fig).
In spite of progress in the field of virus entry, our understanding of the interaction of viral and cellular proteins required for initiation of HCMV infection is still unclear. This may reflect the large number of HCMV glycoproteins and the ability of the virus to infect a wide variety of cell types. Previous studies of early events in HCMV infection were largely limited to a few cell lines. This imposed limitations for identifying host molecules that are important for infection, since virus entry is cell-type dependent. To address this issue, we studied HCMV infectivity in 54 cell lines with diverse genetic backgrounds. The extensive molecular profiling of each of these cell lines along with bioinformatics analysis allowed us to take an unbiased approach to study virus infection instead of screening for single molecules in isolation. The identification of THY-1 as a putative host determinant for HCMV infection in a large set of 54 cell lines, and the subsequent validation by a series of loss-of-function, gain-of-function, and glycoprotein interaction experiments in both malignant and primary cells strongly suggests that THY-1 has an important role in the initial stage of virus infection. THY-1 is expressed in many cell types both in vivo and in vitro, including epithelial and endothelial cells, smooth muscle cells, placenta, neurons, hepatocytes, and hematopoietic stem cells, the same cells that are susceptible to HCMV infection. Therefore, THY-1 likely facilitates HCMV entry in many cell types. On the other hand, THY-1 may not be required for infection of all cell types; instead, it functions in a cell type dependent manner. Other herpesviruses use different receptors to enter different cell types. HSV uses nectin-1 to enter neurons and HVEM to enter lymphocytes [51]. Some cell lines that express very low levels of THY-1 are still susceptible to HCMV infection, particularly at high m.o.i. or after prolonged virus inoculation. It is likely that HCMV enters cells through different pathways, either by direct fusion at the cell surface or by various endocytic pathways, especially when large amounts of virus are used in vitro. This is similar to the case of Lassa virus infection, in which the impairment of virus glycoprotein mediated entry imposed by deletion of host receptor glycosylated α-dystroglycan can be overcome by using high titer virus (m.o.i > 0.5), resulting in virus entry through an alternate pathway involving heparin sulfate, lysosome-resident protein, and pH-dependent endocytosis [52]. Previous studies have shown for other viruses entry dynamics are highly dependent on the m.o.i. Virus internalization occurs much more rapidly when a high m.o.i. (m.o.i 10) is used, compared to a low m.o.i of 0.01–1 [53,54]. For HCMV, infection at low m.o.i. (≤ 0.01) resulted in different profiles of virus replication and signaling as compared with infection at higher m.o.i (0.1–3.0) [55–57]. In the current study, we used a combination of low m.o.i. (between 0.05–1) and short time for infection (60 minutes followed by inactivation of virus remaining on the cell surface) to focus on the most efficient entry pathway(s). As shown in S5 and S6 Figs, only a fraction of the input virus entered cells within 1 to 2 hours at the low m.o.i. Nonetheless, a low m.o.i. is likely more representative of the virus to cell ratio present during natural infection. Since THY-1 is a major cargo protein of clatherin-independent endocytotic carriers [58], it is possible that THY-1 leads virions into the cells by macropinocytosis. Many viruses down-regulate and internalize their receptors from the cell surface through endocytic pathways [59]. Previous studies have shown that THY-1 is down-regulated in fibroblasts [60,61], as well as in mesenchymal stem cells [62] upon HCMV infection in a manner similar to that of PDGFR-α [63].
THY-1 is known to interact with cell proteins that facilitate HCMV entry. THY-1 engages αVβ3 integrin receptors and recruits paxillin [64], and triggers protein kinase dependent signaling pathways such as PI3K and Src [50,65,66]. THY-1 was important for activation of Akt in virus-infected cells and activation of PI3K –Src pathway has been shown to be required for HCMV entry [14,20,22]. Our findings that THY-1 facilitates an early step of HCMV infection, and that down-regulation of THY-1 by siRNA or blocking THY-1 with antibody inhibits HCMV- induced PI3K-Akt activation within the initial 15–45 min of infection, suggests a pivotal role for THY-1 in the coupling of HCMV entry with host signaling, and supports observations that growth factor receptors (PDGFR- α and EGFR) engage integrin/paxilin pathways during HCMV infection [14,16,22,67]. THY-1 protein is localized in lipid rafts through its GPI anchor. Ligand-mediated clustering of THY-1 in cholesterol-rich microdomains is needed to trigger Src-dependent downstream signaling [68,69]. We hypothesize that THY-1 clustering might be induced by interactions between THY-1 and HCMV gB and/or gH, two molecules that have been reported to contribute to signaling during virus entry [22,70]. This is similar to observations that binding of Group B coxackievirus to its receptor decay-accelerating factor (DAF), a GPI anchoring protein, induces DAF clustering to initiate signaling by Src family kinases [71].
We found that THY-1 interacts with both full length and 55 kD cleavage forms of gB, as well as with gH. Both full length and cleaved forms of gB are present on infected cells and virions [72]. Furthermore, THY-1 colocalizes with gB and gH in HCMV-infected cells. However, it is not clear whether THY-1 interacts with gB or gH directly or indirectly. Several studies have shown that exogenous HCMV gB and gH interact [10,73]. gH/gL have been postulated to function as receptor binding proteins, while gB may act as a fusogen; however, gB also binds to ligands and signaling molecules [8,74]. HCMV gB has been identified as a ligand for putative entry mediators, including integrins, PDGFR-α, and EGFR [14,16,22]. Like THY-1, both EGFR and PDGFR-α have been shown to form a complex with αVβ3 integrin [75,76], and are activated when they oligomerize after binding with ligands [69,77,78]. Therefore, THY-1 may be part of a multimolecular complex mportant for the initial phase of CMV infection and signaling (that includes PDGFR- α, EGFR, integrins, paxillin, and viral glycoproteins). However, it is uncertain how THY-1 fits into this complex and the exact form and timing of interaction(s) between THY-1, gB, and gH are unclear. Previous studies of HSV showed recruitment of other viral and host molecules to a complex after gD binds to its receptor. Since THY-1 interacts with several other cellular proteins, including integrins and is important in multiple signaling pathways, it is likely that THY-1 facilitates HCMV infection at an early stage as an entry mediator, rather than a receptor for a viral glycoprotein(s).
HCMV disseminates in leukocytes throughout the body after infecting mucosal epithelial cells, and induces production of inflammatory cytokines and increases permeability of the endothelium. This process is dependent on activation of the PI3K signaling pathway, which promotes extravasation of leukocytes into tissues [67,79,80]. Binding of THY-1 induces vascular permeability and regulates the extravasation of leukocytes during inflammation [81]. THY-1 is expressed in many types of cells that can be productively infected by HCMV as well as CD34+/CD38- stem cells, a putative cellular reservoir for latent infection [62,82]. Therefore, THY-1 may have a central role in mediating HCMV infectivity, coupling integrin/paxillin and leukocyte extravasation signaling, and linking the process of viral entry with signaling modulation of host cells that leads to the virus replication.
Human diploid fibroblast (MRC-5), retinal pigmented epithelial (ARPE-19) cells, and CV1/ EBNA-1 cells were acquired from the American Type Culture Collection (ATCC, Manassas, VA), and maintained in Minimum Essential Medium, F12 medium, or Dulbecco’s Modified Eagle Medium, respectively, with 10% FBS.
HCMV antibodies used were monoclonal anti-gB (Virusys, Taneytown, MD), monoclonal anti-gH (US Biological, Swampscott, MA), rabbit anti-gH antibodies (from Teresa Compton, University of Wisconsin and. David Johnson, Oregon Health & Science University), and mouse anti-ICP 8 antibody (Novus, Littleton, CO). Monoclonal antibodies used to detect THY-1 or isotype control antibody were purchased from Novus (Littleton, CO), Millipore (Billerica, MA) and BioLegend (San Diego, CA). Polyclonal goat anti-THY-1 or rabbit anti-THY-1 were obtained from Novus and GeneTex (Irvine, CA). Monoclonal antibodies for total and phosphorylated Akt were purchased from Cell Signaling Technology (Boston, MA). Expression plasmids encoding HCMV gB and gH were kindly provided by Teresa Compton [21]. pCMV-THY-1 was purchased from Open Biosystems (Huntsville, AL), and pCR2.1-GAPDH was a gift from Dr. Helene Rosenberg (NIAID, NIH). Plasmid expressing VZV full length gB was constructed by PCR amplification of gB ORF from the Oka strain of VZV, cloned into pcDNA3.1 vector with a C-terminal V5 epitope tag, and verified by sequencing.
BAC DNAs for epithelial/endothelial tropic HCMV strains BADrUl131-GFP and TB40E-GFP (kindly provided by Thomas Shenk, Princeton University, NJ; TB40E-GFP is also referred to as GS1783TB40-GFP) [37] were electroporated into ARPE-19 cells and the resulting viruses were propagated in ARPE-19 cells. Towne-GFP and AD169, both fibroblast topic HCMV strains, were propagated in MRC-5 cells. Cell culture supernatants from virus-infected cells were centrifuged at 2000 x g for 30 min at 4°C, and the clarified supernatants were used as virus stocks or further partially purified by centrifugation through a 20% sucrose or sorbitol cushion with a JA25 rotor at 35000 x g at 4°C for 60 min and resuspended in growth medium [83,84]. GFP expressing adenovirus (adenovirus-GFP) and herpes simplex virus 2 (HSV-2-GFP) were used as controls [85,86].
54 cell lines from the NCI-Frederick Tumor Cell Line Repository were purchased through Charles River Laboratories (Frederick, Maryland) and grown in RPMI-1640 medium with 10% FBS [34]. Cells with less than 20 passages after receipt were used in the study. Screening was performed based on previously published methods that have been used to identify other proteins important for the early stage of virus infection [29,30,32,87]. Briefly, cells were infected with GFP-expressing HCMV for 2–3 days, and susceptibility to HCMV was determined by FACS analysis based on GFP positivity, and normalized against infectivity for MRC-5 (for fibroblast tropic virus) or ARPE-19 (for epithelial and endothelial tropic viruses) cells. Bioinformatic analyses to determine correlations between HCMV infectivity and expression of each cellular gene were performed using the COMPARE algorithm and further detailed using MAPP software as described previously [30,38].
HCMV was added to cells on ice for 60 min for virus binding. The temperature was then shifted to 37°C for 60 min to allow virus entry. The cells were then treated with low pH citrate buffer (sodium citrate 40 mM, potassium chloride 10 mM, sodium chloride 135 mM, pH 3.2) at room temperature for 3 min to inactivate any virus that had not yet internalized, washed twice, and cultured for either 6 hrs to detect viral encoded RNAs or 3 days to quantify GFP positivity in cells infected with HCMV expressing GFP. A low m.o.i. (0.05–1) was used for most virus entry experiments since this is likely what occurs during natural infection; a high m.o.i (with a high percentage of cells infected) has been shown to result in different kinetics of entry than a low m.o.i. [53,55,88]. GFP positivity was determined by FACS analysis using a BD FACSCalibur and data was analyzed with FlowJo software (Tree Star, Ashland, OR).
MRC-5 cells infected with HCMV at an m.o.i. of 5 for 3 days or uninfected control cells were lysed in buffer (25 mM Tris, 15 mM NaCl, 0.1% NP40 with or without 5mM EDTA) as described previously [46]. Immunoprecipitation with anti-THY-1 antibody and protein A-Sepharose (Sigma-Aldrich, St. Louis, MO) was carried out at 4°C overnight. After extensive washing, proteins were separated in SDS-PAGE gels under reducing conditions and visualized by Coomassie Blue or silver staining. Specific bands were excised and subjected to mass spectrometry for protein identification (Research Technologies Branch, NIAID, NIH).
Soluble THY-1-His protein, or VZV gE-His control protein, was bound to Ni-NTA or Talon columns at 4°C for 2 hr followed by extensive washing with PBS. HCMV-infected or uninfected cell lysates were centrifuged at 2000 x g at 4°C for 30 min, and the supernatant was added to columns and incubated at 4°C overnight with rotation. Columns were washed with PBS using a peristaltic pump, and eluted with 250mM imidazole. Samples were concentrated by centrifugation at 1600 x g in an Amicon Ultra centrifugal filter unit (3000 MWCO), separated in SDS-PAGE gels, transferred to nitrocellulose membrane, and immunoblotted with antibodies.
Total RNA was extracted using an RNeasy Mini Kit (Qiagen, Valencia, CA) following the manufacturer’s instructions. To eliminate DNA contamination, RNA was treated with DNase I (Roche Applied Science, Indianapolis, IN) and purified a second time with an RNeasy Mini Kit. Quantitative real-time RT-PCR was performed using One-step RT-PCR Master mix reagent (Applied Biosystems, Carlsbad, CA) with a 7500 Real Time PCR machine. Primers and probes for detection of HCMV immediate-early gene UL123 and late gene UL55 were described previously (Boeckh et al., 2004). Primers (5’- GTTAGGCTGGTCACCTTCTG, 5’- GAGATCCCAGAACCATGAACC) and probe (5’- AGACTGTTAGCAGGAGAGCGATGC) for THY-1 were located in exon 1. Primers and probe for GAPDH were purchased from Applied Biosystems (Carlsbad, CA). Serial dilutions of HCMV Bac DNA, THY-1 plasmid, or human GAPDH plasmid were used to generate standard curves, and copy numbers of THY-1 and HCMV RNAs were normalized to copy numbers of human GAPDH amplified from the same wells. DNA contamination was monitored by performing PCR amplification without reverse transcriptase.
THY-1 specific siRNA SmartPools (M-015337-00) and non-specific control pools (Duplex-13), THY-1 single siRNA oligos with targeting sequences CAACUUCACCAGCAAAUAC (THY-1-02) and GGACUGCCGCCAUGAGAAU (THY-1-04), and non-targeting single oligo #4 were obtained from Dharmacon (Lafayette, CO). Cells were transfected with siRNAs (125 pmol per 2 x 106 cells) using nucleofection (Amaxa, Gaithersburg, MD) for 48 hr before infection or harvesting.
DNA corresponding to THY-1 amino acids 20–130 with a C-terminal (His)6 tag was amplified by PCR from plasmid pCMV-THY-1 (using primers 5’-CAGAAGGTGACCAGCC and 5’-GCTCAGAGACAAACTGGTCAAG, and cloned into pDC409 [89]. The THY-1 insert in the resulting plasmid, pDC409-THY-1(20–130)-His, was completely sequenced. THY-1-His soluble protein was expressed in CV1/EBNA 1 cells and purified with a Ni-NTA column (Invitrogen, Grand Island, NY) or Talon resin (Clontech, Mountain View, CA), eluted with 250 mM imidazole, dialyzed against PBS at 4°C overnight, and concentrated with an Amicon Ultra centrifugal filter unit (3000 MWCO) (Millipore, Billerica, MA). Filtrates derived from this filter unit with the same buffer composition, but lacking THY-1 protein, were used as a negative control for experiments. Soluble varicella-zoster virus gE with a C-terminal (His)6 tag, gE-His [46], was used as an additional control in soluble THY-1-His protein experiments.
Cells were fixed in methanol/acetone (1:1) at -20°C for 5 min. After washing in PBS, blocking buffer (4% BSA and 10% normal goat serum in PBS) was added for 1 hr before incubation with mouse monoclonal anti-HCMV gB or gH, and goat anti-THY-1 for 60 min on ice, followed by anti-mouse Alexa-488 or anti-goat Alexa-594 (Invitrogen, Grand Island, NY) on ice for 60 min. For cell surface staining, live cells were treated with blocking buffer on ice for 30 min before incubation with primary and secondary antibodies. After antibody staining, cells were fixed with 2% paraformaldehyde, and mounted with DAPI-Fluoromount-G (Southern Biotech, Birmingham, AL). Confocal imaging was performed with a Leica SP5 X-WLL microscope.
|
10.1371/journal.pcbi.1004614 | Automated Learning of Subcellular Variation among Punctate Protein Patterns and a Generative Model of Their Relation to Microtubules | Characterizing the spatial distribution of proteins directly from microscopy images is a difficult problem with numerous applications in cell biology (e.g. identifying motor-related proteins) and clinical research (e.g. identification of cancer biomarkers). Here we describe the design of a system that provides automated analysis of punctate protein patterns in microscope images, including quantification of their relationships to microtubules. We constructed the system using confocal immunofluorescence microscopy images from the Human Protein Atlas project for 11 punctate proteins in three cultured cell lines. These proteins have previously been characterized as being primarily located in punctate structures, but their images had all been annotated by visual examination as being simply “vesicular”. We were able to show that these patterns could be distinguished from each other with high accuracy, and we were able to assign to one of these subclasses hundreds of proteins whose subcellular localization had not previously been well defined. In addition to providing these novel annotations, we built a generative approach to modeling of punctate distributions that captures the essential characteristics of the distinct patterns. Such models are expected to be valuable for representing and summarizing each pattern and for constructing systems biology simulations of cell behaviors.
| Determining the subcellular location of all proteins is a critical but daunting task for systems biologists, especially when variation between different cell types is considered. Fluorescence microscopy is the main source of information about subcellular location, but large collections of fluorescence images for many proteins are frequently annotated visually and result in assignment only to broad categories. In this paper, we describe automated methods for analyzing images from the Human Protein Atlas to identify nine specific punctate patterns and assign these more specific annotations to 550 proteins many of which previously had little information about subcellular location. We also describe building models of these patterns that will be useful for carrying out systems biology simulations of cellular reactions using accurate spatial distributions.
| Fluorescence microscope images can provide important information about the subcellular location of proteins, and automated systems can be used to assign these proteins to major subcellular location classes with accuracy at or above that of human annotators [1, 2]. However, assigning higher resolution annotations to proteins is more difficult, especially for punctate or vesicular patterns. Punctate subcellular localization patterns may arise either from membrane-bound organelles (e.g., transport vesicles) or from macromolecular complexes of sufficient size (e.g., ribonucleoprotein (RNP) bodies), and they may be quite visually similar. We refer to individual components of these patterns collectively as puncta, to encompass both types of structures. These are important for various cellular tasks such as endocytosis, exocytosis and RNA recruitment, storage or degradation. A critical factor for accomplishing many of those tasks is the association of the vesicles or bodies with cytoskeletal components such as microtubules for intracellular transport. Although microtubules are not necessary for short-range transport, they are required for rapid transport of vesicles [3]. The extent to which the distributions of specific puncta are related to that of microtubules remains unclear, as is the extent to which the distributions vary across different cell lines.
Our understanding of cell behavior and the sources of cellular variation can be significantly aided and tested using cell modeling and simulations [4–6]. For this, we need a mechanism to capture the spatiotemporal behavior of cellular substructures, both as a starting point for simulations and to compare against results. Towards this end, we have previously described systems for building image-derived, 2D or 3D generative models of the distributions of either punctate organelles [7, 8] or microtubules [9] within cells. These models are conditional (dependent) on models of cell and nuclear membranes, but they are independent of each other; that is, they do not consider the relationship between puncta and microtubules.
Here we describe a new computational method that allows us to model this relationship. Our method requires images in which both punctate proteins and microtubules are visualized. The Human Protein Atlas (HPA, http://proteinatlas.org) is a rich source of such images, containing high-resolution images of subcellular location patterns for thousands of proteins in several cell lines [10]. To analyze the patterns of punctate proteins in the HPA, we designed a generative model consisting of compact and interpretable features to characterize the population of puncta within a cell, including measurements of microtubule association, relationship to cell geometry, density, intensity and appearance. We have used the features of these models to discover the major modes of variation among punctate patterns, and to assign subclasses of punctate patterns to unannotated proteins.
We began by creating an image processing pipeline that identified individual puncta and microtubules in 2D confocal microscopy images from the HPA. As illustrated in Fig 1A, an input image (Fig 1C) is processed to create images of puncta and microtubules (shown as a composite in Fig 1D) and of the remaining background protein fluorescence (Fig 1E). One of our major goals was to generate a model of the distribution of puncta that captures their relationship to microtubules. This would presumably reflect the extent to which puncta were bound to microtubules to accomplish transport to or retention in particular regions of the cell. As a simple measure of this association, we computed the distance (d) between each punctum and the nearest microtubule (Fig 1B). We would expect puncta that are bound to microtubules to have a small distance compared to those that are not bound, and perhaps also that the distribution of distances would reflect the extent to which released vesicles diffuse away before being bound again. We added this measure to our previous vesicular object distribution model [8], which included dependence on fractional distance between the nucleus and plasma membrane (r, calculated from L1 and L2) and the angle (α) to the major axis of the cell (see Methods). We also created a model for background intensity that was similarly dependent on microtubules and cell shape (see Methods). We combined the estimated parameters from these models with five parameters that describe puncta size and shape and two parameters that measure the amount of fluorescence in puncta and background. This resulted in twenty-two parameters (S1 Table) that can be readily determined from each image of a protein’s subcellular distribution in an individual cell. We used these parameters both as features to describe protein patterns and, later, to construct generative models of punctate patterns.
A number of proteins in the HPA are assigned annotations of “vesicles” or “cytoplasm”. We considered whether we could use HPA images to assign these proteins to a more specific organelle or structure. By examining UniProt annotations and primary literature for proteins whose subcellular location has been reasonably well characterized, we selected eleven proteins that are found in eleven specific types of punctate patterns (Table 1) (we refer to these proteins as “founders” since they enabled us to define specific subtypes). We chose these patterns due to the fact that the proteins showed a similar pattern across all three cell types in the HPA and they represent a wide range of membrane and non-membrane bound compartments (although there are of course additional punctate patterns for which we did not find appropriate founders). In particular, they cover all main compartments of the endomembrane system. We calculated the feature values for all cells for each combination of the eleven proteins and three cell lines. We verified that the features accurately reflect the relationship between vesicles and microtubules by comparing the cumulative distribution of the experimentally measured distance between puncta and microtubules with that calculated from the model; the distributions were very similar for all eleven patterns (S1 Fig). We then asked whether these patterns could be distinguished from each other in HPA images. To provide a visual basis for illustrating how the proteins differed in the features, we calculated the first three principal components. Fig 2 shows the position of each antibody-cell line combination in two projections of this three-dimensional space, as well as representative images along each principal axis. For a given cell line, the eleven patterns are roughly separable, although the position of a given protein sometimes varies from cell line to cell line. For example, proteins 2, 3, 6 and 7 are close together in pc1 and pc2 but separated by pc3. From inspection of the projection of each numerical feature onto the three most significant principal component axes, as well as the example images, it appears that the first component primarily represents variation in features 12, 13 and 5, which capture relationship to microtubules and variation in intensity. The second primarily represents variation in features 21, 22, 2, and 8, which capture intensity and distance from the nucleus, while the third principal component represents variation in features 1, 3, and 4, which capture puncta size and variation in size. This figure does not permit accurate assessment of the overlap between patterns, but is presented to give a visual overview of the major modes of variation with the patterns.
These results suggest that the feature set may be a reliable basis for measuring variation in punctate patterns, and we therefore sought to determine whether we could use them to predict the compartmental localization of other proteins to one of the eleven patterns. To do this, we first used the features to construct a classification accuracy-derived separability statistic to compare two collections of cells (see Methods) and assessed the extent to which the eleven patterns could be distinguished. We used a classification approach based on Bayes error rate in order to avoid problems with imbalance between the numbers of proteins in each class and to allow for class-specific differences in scale for different features (see Methods).
For each cell type separately, we classified each image as belonging to one of the eleven patterns using hold-out-image cross validation: for each held-out image, we calculated the separability between the cells contained in that image and the cells of each of the founder patterns. The image was given the label of the pattern that was least separable from it. Using this method for each cell type we achieved an average class accuracy of 86.9% (Table 2). We compared these results to those using the same classification procedure but excluding the features relating to microtubule distribution, which resulted in 82.8% average accuracy. This demonstrates that the relationship to microtubules provides information that improves our ability to distinguish punctate patterns. Further examination of Table 2 reveals that the coated pits pattern is the only one that is consistently difficult to distinguish. This may in part be due to the fact that 2D confocal images were used, and thus the features cannot easily distinguish whether puncta are on the surface or inside the cell (for the other surface puncta pattern, caveolae, their distribution or size must allow them to be distinguished).
We next asked whether the classification approach could be used to assign a punctate subpattern annotation to an image of proteins other than the founders. We did not want to simply assign the subcellular location of the class that a protein was most similar to (since the protein might not actually be from any of our classes), but wanted to ensure that we only assigned annotations for proteins with a high degree of similarity to one of the founders. For each cell type, we determined a threshold on the separability statistic that could be used to determine whether or not a new protein should be assigned to a particular class. This threshold was determined as the optimal point of the receiver operating characteristic curve (see Methods and S2 Fig) for each cell type.
To assign subcellular location to a new image, we measured the separability between it and each founder pattern. If the value for one of the patterns was below the threshold, we assigned the corresponding pattern label to that image. In the rare case of an image being below the threshold of multiple patterns, we assigned it the label “ambiguous.” This classification procedure was applied to the remainder of images in the HPA dataset; the results are contained in S1 Dataset. One hundred and twenty-five proteins were identified as belonging to one of the eleven classes in A-431, 60 in U-2OS, and 365 in U-251 MG. The list of the most confident assignments is shown in Table 3. With the goal of providing improved annotations for protein databases, we also generated an XML file that can be used to update those databases. The file (S2 Dataset) contains information on HPA antibody IDs, gene targets and proposed annotation. Due to the nature of immunofluorescence tagging, a sequence-specific tag may be present on more than one protein isoform, each of which may show a condition-specific localization pattern. With that in mind, we also report the known protein gene products provided by ENSEMBL 79, and the percentage of matching peptides after alignment between the gene-product and antigen sequences in the region spanned by the antibody. We also provide annotations to all protein isoforms that match the antibody sequence. For those proteins and isoforms that have a high confidence location assignment, we also provide an XML file for updating their UniProt record (S3 Dataset).
In order to provide an independent assessment of the accuracy of the annotation procedure, we searched for literature describing the localization of the most confident annotations. We were able to find literature supporting our proposed labeling for many of the proteins (although they had often only been analyzed in other cell types). For example, of the top hits for A-431 cells, BRD4, has been suggested to be involved in the lysosome protolytic pathway [11]. For U-2OS, top hit RAB5C is a classic early endosomal protein [12], and prohibitin (PHB) is a multifunctional membrane protein [13] one of whose roles is in regulation of degradation of PAR1 [14]. For U-251MG cells, the top hits include cathepsin H (CTSH), a lysosomal enzyme, DTX3L, which regulates endosomal sorting [15], and LY6K, which, like other Ly6 antigens, is associated with glycosylphosphatidyl inositol-anchored glycoproteins (such as TEX101 [16]) that are typically found in caveolae. These findings increase our confidence in the proposed annotations.
Many of the proteins analyzed (which were all proteins assigned “vesicles” or “cytoplasm” annotations) were not assigned with high confidence to any of the 11 patterns. There are at least three potential reasons for this. First, the staining may be of low enough intensity or quality that foreground cannot be adequately identified. Second, the unassigned proteins may be cytoplasmic proteins without a discernible punctate pattern, or vesicular proteins from an organelle that we have not considered. Third, they may be present in more than one of the eleven patterns, such that their pattern does not match well enough to any of them.
Our models allow us to ask whether different punctate subclasses differ in their relationship to microtubules. We performed a simple characterization of this relationship by calculating the average actual distance of each punctum from microtubules, as well as the average distance from microtubules predicted by our fitted model. S3 Fig shows a comparison of these two distances for each pattern across all cell types and for each combination of pattern and cell type. A confidence interval on the average distance from microtubules was determined via the Tukey-Kramer method after two-way ANOVA [17] (across proteins and cell types). All of the symbols are quite near the diagonal, indicating that the model is in high agreement with the measurements. When averaged across all three cell types, retromer, recycling endosomes, and early endosomes show the closest association with microtubules, and RNP bodies, COPI vesicles and coated pits show the least. When each combination of protein and cell type is considered separately, we see greater variability in the distances (perhaps due to differences in microtubule-binding proteins or cell size or shape). COPII, lysosomes and COPI show the least variation across the three cell types, and coated pits and recycling endosomes show the greatest.
Another way in which we can compare the different patterns is by examining the differences in the model features among them. A simple visualization of this is shown in Fig 3, in which the relative values of each feature are shown for each pattern. In U-2OS, for example, the first four features (relating to size and intensity) clearly distinguish the group of RNP bodies, late endosomes, recycling endosomes, lysosomes and COPII from the others, and a high value for mx5 (number of puncta) separates RNP bodies from this group. Other distinguishing features or feature combinations can also be identified, such as retromers having the lowest value for mx12 (consistent with their close association with microtubules). These differences provides a interpretable rationale for the ability of the classifiers to distinguish the patterns.
A difficult question that frequently gives rise to controversy is how to best describe the subcellular pattern of a given organelle or structure (especially a novel one). Descriptions using unstructured text or Genome Ontology terms defer the question by assuming that the words will be sufficient for the reader to be able to mentally construct the pattern. An alternative is to show an example image, but this does not give an idea of the variation in the pattern (one can find differences between any two example images, but this does not address whether those differences are statistically significant). Unfortunately these two methods of conveying information about the distribution and variation in protein pattern do not provide a quantitative, or much less a probabilistic or statistical representation of the observed pattern. Alternatively, one can give values for a descriptive feature vector or matrix for each pattern (which can be used for a classifier) but this allows one only to recognize new examples but not to produce an example of the pattern. Feature vectors also do not necessarily allow an explicit model of the relationship between cell components. Of course, none of the approaches above are helpful if we desire an in silico representation of the cell geometry and expressed patterns (i.e., the consumer of the representation is a computer rather than a cell biologist). For example, information about subcellular patterns is needed for accurate mathematical simulations of cell biochemistry and behavior [4–6]. As a solution, we have introduced the building of generative models of cell organization directly from images [7, 8, 18–20]. The intent is for these models to capture the underlying properties of a particular pattern; in statistical terms, to capture the distribution from which all examples of that pattern are drawn. Such a model can be used to synthesize new cell images from that distribution.
We therefore constructed a generative model of punctate patterns whose structure is shown in Fig 4. The model starts with models of nuclear and cell shape (dn, dc) and microtubule distribution (dm) and links them to models of puncta distribution using mx7 through mx11 to capture dependence on cell shape and mx12 and mx13 to capture dependence on microtubules (see Methods). Additionally the size, shape and intensity of vesicles are modeled independently of the cell shape and microtubules with mx1 through mx6. The background intensity is similarly modeled dependent on cell shape and microtubules (mx14 through 20) and scaled to match the fraction of intensity with mx21 and mx22. We illustrate that the images generated from the models learned for each of the pattern classes are similar to real images in Fig 5 and S4 Fig.
Assuming that the distributions of the eleven punctate patterns are independent of each other, we can combine the models and synthesize cells containing all eleven. Fig 6 shows an example of a “typical” cell under this assumption (using the average values of all model parameters).
With the development of systems to fluorescently tag and acquire images of thousands of subcellular protein patterns, a need arose for automated methods to analyze and model the patterns in these images [21]. The goals of such analyses include, but are not limited to, determining the organelles to which different proteins localize and studying the statistical dependency between different protein patterns. However, previous methods have not been able to recognize subpatterns of the major organelle types. Furthermore methods are needed to describe the relationships between cellular components in a way that is not only human-interpretable, but allows us to generate new examples of these patterns for future use in cell simulations [22].
Here we have described a new framework to build models of subcellular punctate patterns conditional on cell geometry and microtubules. These models use interpretable features that capture specific ways in which punctate subpatterns differ between cell types (such as the differences noted at the beginning of the Results) and can generate synthetic cell instances representative of the modeled population. We demonstrated the value of this framework by learning models directly from images of eleven well-characterized punctate protein patterns in three cell types. We showed that the major variation in these patterns corresponded to dependence on microtubules, total intensity, and puncta size and shape. Given the model parameters we constructed a pipeline demonstrating both the high discriminative ability of this model across patterns of the same cell type and the ability to automatically assign annotations to 550 proteins (many of which had been poorly characterized previously with respect to subcellular location).
High-content screening and analysis have become increasingly frequent, including subtle analysis of location changes induced by chemical compounds or inhibitory RNAs and proteome-scale analysis of patterns. The features we have described should be useful for refining the ability to distinguish different vesicular and punctate patterns, and, most importantly, to provide an interpretable and portable basis for comparing them.
The work presented here represents an important step towards bridging detailed models learned from large collections of images for proteins contained in discrete objects with models of microtubule network growth learned by inverse modeling [9, 18]. It serves as an important component of our CellOrganizer project (http://cellorganizer.org/) [20], which aims at capturing a detailed model of the spatial organization and relationships between different subcellular location patterns. We plan to extend this work by merging it with models of subcellular pattern dynamics, as well as extend the model to capture further dependency between components. It is hoped that approaches like this will enable the construction of models that capture essential cell behaviors without requiring the simultaneous measurement of the thousands of different proteins in the same living cell, something that is infeasible with current technology.
The data used here were confocal immunofluorescence microscopy images of fixed cells from A-431, U-2OS and U-251MG cell lines from HPA [10]. All antibodies whose subcellular pattern was annotated as “vesicles” or “cytoplasm” were chosen (a total of 2357, 3038, and 1730 proteins for each line; S1 Dataset contains the complete list of proteins analyzed). The images were analyzed as 8-bit TIFF images with three channels each obtained using a different emission wavelength of fluorescence from a single image field. The three channels show the locations of a specific punctate protein, a nuclear stain, and microtubules. Each of the images is 1728 × 1728 pixels and the pixel size corresponds to 0.08 microns in the sample plane. Founder proteins for eleven patterns were chosen as described in the Results. After segmenting the image fields for these proteins into single cell regions using a seeded watershed method [2], the set of founder images was found to contain 1099 cells, 333 from A-431, 327 from U-2OS and 439 from U-251MG (the number of cells for each of the 33 combinations of antibody and cell line varied from 12 to 85).
In cell images, due to variation in fluorescence intensity in the cytoplasm, segmentation of puncta and microtubules from protein pattern images poses a difficult problem where global threshold-based methods may over-threshold regions of the cytoplasm containing low-intensity structures. The input cell image was de-noised by blurring with a Gaussian filter with standard deviation of 0.75. We isolated high spatial-frequency foreground and low spatial-frequency background intensity images by low pass filtering the smoothed image with a Gaussian filter of 4-pixel standard deviation, and subtracted this background image from the smoothed image, resulting in an image of high-frequency foreground signal (i.e. puncta). The negative-valued pixels of the foreground signal were removed, and the foreground image was subtracted from the first smoothed image, to get the background image (both of which sum to the total image intensity). To increase the speed at which a Gaussian mixture model could be fit over the foreground image, we excluded all pixels below the Ridler-Calvard threshold [23] and all single-pixel objects. We used the skeletonized foreground signal of the microtubule image to model the distances of objects from microtubules. This approach resulted in reasonable definition of both puncta and microtubules and was sufficient to capture variation across the founder patterns analyzed in this paper.
The centroids of all puncta were computed by fitting a mixture of Gaussians to distinguish overlapping puncta [7]. The distance between the centroid of each punctum and its nearest microtubule was found using a distance transform of the skeletonized microtubule image.
A probability density function (PDF) for the position of puncta (pp) relative to the cell geometry and microtubules was estimated by extending the model previously described [8] by adding a terms describing the distance from microtubules, d:
P(r,a,d) = eβ0+β1r+β2r2+β3sinα+β4cosα+β5d+β6d21+eβ0+β1r+β2r2+β3sinα+β4cosα+β5d+β6d2
(1)
The terms β1 through β4 describe the dependency of objects on radial and angular coordinates in relation to the shape of the cell [2, 8], and β5 and β6 describe the dependency of objects to be localized in relation to the microtubules. We similarly constructed a PDF for the background intensity (which presumably results from soluble, non-punctate protein).
The Bayesian hierarchical framework for the generative model for puncta is shown in Fig 3 as a graphical model. A multivariate statistical model was constructed from the independent distributions of values of the following statistics from each cell: puncta size (sp), puncta per cell (np), and intensity (ip).
Synthetic cell instances were created starting from the cell and nuclear boundaries and microtubule image of a randomly-selected cell. (They can also be created by first generating cell and nuclear boundaries and microtubule distributions using models learned previously for the three cell lines [18].) To add puncta to a cell, values were sampled for the number of puncta per cell (np) and the size (sp) and fluorescence intensities (ip)) for each punctum from distributions learned from 2D HPA data. These were used to generate puncta using the Gaussian object based generative model [8]. Positions for them were sampled from the vesicle position PDF from the model above after morphing to the specific cell geometry. Background fluorescence was added using the learned PDF from the background images, scaled to match a draw from the total background intensity distribution learned from images.
The assignment of subcellular annotations to images of cells is a classification task with complications found in many biological contexts; specifically being the structured nature of data (cells with the same antibody should all be assigned the same label), the inseparability of class data (proteins with different biochemical properties may have similar localization patterns), and imbalanced number of observations(some images may contain many cells while others have few). We designed a classification method to specifically address the above complications.
Given pattern parameterizations corresponding to cells of two collections (all cells contained in two images), we perform a balanced classification task to determine how distinguishable the two collections are. For each pair of images, we hold out a subset of cells and train an SVM by weighting the training data such that there is a uniform prior across the classes. We then classify the hold-out and count the frequency at which the hold-out was assigned the correct collection, approximating the Bayes Error rate [24]. This approach is similar to other methods used in genomics [25]. We take the average classification accuracy across all cell classification tasks (whether or not the cells belonging to the two images are assigned the same subcellular pattern) as a measure of how distinguishable the two collections are, resulting in a possible range of values from 1 (totally separable) to 0 (completely inseparable). In virtually all cases, the measure of difference lies between 0.5 and 1. We will refer to this measure as “dissimilarity”.
To determine a threshold on dissimilarity, at which we can say two collections belong to the same or different patterns, the pipeline treats images of each of our basis patterns as their own collection (with multiple images of each pattern) and performs the above classification task using cells contained in each image. An ROC curve is constructed, indicating the true and false positive classification rates as a function of increasing dissimilarity. For each cell type we constructed an upper-bound of dissimilarity (above which is considered “not the same annotation”) by the cutoff determined at the location where the upper-left-most point of the ROC curve intersects with a slope of TN+FPTP+FN, where TN, FP, TP and FN are the counts of true negative, false positive, true positive and false negatives respectively. When comparing our basis set to images containing cells of unknown protein localization, we assign the unknown pattern the label of any basis pattern that is within the similarity threshold. These thresholds were 0.78846, 0.70588 and 0.72093 for A-431, U-2OS, and U-251MG, respectively.
All software and data used for this work is available as a reproducible research archive (http://murphylab.web.cmu.edu/software). The software will also be available as part of the open source CellOrganizer system (http://CellOrganizer.org). The segmentation and feature calculation pipeline can be used separately.
|
10.1371/journal.ppat.1004366 | Sialylation of Prion Protein Controls the Rate of Prion Amplification, the Cross-Species Barrier, the Ratio of PrPSc Glycoform and Prion Infectivity | The central event underlying prion diseases involves conformational change of the cellular form of the prion protein (PrPC) into the disease-associated, transmissible form (PrPSc). PrPC is a sialoglycoprotein that contains two conserved N-glycosylation sites. Among the key parameters that control prion replication identified over the years are amino acid sequence of host PrPC and the strain-specific structure of PrPSc. The current work highlights the previously unappreciated role of sialylation of PrPC glycans in prion pathogenesis, including its role in controlling prion replication rate, infectivity, cross-species barrier and PrPSc glycoform ratio. The current study demonstrates that undersialylated PrPC is selected during prion amplification in Protein Misfolding Cyclic Amplification (PMCAb) at the expense of oversialylated PrPC. As a result, PMCAb-derived PrPSc was less sialylated than brain-derived PrPSc. A decrease in PrPSc sialylation correlated with a drop in infectivity of PMCAb-derived material. Nevertheless, enzymatic de-sialylation of PrPC using sialidase was found to increase the rate of PrPSc amplification in PMCAb from 10- to 10,000-fold in a strain-dependent manner. Moreover, de-sialylation of PrPC reduced or eliminated a species barrier of for prion amplification in PMCAb. These results suggest that the negative charge of sialic acid controls the energy barrier of homologous and heterologous prion replication. Surprisingly, the sialylation status of PrPC was also found to control PrPSc glycoform ratio. A decrease in PrPC sialylation levels resulted in a higher percentage of the diglycosylated glycoform in PrPSc. 2D analysis of charge distribution revealed that the sialylation status of brain-derived PrPC differed from that of spleen-derived PrPC. Knocking out lysosomal sialidase Neu1 did not change the sialylation status of brain-derived PrPC, suggesting that Neu1 is not responsible for desialylation of PrPC. The current work highlights previously unappreciated role of PrPC sialylation in prion diseases and opens multiple new research directions, including development of new therapeutic approaches.
| The central event underlying prion diseases involves conformational change of the cellular form of the prion protein (PrPC) into disease-associated, transmissible form (PrPSc). The amino acid sequence of PrPC and strain-specific structure of PrPSc are among the key parameters that control prion replication and transmission. The current study showed that PrPC posttranslational modification, specifically sialylation of N-linked glycans, plays a key role in regulating prion replication rate, infectivity, cross-species barrier and PrPSc glycoform ratio. A decrease in PrPC sialylation level increased the rate of prion replication in a strain-specific manner and reduced or eliminated a species barrier when prion replication was seeded by heterologous seeds. At the same time, a decrease in sialylation correlated with a drop in infectivity of PrPSc material produced in vitro. The current study also demonstrated that the PrPSc glycoform ratio, which is an important feature used for strain typing, is not only controlled by prion strain or host but also the sialylation status of PrPC. This study opens multiple new directions in prion research, including development of new therapeutic approaches.
| Prion disease is a family of lethal, neurodegenerative maladies that can be sporadic, inheritable or transmissible in origin [1]. The key molecular event underlying prion diseases involves conformational change of the normal, cellular form of the prion protein denoted PrPC into the disease-associated, self-propagating, transmissible form denoted PrPSc [2]. Upon expression in the endoplasmic reticulum, PrPC undergoes posttranslational modifications, including attachment of up to two N-linked carbohydrates to residues Asn-181 and Asn-197 and of glycosylinositol phospholipid anchor (GPI) to the C-terminal residue Ser-231 (residue numbers are given for hamster PrPC) [3]–[5]. These posttranslational modifications are intact upon conversion of PrPC into PrPSc [4], [6], [7].
Since the discovery that the PrPSc and PrPC glycans are sialylated more than 25 years ago [6], [8], the potential role of sialylation in PrPC function, prion replication or its pathogenesis remains uncertain. The two N-linked carbohydrates can carry from zero to four terminal sialic acid residues each [8], [9]. While the PrP polypeptide has a strong positive charge, the isoelectric points (pI) of PrPC can vary significantly in part due to variation in sialylation of the glycans [10]–[15]. In glycans sialic acid residues are linked to the galactose residues at the C-6 or C-3 positions [8]. The detailed site-specific characterization of mouse PrPC revealed that the majority of glycans at Ans-180 have bi- and triantennary structures and are sialylated to a lesser degree than the glycans at Ans-196, a majority of which are tri- and tetraantennary structures [16]. While the relative proportion of bi-, tri-, and tetra-antennary glycans appears to differ slightly in PrPC and PrPSc, the relative proportions of sialylated glycans was found not to be statistically different between PrPC and PrPSc [9]. Due to diverse structure and composition of oligosaccharides, PrPC primary structure consists of more than 400 different glycoforms. In addition to sialylation of both glycans, a single sialic acid was also found on a GPI anchor of PrPC and PrPSc [3].
The ratio of di-, mono-, and unglycosylated PrPC glycoforms was found to change in favor of di-glycosylated forms in the course of neuronal differentiation, as well as upon an increase in the density of cells cultured in vitro [17], [18]. While diglycosylated PrPC is the dominant glycoform in adult brain, the ratio of di-, mono-, and unglycosylated PrPC glycoforms was found to vary in different brain regions [19]. 2D-gel electrophoresis analysis revealed variations in isoelectric points (pI) of PrPC isoforms expressed in different brain regions [10], a variation that could presumably be attributed, at least in part, to the region-specific differences in sialylation status of glycans. Moreover, as probed by binding of 19 lectins specific to different sugars including sialic acid residues, the composition of PrPC glycans was found to change with normal aging [20].
In the last decade, numerous studies illustrated the essential role of protein sialylation in immunity including its role in cell signaling, cell activation, differentiation, and pathogen recognition (reviewed in [21]-[23]). While sialylation of cell surface proteins is also involved in a number of functions of central nervous system, including cell differentiation, adhesion and neuronal plasticity, a big gap in understanding the role of sialylation in the normal and pathological function of PrPC exists. Sialylation of PrPC glycans was shown to prevent binding of PrPC to selectins, a family of cell surface proteins that interact with carbohydrates in a Ca2+-dependent manner and participate in cell adhesion and migration [24]. A recent study examined the role of Siglec-1, a sialic acid-binding immunoglobulin-type lectin, expression of which is restricted to mononuclear phagocytes, in prion diseases [25]. While mononuclear phagocytes are known to be important for prion uptake and trafficking to/within lymphoid tissue and possibly prion clearance, no effect of Siglec-1 knockout on peripheral prion disease pathogenesis was observed [25]. Another study examined possible involvement of GPI sialylation in neurodegeneration and found that a dense clustering of sialic acid-containing GPI anchors in the plasma membrane resulted in alteration of membrane composition and synapse damage [26], [27]. The presence of sialic acid in the GPI was a requirement for the toxic effect expressed by clustering of PrPC molecules on cell surface [26], [27].
In the current work, we examined the role of sialylation of PrPC on prion replication, a topic that has not been addressed in previous studies. We showed that charge heterogeneity in brain-derived PrPC and PrPSc was due to sialylation and that undersialylated PrPC molecules (sialylated less than the statistical average for PrPC) were a preferable substrate for prion amplification in PMCAb. As a result, PrPSc produced in PMCAb was less sialylated than brain-derived PrPSc and also showed longer incubation time to disease. Consistent with the idea that sialylation of PrPSc is important for prion infectivity, PMCAb material produced using desialylated PrPC was not infectious. Nevertheless, in support of the hypothesis that PrPC sialylation controls prion replication rate, de-sialylation of PrPC was found to speed up considerably PrPSc amplification in PMCAb, with the magnitude of this effect found to be strain-dependent. Moreover, de-sialylation of PrPC reduced or eliminated a species barrier of prion amplification in PMCAb. Furthermore, 2D analysis suggested that sialylation status of brain-derived PrPC was different from that of spleen-derived PrPC. Surprisingly, the PrPSc glycoform ratio was found to be controlled by the sialylation status of PrPC, with a decrease in PrPC sialylation levels resulting in a higher percentage of the diglycosylated glycoform in PrPSc presumably due to a decrease in density of negatively charged groups on PrPSc surface. The current study exposes the previously underappreciated role of PrPC sialylation in a number of key aspects of prion diseases, including its role in controlling prion replication rate, its infectivity, species barrier and PrPSc glycoform ratio.
In the absence of posttranslational modifications the prion protein has a strong positive charge at physiological pH. Theoretical calculation of the isoelectric point for full-length Syrian hamster PrP predicts a value 9.58. However, due to posttranslational modifications and, primarily, sialylation of N-linked glycans and the GPI anchor, the actual isoelectric points of PrPC molecules could be substantially lower than 9.58 [14]. In fact, because each of the two N-linked glycans contains up to four terminal sialic acids (Figure 1E), brain-derived PrPC molecules are heterogeneous with respect to their charge as confirmed by 2D gel-electrophoresis (Figure 1A).
To test whether sialic acid residues indeed account for broad charge heterogeneity, Syrian hamster normal brain homogenate (NBH) was treated with A. ureafaciens sialidase (sialidase-treated NBH will be referred to as dsNBH), an enzyme that cleaves off terminal α2,3- and α2,6-linked sialic acid residues. While in non-treated NBH PrPC molecules are spread between pI 3 and 10 (Figure 1A, top), enzymatic desialylation resulted in a substantial shift of PrPC towards pI 10. A relatively intense spot at acidic pH in dsNBH appears to be due to aggregation of PrPC at low pH. Nevertheless, this experiment illustrates that PrPC charge heterogeneity is attributable at least in part to its variable sialylation status. Consistent with a previous study [20], the diglycosylated form of PrPC in dsNBH migrated slightly faster on a 1D gel than that in non-treated NBH (Figure 1D).
Brain-derived, proteinase K (PK)-treated scrapie material from animals infected with the strains of natural or synthetic origin 263K or SSLOW [28], respectively, also showed broad charge heterogeneity on 2D gels (Figure 1B,C). When compared to the 2D profile of PrPC, the charge distributions of PK-treated 263K and SSLOW were found to shift toward pI 10, despite an expected shift toward acidic pH due to proteolytic cleavage of the positively charged N-terminal region. The reason behind such a shift is difficult to explain. There is a possibility that PrPSc is less sialylated than PrPC, although no notable differences in sialylation status of PrPC and PrPSc were found in previous study [9]. Alternatively, a fraction of PrPC molecules could be subjected to posttranslational modifications including deamidation of Asn and Gln to Asp and Glu [29], [30], respectively, phosphorylation of serine 43 [15], or modification of amino groups of Lys and Arg by reducing sugars resulting in advanced glycation end-products [5], [31]. Such modification would increase PrPC charge heterogeneity and account for spreading PrPC to acidic pH on 2D. Attempts to remove sialic acid in PrPSc by treatment with sialidase were not successful, presumably due to high aggregation status of PrPSc (data not shown).
Because previous studies showed that properties of PrPSc change during PMCA [32], [33], we decided to compare charge distribution of PMCAb-derived and brain-derived PrPSc. To rule out any interference of the initial brain-derived PrPSc seeds, twenty four serial PMCAb rounds were performed with a dilution 1∶10 between rounds to produce PMCAb-derived material. Surprisingly, both 263K and SSLOW PMCAb-derived materials showed a considerable shift towards basic pI when compared to that of brain-derived PrPSc (Figure 1B,C). Moreover, consistent with the previous study [32], the percentage of monoglycosylated glycoforms decreased in PMCAb-derived material comparing to those of brain-derived PrPSc. These results suggest that (i) undersialylated PrPC molecules are selected during in vitro amplification at the expense of overersialylated PrPC (sialylated more than the statistical average for PrPC) and (ii) a decrease in PrPSc sialylation level reduces the negative charge on PrPSc surfaces that might lead to an increase in percentage of diglycosyated molecules incorporated into PMCAb-derived material.
To provide independent support that undersialylated PrPC is a preferable substrate for PrPSc amplification in vitro, we tested whether desialylation of PrPC increases the rate of amplification using two alternative PMCAb formats. dsNBH prepared by treatment of NBH with A.ureafaciens sialidase (Figure 1A) was used as a substrate in PMCAb along with 10% non-treated NBH. In the first format, increasing dilutions of brain-derived 263K, Hyper, Drowsy and SSLOW materials were subjected to a single round of PMCAb conducted in dsNBH or NBH. The range of seed dilutions was chosen individually for each strain according to the previously published results [34]. For the hamster-adapted strains of natural origin (263K, Hyper (HY), Drowsy), the reactions conducted in dsNBH detected approximately 10-fold higher seed dilutions than the reactions conducted in NBH (Figure 2A). Surprisingly, for the synthetic strain SSLOW, the reaction conducted in dsNBH detected 104-fold higher seed dilutions than the reactions conducted in NBH (Figure 2A). In fact, in dsNBH 108-fold diluted SSLOW brain material was persistently detected in a single PMCAb round. For two other strains of synthetic origin LOTSS and S05 [35], [36], the amplification rate also increase by at least four orders of magnitude in dsNBH compared to that in NBH (data not shown).
In a second format, a set of serial PMCAb reactions were conducted for 263K or SSLOW in NBH or dsNBH with the dilution folds between serial rounds ranging from 1∶30 to 2∶108. The amplification rate is defined operationally as the highest dilution between PMCAb rounds at which amplification was still capable of compensating for the effect of dilution [34]. For 263K, the amplification rate increased approximately 50 fold, from 100-fold in NBH to 5000-fold in dsNBH (Figure 2B). For SSLOW, the amplification rate increased more than 5×105 fold, from approximately 100-fold in NBH to at least 5×107-fold in dsNBH (Figure 2B). Consistent with previous studies [32], [34], SSLOW showed an increase in signal intensity in serial PMCAb suggesting that it undergoes fast ‘adaptation’ to the PMCAb environment.
Both experimental formats showed that desialylation of PrPC increases the rate of prion amplification in PMCAb, while the magnitude of an increase was strain-dependent. This effect was considerably higher for synthetic strains than strains of natural origin.
In previous studies prion amplification in PMCA was shown to mimic key features of prion transmission including the species barrier [37]–[39]. A drop in amplification efficiency in PMCAb seeded with heterologous PrPSc will be referred to as the cross-seeding barrier. Considering that desialylation of PrPC increases the rate of PrPSc amplification, we were interested to test whether desialylation of PrPC eliminates the cross-seeding barrier in PMCAb. When serial PMCAb reactions in mouse NBH were seeded with hamster strains 263K or HY, stable replication was observed only after four or six serial rounds, respectively, a sign of a significant cross-seeding barrier (Figure 3A). However, when mouse dsNBH was used as a substrate for 263K and HY, stable amplification was observed starting from the first round for both strains with no signs of cross-seeding barrier (Figure 3A).
In a reverse transmission experiment, two mouse strains 22L and ME7 were subjected to amplification in hamster NBH. No signal was observed for ten serial rounds suggesting that these two strains could not cross the barrier under the current experimental conditions (Figure 3B). Surprisingly, when 22L and ME7 were subjected to amplification in hamster dsNBH, stable amplification was observed starting from the third or fourth serial round, respectively (Figure 3B). Both experiments show that reducing sialylation levels of PrPC of a host species helps to eliminate or significantly reduce the barrier that prevents cross-seeding.
Careful comparison of mouse-adapted 263K or HY revealed that the relative ratio of diglycosylated vs. monoglycosylated glycoforms was higher in dsNBH-amplified products than in NBH-amplified products (Figure 3A,C). These changes suggest that the ratio of di-, mono- and unglycosylated glycoforms is not only a function of prion strain or host, but also depends on PrPC sialylation status. Due to glycan sialylation, the surface of PrPSc particles has a dense negative charge that creates electrostatic repulsion and, presumably, limits the percentage of diglycosylated PrPC able to be recruited by PrPSc. We propose that desialylation of PrPC eliminates electrostatic repulsion permitting a higher percentage of the diglycosylated glycoform to be accommodated.
To test the hypothesis that PrPC sialylation controls the ratio of glycoforms within PrPSc, we examined the glycosylation profile of two mouse strains 22L and ME7 after their amplification in mouse NBH or dsNBH. Mouse strains were chosen because, in contrast to hamster strains, they have equal or even slightly higher percentage of monoglycosylated form relative to that of diglycosylated form. Upon amplification in dsNBH, the glycosylation profile of both 22L and ME7 changed immediately from predominantly monoglycosylated to predominantly diglycosylated (Figure 4A). For comparison, the glycosylation profile of NBH-amplified 22L PMCAb-derived material remained very similar to that of brain-derived 22L (Figure 4B). Due to very low amplification efficiency of ME7 in NBH (Figure 4A), it was difficult to compare the glycosylation profile of ME7 NBH- versus dsNBH-amplified material directly. Nevertheless, after adjusting the amount of material loaded on a gel, ME7 amplified in dsNBH showed a considerably higher ratio of di- versus mono- or unglycosylated glycoforms than those observed in ME7 NBH-amplified or brain-derived material (Figure 4B). Similar trend was observed for both hamster strains tested (263K and SSLOW). Monoglycosylated glycoform was well represented in brain-derived and in lower proportion in PMCAb-derived material, but absent in PMCAb-derived material produced in dsNBH (Figure 4C).
Taken together these results indicate that (i) the glycoform ratio within PrPSc is not only controlled by the strain or host but also by the sialylation status of PrPC; (ii) a decrease in PrPC sialylation levels results in a higher percentage of diglycosylated glycoforms in PrPSc; (iii) a shift toward diglycosylated glycoforms appears regardless of the host species, however the extent of the shift is likely determined by the strain-specific conformation.
The lymphoreticular system plays an important role in prion pathogenesis, because: 1) prion replication in lymphoid tissues precedes neuroinvasion [40], [41], 2) lymphoid organs are targeted upon cross-species transmission and appear to be more permissive than central nervous system [42], and 3) inflammation facilitates prion invasion [43], [44]. Considering that desialylation of PrPC facilitates PrPSc replication, we were interested in comparing the sialylation pattern of brain-derived and spleen-derived PrPC.
The 2D analysis of spleen tissues was very challenging because the level of PrPC expression in the spleen is 20 to 50-fold lower than that in the brain. Moreover, most of spleen-derived PrPC was proteolytically processed and formed a C2 fragment (residues ∼100–231) that was immunoreactive with 3F4 (epitope 109–112) and SAF-84 (epitope 160–170) antibodies, but not Ab3531 (epitope 90–102) antibody (Figure S1). This finding was in agreement with a previous report on N-terminal trimming of spleen-derived prion protein [45]. In addition, we found that the spleen-derived C2 fragment is highly prone to aggregation, as a significant proportion of it appeared as a dimer on SDS-gels (Figure S1). For the above reasons, 2D analysis of brain- and spleen-derived PrPC was performed using the SAF-84 antibody that reacts with full-length PrPC as well as C1 (residues 111–231) and C2 fragments (Figure 5).
Consistent with 1D SDS gel analysis, a significant percentage of full-length PrPC and the C2 fragment were persistently seen as dimers on 2D gels of spleen tissues (Figure 5A). Pretreatment of spleen homogenates with sarcosyl did not help to reduce the amount of dimers. Nevertheless, a notable difference could be observed with respect to the charge distribution between spleen- and brain-derived monomeric full-length PrPC (Figure 5A,B). In spleen material, the distributions of charge isoforms were shifted toward basic pH for both full-length PrPC and the C2 fragment. While the details of sialylation of spleen-derived PrPC remain to be investigated, this result suggests that the metabolism of PrPC sialylation in spleen might be different from that in a brain.
To test whether PMCAb-associated changes in sialylation status affect prion infectivity, Syrian hamsters were inoculated with 263K brain-derived and PMCA-derived materials produced using NBH or dsNBH. We also attempted to produce desialylated PrPSc directly by treating brain-derived 263K with sialidase. However, this approach was not successful. For PMCAb conducted in NBH, 103-fold diluted 263K brain material was subjected to 24 serial rounds with 1∶10 dilution between rounds that resulted in a final dilution of brain material of 10−27. For PMCAb conducted in dsNBH, 105-fold diluted 263K brain material was subjected to 7 serial rounds with 1∶1000 dilutions between rounds that resulted in a final dilution of brain material of 10−26. On 2D gels, the charge distribution of products generated in PMCAb with dsNBH showed substantially more significant shift toward pI 10 and very little charge heterogeneity in comparison to the brain-derived or PMCAb-derived materials (Figure 6A).
The incubation time to the clinical signs of disease was 83±2 and 106±12 days in groups inoculated with brain-derived 263K and PMCAb-derived material, respectively. Two animals inoculated with PMCAb-derived material produced in dsNBH were sacrificed at 223 days p.i., and no PK-resistant material was found in their brains (Figure S2A). The remaining four animals from this group did not show any clinical signs and were sacrificed at 341 days p.i. No PK-resistant products were detected in their brain using 3F4 or SAF-84 antibodies (Figure S2B). This result suggests that in contrast to the material produced in PMCAb with NBH, the material produced in PMCAb with dsNBH had no detectible infectivity despite being subjected to fewer PMCAb rounds.
Lack of infectivity of PMCAb material produced in dsNBH could be due to changes in physical properties, particularly the high sensitivity to proteolytic clearance. To examine proteolytic resistance, PMCAb products generated in NBH or dsNBH were treated with increasing concentrations of PK (Figure 6B). Indeed, for both strains 263K and SSLOW, the products formed using desialylated substrate were substantially less resistant than PMCAb products produced in NBH (Figure 6B). Increased proteolytic sensitivity could lead to faster clearance of such inocula.
In previous studies, we showed that 263K gave rise to a novel PrPSc conformation referred to as 263KR+ after 263K brain material was subjected to 12 serial PMCAb rounds in RNA-depleted NBH and then to additional 14 PMCAb rounds in NBH [46]. While 263KR+ amplified very fast in vitro, no clinical disease was observed upon inoculation of 263KR+ in first and second serial passages. We were interested in testing whether lack of infectivity could be due to changes in sialylation status of 263KR+. 2D analysis revealed that the distribution of 263KR+ charge isoforms was considerably shifted to the basic pI in comparison to that of PMCAb-derived 263K (Figure 6A). Therefore, for both 263KR+ and PMCAb-derived material produced in dsNBH the lack of infectivity correlates well with their low sialylation status.
The catabolism of sialoglycoconjugates is regulated by four sialidases (also referred to as neuraminidases Neu1, Neu2, Neu3 and Neu4), all of which catalyze the removal of terminal sialic acid residues from carbohydrates of glycoprotein or glycolipids [47]. While all four sialidases are expressed in neuronal tissues, their levels and subcellular localization differ greatly. Because Neu1 is the most abundant, is expressed in the brain and is localized in lysosomes and on the cell surface, we decided to examine its role in regulating PrPC sialylation status. Brain materials from Neu1−/−, Neu1+/− and wild type mice of two genetic backgrounds (FVB and BL6) were compared using 2D analysis. No notable differences with respect to PrPC charge distribution were observed between wild type, Neu1−/− or Neu1+/− mice of the two groups (Figure S3). The lack of effect could be because (i) Neu1 is not involved in PrPC desialylation, (ii) Neu1 deficiency is compensated by other neuraminidases, or (iii) PrPC molecules are degraded very fast in lysosomes, so the relative contribution of desialylated PrPC in the total pool of PrPC is very small. To probe the possible role of Neu1 in PrP catabolism further, we also examined the sialylation profile of PrP proteolytic fragment C1 (residues ∼111–231) using SAF-84 antibody. C1 is present in large amounts in mouse brain and similar to PrPC, C1 can be found in di-, mono- and unglycosylated forms; however, the dynamics of the cellular clearance of C1 and its cellular localization could be different from those of full-length PrPC. No differences in sialylation status of C1 fragments were found between Neu1−/−, Neu1+/− and wild type mice (Figure S3). Because the life-span of Neu1−/− mice is limited to ∼150 days, it was not possible to test whether knocking out Neu1 affected the incubation time to prion disease.
Sialic acids are the most abundant terminal monosaccharides in cell membrane glycans [21], [22]. Sialylation plays an essential role in key cellular functions including cell signaling, adhesion, differentiation, neuronal plasticity, cell-cell and cell-pathogen recognition, and the activation and trafficking of B and T lymphocytes, among other things [21], [22]. The sialic acid content is the highest in embryonic and perinatal phases, but drops gradually during adulthood [48], [49]. Brain and immune tissues including spleen have considerably higher amounts of sialic acid in their membrane fraction than other organs such as heart or kidney [48].
In the current study we showed that de-sialylation of PrPC increases PrPSc amplification rates in PMCAb. Faster amplification of desialylated substrates was likely due to removal of electrostatic repulsion between glycan moieties, which can carry up to 4 negatively charged sialic acid residues each (Fig. 7A,B). This work suggests that a dense negative charge on the surface of PrPSc particles due to sialylated glycans prevents efficient PrPSc replication. Previous studies revealed that partial removal of N-linked glycans from PrPC using treatment with PNGase F or replacing a mixture of di-, mono- and unglicosylated PrPC with only unglicosylated form have a negative effect on PrPSc amplification in PMCA [50], [51]. Taken together, these results indicate that while glycans are important for efficient amplification of PrPSc, the terminal sialic acid residues have a negative impact.
Notably, the positive effect of PrPC desialylation on the replication rate, while present in all prion strains probed in this study, differed considerably in magnitude. Approximately 10–50 fold increases in the rate for strains of natural origin (263K, HY, Drowsy) was strikingly different from the 10,000-fold increase for the strains of synthetic origin. Such drastic difference between strains of the two classes indicate that steric clashes between sialic acid residues in neighboring glycans are much more substantial in synthetic strains than in strains of natural origin (Figure 7A,B). The elimination of the negative charges from PrPSc surface led to a much more significant drop in polymerization energy costs for synthetic strains than natural strains. This result highlights structural differences between the two classes of prion strains.
The hypothesis that electrostatic repulsion between sialic residues controls PrPSc amplification rate explains why undersialylated PrPC molecules are preferentially recruited during in vitro amplification at the expense of oversialylated PrPC. As a result, the sialylation status of PrPSc changes during PMCAb becoming less sialylated in comparison to brain-derived PrPSc. In previous studies, prion specific infectivity (the ratio of the infectivity titer to the amount of PrPSc) was found to decrease gradually during amplification in serial PMCA [52]. A decrease in specific infectivity correlates well with a drop in sialylation status of PMCAb-derive material observed here. In support of this correlation, the current study observed longer incubation time to disease for PMCAb-derived material relative to that of brain-derived PrPSc, and a lack of clinical signs for PMCAb-derived material produced using desialylated substrate. Recent studies reported a gradual change in strain-specific secondary structure during serial amplification in PMCAb [33]. The relationship between changes in PrPSc conformation and sialylation status during serial PMCAb is not clear. Nevertheless, the fact that PMCAb material produced in NBH caused prion disease after 24 PMCAb rounds, whereas PMCAb material amplified in dsNBH did not cause the disease after only 7 amplification rounds, suggests that it is the loss of sialylation rather than a number of PMCAb rounds that had deleterious effect on infectivity.
Progression of prion diseases is determined by a number of factors including PrPC-to-PrPSc conversion rate, PrPSc clearance, PrPSc deposition sites, and relative toxicity and size of PrPSc aggregates. Lack of clinical disease in animals inoculated with desialylated PMCAb products can be attributed in part to their high clearance rate. Consistent with this hypothesis, PMCAb materials produced with de-sialylated substrate were more sensitive to proteolytic digestion than standard PMCAb-derived material. Alternative mechanisms that involve interaction with microglia and cells of the immune system might also contribute to the lack of infectivity of PMCAb material produced in dsNBH. The mammalian immune system uses terminal sialylation of cell surface glycoproteins to identify pathogenic microorganisms to set them apart from their own cells, as microorganisms generally lack enzymes essential for sialic acid synthesis [21], [22]. In the absence of terminal sialic residues, galactose is exposed as the terminal residue of glycans of microbial glycoproteins and serves as a signal for activating the immune response and phagocytotic clearance by macrophages [21], [22]. If clearance of PrPSc involves mechanisms that are involved in clearance of microorganisms, desialylated PrPSc should be cleared much faster than sialylated PrPSc. Notably, some microbial pathogens recruit sialic acid from the host and sialylate their own glycoproteins in order to become “invisible” to the host's immune systems [23]. In addition to the clearance by macrophages, recent studies revealed that glycoclusters with terminal sialic acid were stable upon injection into mice and accumulated in the spleen, while the same clusters without sialic acid residues were rapidly excreted via the urinary tract [53]. It remains to be determined whether any of the above mechanisms account for lack of infectivity of desialylated PMCAb material. Nevertheless, fast clearance of PrPSc with low level of sialylation in a brain and luck of such clearance in PMCAb could explain why the sialylation levels of brain-derived and PMCAb-derived PrPSc are different.
In previous studies, prions with high infectivity titers that lacked sialylation were generated in vitro using recombinant PrP [54], [55]. Because entire carbohydrate groups were missing in PrPSc produced from recombinant PrP, it is unlikely that the immune system and microglia can identify these synthetic prions as potential pathogens in the same manner as it deals with desialylated PMCAb products. Consistent with this hypothesis, scrapie brain material from transgenic mice deficient in PrP glycosylation at both sites was found to be capable of infecting wild type mice [56]. Notably, transgenic mice that lacked glycosyls at both sites displayed a dramatic increase in the incubation time, incomplete attack rate or lack of infection [56]. These results suggest that in the absence of glycosylation/sialylation, PrPC of the host does not support well the infection or the newly formed PrPSc is not toxic.
That PrPC sialylation controls PrPC-to-PrPSc conversion rate has far reaching implications. A decrease in PrPC sialylation could lead to a dramatic plunge of PrPC-to-PrPSc barriers in vivo and provide favorable conditions for (i) lowering the energy barrier of the spontaneous PrPC-to-PrPSc conversion in sporadic prion diseases; (ii) successful infection of a host or tissues with abnormally low sialylation status by low prion doses; and (iii) crossing the species barrier. In support of the last hypothesis, the current study revealed that hamster-to-mouse or mouse-to-hamster cross-seeding barriers can be reduced or abolished entirely in PMCAb, if de-sialylated PrPC is used as a substrate. While PMCAb can not predict the outcomes of transmission species barrier effects in whole organisms, our work opens an intriguing possibility that a species barrier is not only controlled by PrP amino acid sequence and PrPSc strain-specific structure but also by PrPC sialylation status. Noteworthy, because of an irreversible mutation in the gene encoding human N-acetylneuraminic acid hydroxylase, humans and the rest of mammalian species use different sialyc acid residues: humans produce only N-acetyl neuraminic acid (Neu5Ac), while other mammals produce Neu5Ac and N-glycolylneuraminic acid (Neu5Gc) [57]. The difference in sialic acid structure affects interaction of pathogenic microbes with the immune systems of humans and other mammalian species. This difference might also contribute to a previously unappreciated mechanism that controls the prion transmission barrier between mammals and humans.
Lymphoid organs are targeted upon cross-species transmission and appear to be more permissive than central nervous system [42]. 2D analysis of PrPC charge distribution revealed that spleen-derived PrPC was different from that of brain-derived PrPC with respect to their sialylation pattern, although precise comparison of the two tissues was complicated because of the low expression level, high tendency for aggregation and formation of the C2 proteolytic fragment by spleen-derived PrPC. Further research is needed more sensitive and accurate tools to confirm whether hyposialylation of PrPC in spleen makes the spleen more susceptible to prion infection than brain. Noteworthy, endogenous sialidase activity was found to increase in cells of the immune system, including lymphocytes and monocytes during cell activation and differentiation leading to undersialylation of cell surface glycoproteins [58], [59]. Because inflammatory conditions support prion replication [43], [44], it would be interesting to examine in future studies whether inflammation-induced activation of sialidase gives rise to undersialylated PrPC and facilitates prion infection.
Sialylation status of glycoproteins is controlled by sialyltransferases and sialidases (also called neuraminidases), two classes of enzymes that transfer or cleave terminal sialic acids to/from glycoproteins, respectively [60]. In mammals, there are four sialidases (Neu1, Neu2, Neu3 and Neu4) that are expressed in a tissue-dependent manner and differ with respect to their cellular localization and enzymatic properties [60]. Among the four sialidases, Neu1 is the most abundant and ubiquitously distributed. It is a part of a multi-enzyme, 1200 kDa hydrolase complex, which is localized predominantly in lysosomes and to lesser extent on the cell surfaces of many tissues and organs including brain [61]. To test whether Neu1 is responsible for desialylation of PrPC, brain materials from Neu1-/- knockout and Neu1+/− mice generated in two genetic backgrounds (FVB and BL6) were analyzed using 2D gels. No significant changes in PrPC sialylation patterns were observed in Neu1−/− or Neu1+/− in comparison to those of corresponding wild type mice (Figure S1). These results indicate that either Neu1 is not involved in PrPC desialylation or Neu1 deficiency is compensated by other neuraminidases. Alternatively, PrPC molecules might be degraded so fast following desialylation, that the relative contribution of desialylated PrPC in the total pool of PrPC is too small to be detected by the current approach. In either case, Neu1 might not be the right target if one wants to alter the sialylation status of PrPC in vivo for therapeutic intervention.
The ratio of di-, mono- and unglycosylated glycoforms within PrPSc is believed to be an intrinsic property of a prion strain or PrPSc subtype (in sporadic prion diseases) [62]–[65]. As such, the PrPSc glycoform ratio is used widely for strain typing and classification of PrPSc subtypes in sporadic CJD [64], [65], and changes in the glycoform ratio are thought to be indicative of a strain mutation or strain adaptation to a new host or environment [66], [67]. Surprisingly, the current study revealed that the PrPSc glycoform ratio is not only controlled by prion strain or the host, but also by the sialylation status of PrPC. A decrease in PrPC sialylation levels resulted in a shift of the glycoform ratio toward diglycosylated forms at the expenses of mono- and unglycosylated glycoforms for both mouse and hamster strains. Such a relationship is explained well by the model that postulates that electrostatic repulsion created by sialic acid residues on the surface of PrPSc particles limits the percentage of diglycosylated molecules that can be accommodated within PrPSc (Figure 7C). A decrease in sialylation levels reduces electrostatic repulsion leading to an increase in the percentage of diglycosylated molecules. In previous studies, treatment of prion-infected cultured cells with swainsonine, a compound that blocks synthesis of complex N-linked glycans, was shown to select minor strain variants or “mutants” resistant to swainsonine [68]. This process was accompanied by a change in the PrPSc glycoform ratio in favor of diglicosylated forms at the expense of monoglycosylated PrPSc glycoforms [68], [69]. The extent to which swainsonine-related selection of minor variants and changes in glycoform ratio were due to lack of sialic acid residues is unclear.
Recent studies established a possible link between protein sialylation and Alzheimer's diseases [70]. Deficiency of the lysosomal sialidase Neu1 was found to lead to the spontaneous occurrence of an Alzheimer's disease-like amyloidogenic process in mice. Loss of Neu1 resulted in accumulation of an over-sialylated amyloid precursor protein in lysosomes and excessive release of Aβ peptides by lysosomal exocytosis [70].
The current study opens a new avenue in prion research that might shed new light on the mechanism of prion replication and contribute to development of new therapeutic approaches. A number of sialic acid metabolic precursors or sialidase inhibitors are currently available and approved by FDA. Nevertheless, the impact and effectiveness of pharmacological intervention that target PrPC sialylation on progression of prion diseases are difficult to predict. The sialylation status of PrPC does not only control the rate of prion replication and magnitude of the species barrier, but is also likely to affect prion uptake and transport by macrophages, prion clearance rate, toxicity of PrPSc particles, and interaction of PrPSc with cells of the immune system and microglia. These topics have to be investigated in future studies.
In summary, the current work demonstrated that hyposialylated PrPC molecules are a preferable substrate for prion amplification in PMCAb. PMCAb-derived PrPSc is less sialylated than brain-derived PrPSc. De-sialylation of PrPC significantly speeds up PrPSc amplification in a strain-dependent manner and significantly reduces or eliminates the species barrier. A decrease in PrPSc sialylation correlates with a drop in infectivity of PMCAb-derived material. The sialylation status of brain-derived PrPC appears to differ from that of spleen-derived PrPC. The sialylation status of PrPC controls the PrPSc glycoform ratio with a decrease in PrPC sialylation levels resulting in a higher percentage of the diglycosylated glycoform in PrPSc. Knocking out lysosomal sialidase Neu1 does not change the sialylation status of PrPC. The current work highlights the previously unappreciated role of PrPC sialylation in prion diseases and opens new directions in prion research, including development of new therapeutic approaches.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Institutional Animal Care and Use Committee of the University of Maryland, Baltimore (Assurance Number A32000-01; Permit Number: 0309001).
Hyper and Drowsy scrapie brain materials were kindly provided by Richard Bessen (Colorado State University, Fort Collins, CO); 263K, 22L and ME7 scrapie brain materials were kindly provided by Robert Rohwer (Veterans Affair Maryland Health Care System, Baltimore, MD); SSLOW scrapie brain homogenate was prepared using animals from the 4th passage of SSLOW [71]. Neu1−/−, Neu1+/− and wild type mouse brains were collected from four month old FVB mice and five month old BL6 mice [72]. The mice were perfused with 20 ml PBS/5mM EDTA (pH 7.4), and then brains were collected and frozen in liquid nitrogen.
Weanling Golden Syrian hamsters (all males) were inoculated intracerebrally under 2% O2/4 MAC isoflurane anesthesia. Each animal received 50 µl of brain homogenate or PMCAb products. After inoculation, hamsters were observed daily for disease using a ‘blind’ scoring protocol. Hamsters were euthanized as they approach the terminal stage of the disease.
10% normal brain homogenate (NBH) from healthy hamsters was prepared as described previously [35] and used as a substrate for PMCAb [39]. The sonication program consisted of 20 sec sonication pulses delivered at 170W energy output applied every 20 min during a 24 hour period. For each subsequent round, 10 or 20 µl of the reaction from the previous round were added to 90 or 80 µl of fresh substrate, respectively. Each PMCAb reaction was carried out in the presence of two 2/32” Teflon beads (AmazonSupply.com).
To analyze production of PK-resistant PrP material in PMCAb, 10 µl of sample were supplemented with 5 µl SDS and 5 µl PK, to a final concentration of 0.25% SDS and 50 µg/ml PK, followed by incubation at 37°C for 1 hour. The digestion was terminated by addition of SDS sample buffer and heating the samples for 10 min in a boiling water bath. Samples were loaded onto NuPAGE 12% BisTris gels, transferred to PVDF membrane, and probed with SAF-84, Ab3531 or 3F4 antibodies.
To produce de-sialylated substrate, 10% NBH from healthy hamsters prepared for PMCAb was treated with Arthrobacter ureafaciens sialidase (cat # N3786, Sigma-Aldrich, St. Louis, MO) as follows. The lyophilized enzyme was dissolved in MilliQ water to the final concentration of 500mIU/ml. After preclearance of NBH at 500 × g for 2 min and addition of the buffer supplied by manufacturer, 7mIU/ml sialidase were added to the supernatant, and the reaction was incubated on a rotator at 37 °C for 5 h. The resulting substrate was used in dsPMCAb using the sonication protocol described for PMCAb. To prepare mock sialidase treated PMCAb substrate, the procedures were the same with adding MilliQ water instead of sialidase solution.
Samples of 10 µl volume prepared in 1xSDS sample loading buffer as described above were solubilized for 1h at room temperature in 80 µl solubilization buffer (8M Urea, 2% CHAPS, 5mM TBP, 20mM Tris pH 8.0), alkylated by addition of 135 µl 0.5M iodoacetamide and incubated for 1h at room temperature. Then, 1150 µl of ice-cold methanol was added, and samples were incubated for 2h at −20°C. After centrifugation at 13,000 rpm and 4°C, supernatant was discarded and the pellet was re-solubilized in 160 µl rehydration buffer (7M urea, 2 M thiourea, 1%DTT, 1% CHAPS, 1% Triton X-100, 1% ampholyte, trace amount of Bromophenol Blue). Fixed immobilized pre-cast IPG strips (cat. # ZM0011, Life Technologies, Carlsbad, CA) with a non-linear pH gradient 3–10 were rehydrated in 155 µl of resulting mixture overnight at room temperature inside IPGRunner cassettes (cat. # ZM0008, Life Technologies, Carlsbad, CA). Isoelectrofocusing (first dimension separation) was performed at room temperature with rising voltage (175V for 15 minutes; 175–2,000V linear gradient for 45 minutes; 2,000V for 30 minutes) on Life Technologies Zoom Dual Power Supply using the XCell SureLock Mini-Cell Electrophoresis System (cat. # EI0001, Life Technologies). The IPG strips were then equilibrated for 15 minutes consecutively in (i) 6 M Urea, 20% glycerol, 2% SDS, 375mM Tris-HCl pH 8.8, 130mM DTT, and (ii) 6 M Urea, 20% glycerol, 2% SDS, 375mM Tris-HCl pH 8.8, 135mM iodoacetamide, and loaded on 4–12% Bis-Tris ZOOM SDS-PAGE pre-cast gels (cat. # NP0330BOX, Life Technologies). For the second dimension, SDS-PAGE was performed for 1h at 170V. Immunoblotting was performed as described above.
|
10.1371/journal.ppat.1007224 | Molecular basis for CesT recognition of type III secretion effectors in enteropathogenic Escherichia coli | Enteropathogenic Escherichia coli (EPEC) use a needle-like injection apparatus known as the type III secretion system (T3SS) to deliver protein effectors into host cells. Effector translocation is highly stratified in EPEC with the translocated intimin receptor (Tir) being the first effector delivered into the host. CesT is a multi-cargo chaperone that is required for the secretion of Tir and at least 9 other effectors. However, the structural and mechanistic basis for differential effector recognition by CesT remains unclear. Here, we delineated the minimal CesT-binding region on Tir to residues 35–77 and determined the 2.74 Å structure of CesT bound to an N-terminal fragment of Tir. Our structure revealed that the CesT-binding region in the N-terminus of Tir contains an additional conserved sequence, distinct from the known chaperone-binding β-motif, that we termed the CesT-extension motif because it extends the β-sheet core of CesT. This motif is also present in the C-terminus of Tir that we confirmed to be a unique second CesT-binding region. Point mutations that disrupt CesT-binding to the N- or C-terminus of Tir revealed that the newly identified carboxy-terminal CesT-binding region was required for efficient Tir translocation into HeLa cells and pedestal formation. Furthermore, the CesT-extension motif was identified in the N-terminal region of NleH1, NleH2, and EspZ, and mutations that disrupt this motif reduced translocation of these effectors, and in some cases, overall effector stability, thus validating the universality of this CesT-extension motif. The presence of two CesT-binding regions in Tir, along with the presence of the CesT-extension motif in other highly translocated effectors, may contribute to differential cargo recognition by CesT.
| Enteropathogenic Escherichia coli injects effector proteins into host cells using a type III secretion system (T3SS). The translocated intimin receptor (Tir) is the first effector delivered into host cells and imparts efficient secretion of other effectors. However, the mechanism for Tir-dependent modulation of the T3SS is poorly understood. We provide evidence that the multi-cargo chaperone CesT binds to two regions in Tir at the N- and C-terminus through a specific recognition motif, and show that CesT binding to the Tir C-terminus is important for host translocation. Furthermore we show that the CesT-specific motif is conserved in a subset of highly translocated effectors. This study highlights the multi-faceted role that T3SS chaperones play in effector secretion dynamics.
| Enteropathogenic and enterohemorrhagic Escherichia coli (EPEC and EHEC) cause acute gastroenteritis in humans and are a common source of outbreaks [1]. EPEC is a significant pathogen in the pediatric population, especially in areas with limited access to healthcare and clean water, whereas EHEC is a common food- or water-borne contaminant in industrialized nations [1]. EPEC and EHEC contain a genomic island called the locus of enterocyte effacement (LEE) that encodes a type III secretion system (T3SS) [2] necessary for the formation of attaching and effacing (A/E) lesions on epithelial cells [3]. The T3SS is a needle-like protein injectisome used by Gram-negative bacteria to deliver effector proteins into host cells directly from the bacterial cytosol [4], where they target specific host processes to allow for attachment, survival, and propagation of the bacteria [5, 6]. Enteric pathogens that use a T3SS for host attachment, infection, and/or colonization are significantly attenuated when lacking their encoded T3SS [7], identifying it as a key mediator of host-pathogen interactions.
Structural biology efforts have advanced our understanding of the assembly, structure, and function of the T3SS [8, 9]. The T3SS contains ~25 proteins assembled into distinct structures including, (i) an extracellular needle filament capped at the distal end by hydrophobic translocon proteins, (ii) a basal body comprised of inner and outer membrane-spanning rings, (iii) an ATPase-containing sorting platform complex at the cytoplasmic face of the basal body, and (iv) cytosolic chaperones that bind, protect, deliver, and control effector secretion [10]. Three classes of non-flagellar T3SS chaperones have been described. Class I chaperones bind translocated effectors, class II chaperones bind the hydrophobic translocators, and class III chaperones escort and prevent cytosolic polymerization of the extracellular needle filament [11, 12]. The class I chaperones are further subdivided into class IA and IB. Class IA chaperones are usually specific for one effector and are located adjacent to the gene that encodes the cognate effector [11, 13]. Class IA chaperones that bind multiple effectors have been reported, including EPEC CesT and Salmonella SrcA [14, 15], and are referred to as multi-cargo chaperones [16]. Class IB chaperones bind multiple effectors and are usually encoded within large operons that contain structural components of the T3SS instead of being adjacent to a specific effector gene [11, 13]. Class IB chaperones appear to be functionally interchangeable between species and recognize a specific sequence motif [17].
Multi-cargo chaperones play a significant role in T3SS-dependent infection biology as mutants lacking these proteins are attenuated in animal and plant models of infection [16]. CesT from EPEC and EHEC was originally thought to be a class 1A chaperone specific for the translocated intimin receptor (Tir) [18, 19]. However, it was later reclassified as a multi-cargo chaperone because it interacts with at least 9 other effectors [14, 20, 21], most of which require CesT for translocation into host cells [22, 23]. Recent work has indicated that the effector binding and secretion activities mediated by CesT can be functionally separated. For example, mutants in the C-terminal domain of CesT retain their ability to bind effector cargo, yet exhibit reduced effector secretion [24]. This C-terminal domain was also identified as a site for tyrosine phosphorylation in a phosphotyrosine-proteome study [25], in which tandem tyrosine phosphosites (Y152 and Y153) influenced NleA or global effector secretion, respectively [26]. Furthermore, host-cell contact has been proposed to liberate free CesT in the bacterial cytosol that can then bind and antagonize CsrA repression of the nleA 5’UTR [27]. This interaction is facilitated by the C-terminal domain of CesT [28], allowing for greater control over the timing and translocation efficiency of the NleA effector. Notwithstanding the requirement of CesT for effector secretion, Tir has been implicated in effector secretion hierarchy. Deletion of tir in the hyper-secreting ΔsepD strain of EPEC significantly reduced the level of at least 6 effectors in culture supernatants [21]. A similar but more modest effect was also seen for this subset of effectors translocated into host cells when only tir was deleted [23].
Given the primary importance of the Tir-CesT complex in orchestrating secondary effector secretion in E. coli, we initiated structural studies to characterize the Tir-CesT interaction and to delineate the role that this effector-chaperone pair plays in protein translocation. Here, we present the co-crystal structure of a C-terminal truncation of CesT in complex with an N-terminal fragment of Tir. This structure allowed us to define a CesT-extension motif, leading to the identification of a second CesT-binding region in the C-terminal domain of Tir, which we verified using biochemical and molecular assays. Furthermore, we identified the CesT-extension motif in the N-terminus of a subset of other effectors and demonstrated the function of this motif in effector translocation.
The first ~20 amino-terminal residues of E. coli T3SS effectors contain a T3SS-specific secretion signal that can be predicted bioinformatically [29]. Downstream of the T3SS secretion signal, but within the first ~100 residues, is an unspecified CesT-binding domain that has been identified in Tir, Map, and NleH [18, 20, 21]. Despite the fact that CesT binds to the N-terminus of these effectors, sequence alignments have not identified a consensus motif within this region. To determine the minimal recognition sequence of the CesT-binding region, various His6-tagged N-terminal Tir constructs were tested for their ability to co-purify CesT (Fig 1). Tir fragments containing residues 23–80, 32–80, and 35–77 co-purified CesT as seen by Ni2+-affinity pull-down and immunoblotting (Fig 1B and S1A Fig), whereas CesT alone was never pulled-down in the absence of Tir by the Ni2+-affinity resin, thus confirming specificity of our assay. When these Tir fragments were truncated further to residues 32–73, 37–80, and 37–73, they lost the ability to co-purify with CesT (Fig 1B). To determine the molecular basis of the Tir-CesT interaction, we carried out crystallization trials with the three Tir fragments that co-purified CesT (S1A Fig), however none of the complexes produced crystals. CesT contains a unique C-terminal extension that is not conserved among closely related chaperones, such as SrcA from Salmonella [15]. This C-terminal extension was shown to be important for effector secretion but was dispensable for effector binding [24]. We hypothesized that this C-terminal region of CesT was either disordered or heterogeneous from differential phosphorylation, possibly preventing favourable crystallization contacts. To address this, we truncated CesT at residue 138 (CesT138) and tested whether this variant could co-purify the same Tir peptides as full-length CesT. Tir peptides 23–80, 32–80, and 35–77 retained their ability to co-purify with CesT138 indicating that the C-terminus of CesT was not required for this interaction (Fig 1C and S1B Fig). However, the shorter Tir peptide 32–73 was now able to co-purify CesT138 (Fig 1C and S1B Fig), whereas the Tir peptides 37–80 and 37–73 were unable to co-purify CesT138. Gel filtration chromatography confirmed that both CesT and CesT138 existed in a dimeric configuration (S2 Fig), which is the functional unit of T3SS chaperones [30]. Taken together, these data suggest that the minimal CesT-binding region is located between residues 35–77 and that the C-terminus of CesT may interfere with binding of Tir residues 73–80.
The structural basis for how CesT binds and interacts with multiple T3SS effectors is not known. To determine the molecular determinants behind this interaction we conducted crystallization trials for all of the successful Tir-CesT co-purifications (S1 Fig). The Tir32-80-CesT138 complex produced crystals in the trigonal space-group P322 with one molecule of the complex in the asymmetric unit. Diffraction data were collected to 2.74 Å resolution and the structure was determined by molecular replacement (Table 1). Structural refinement produced a final model with good geometry and R factors (Rwork and Rfree of 20.8% and 25.5%, respectively) (Table 1). Residues 130–138 of CesT138, the N-terminal histidine tag, and residues 32–34, 54–64, and 76–80 of Tir were not included in the final model due to poor or absent electron density. Tir32-80 adopts minimal regular secondary structure that is limited to two small β-strands, β1’ and β2’ (Fig 2A). The Tir32-80 fragment binds CesT138 in two distinct locations and is separated by a break in the peptide chain likely due to residue mobility in the crystal. Tir residues 35–53 adopt a β-hairpin-like fold and extend the 5-stranded β-sheet core of CesT138, while also being pinched between α1 and orthogonally below by α3 of CesT138 (Fig 2A). The interaction between Tir32-80 β2’ and CesT138 occurs through a conserved 3-amino acid β-motif, adopting the consensus sequence of Φ-X4-Φ-x-Φ where x is any amino acid and Φ represents a hydrophobic residue. The β-motif was originally identified in the SipA-InvB complex [31, 32], but appears to be a conserved mode of binding present in all class I chaperone-effector complexes [33]. Slight differences have been observed in the β-motif, most notably that one to four residues can separate the first and second hydrophobic residues (ie. Φ-(X1-4)-Φ-x-Φ). Tir residues 65–75 are bound along the concave surface of the β-sheet core of CesT138 (Fig 2A). Despite CesT138 having a global acidic surface potential (Fig 2B), Tir32-80 binding is mediated through distinct hydrophobic-hydrophobic contacts (Fig 2C). Specifically, Tir residues I38 (purple), L44 (cyan), and L49 (cyan) anchor the β-hairpin-like peptide to CesT138 (Fig 2D); and L69 plus three additional proline residues make a second point of contact with the β-sheet core of CesT138 (Fig 2E).
CesT and CesT138 form a dimer in solution (S2 Fig) consistent with previous reports [30], and is a property conserved among T3SS class I chaperones. Although only one molecule of CesT138 was present in the crystallographic asymmetric unit, the dimer interface is clearly present along the principle 2-fold axis of symmetry (S3A Fig). Recently, the structure of CesT in complex with CsrA was reported [28]. CesT in the CsrA-CesT complex also adopts the same dimer orientation as observed in the Tir32-80-CesT138 complex, providing further evidence that the domain swapped dimer of the previous unladen EHEC CesT structure is likely a crystallographic artifact (S3 Fig). Furthermore, structural alignment of Tir32-80-CesT138 to the CsrA-CesT complex reveals significantly different binding modes for Tir and CsrA to CesT (Fig 3A; a monomer of CesT is shown for simplicity). CsrA binds CesT predominantly through residue contacts along CesT α3 and α4 (red), in which the latter comes from the second molecule of the CesT dimer (Fig 3A). In contrast, Tir32-80 binds the cleft formed between CesT α1 and β1 (Fig 3A). CsrA doesn’t directly occlude binding of Tir residues 35–53 to CesT, but residues K26 and R31 come within very close proximity, 2.9 Å and 2.2 Å, from Tir residues G45 and S46, respectively (Fig 3B). The C-terminus of CesT (cyan) from the CsrA-CesT complex, which is absent from CesT138, self-associates by binding along the concave surface of CesT (Fig 3A and 3C). Furthermore, the C-terminus of CesT also forms α4 (red) that interacts with CsrA, locking the C-terminal CesT peptide (residues I32 to Y153) in place (Fig 3A and 3C). Interestingly, the CesT C-terminus occupies the same binding surface as Tir residues 65–75 (pink, Fig 3C), despite significantly different sequences. These findings likely explain why none of the Tir peptide-CesT complexes crystallized, as CesT residues I32 to Y153 would compete for the same binding grove as Tir residues 65–75, and thus required the truncation of the CesT C-terminus (CesT138). Furthermore, this explains why only CesT138 could co-purify with Tir32-73, as the self-associated C-terminus of CesT likely out-competes the smaller Tir peptide for binding along the same concave surface. Taken together, the structural data from the Tir32-80-CesT138 and CsrA-CesT complexes suggest that (i) CesT exists as an unswapped dimer, (ii) the Tir binding region of CesT exhibits significant plasticity that could accommodate the binding of multiple effectors with varying sequences, and (iii) the C-terminal extension of CesT is required for CsrA interaction that in turn could also prevent CesT from binding Tir.
Tir32-80 binds the same hydrophobic surface in each monomer of the CesT138 dimer, producing a Tir32-80:CesT138 stoichiometry of 2:2 in solution, that was validated by gel filtration chromatography (S2 Fig). This crystal packing orientation was observed for the chaperone-effector fragment complexes of SycH-YscM2 [34], SycH-YopH [35], and ShcA-HopA1 [33]. However, gel filtration chromatography of the Tir23-550-CesT complex suggests that only a monomer of Tir23-550 binds a dimer of CesT (S2 Fig), consistent with a 1:2 Tir:CesT stoichiometry reported recently [36]. Since one molecule of Tir binds a dimer of CesT, but our crystallographic data suggest that two Tir32-80 fragments can bind a dimer of CesT, these data could be reconciled if full-length Tir contained a second uncharacterized CesT-binding region. Co-expression pull-down assays support this hypothesis as Tir81-550, which lacks the N-terminal CesT-binding region, retained the ability to pull-down CesT (Fig 4A). This was consistent with previous data showing that CesT can interact with N-terminal truncations of Tir in EHEC [37]. Furthermore, bacterial adenylate cyclase two hybrid (BACTH) assays showed that fusion of T18 to Tir23-80, Tir23-550, and Tir81-550 all had a positive interaction with CesT and CesT138 fused to T25 (blue colonies), but not to T25 alone (white colonies) (Fig 4B). We also observed CesT-CesT and Tir-Tir interactions in these assays consistent with previous reports [38, 39]. To identify the second CesT-binding region of Tir we used Tir residues 29–80 to conduct sequence alignments on a sliding window of ~50–80 amino acids. The carboxy-terminus of Tir (residues 490–550) had 20% sequence identity to Tir 29–80 and contained a TGRLIGT sequence similar to the sequence that forms β1’ in the N-terminal region of Tir (Fig 4C). T18 fused to Tir490-550 showed a strong interaction with CesT fused to T25 by BACTH assays (Fig 4B), and was able to pull-down CesT in co-expression pull-down experiments (Fig 4D), confirming this site as a second CesT-binding region. Taken together these data suggest that Tir is unique among E. coli effectors in that it contains a second carboxy-terminal CesT-binding region that is sufficient for interaction with CesT. Furthermore, both of the Tir CesT-binding regions have a conserved sequence motif distinct from the known chaperone binding β-motif.
Previous studies on the Salmonella effectors SipA and SptP showed that disruption or deletion of the β-motif prevented chaperone binding and subsequent effector secretion through the T3SS [31, 40]. To probe the function of the N- and C-terminal CesT-binding regions in Tir, we constructed leucine to glutamate mutants, L49E and L514E, within the N- and C-terminal β-motifs (cyan, Fig 4C). Since we showed that Tir lacking the N-terminal CesT-binding region (Tir81-550) could still interact with CesT, we first tested if β-motif variants of the individual N- and C-terminal Tir-peptide fragments retained CesT binding. Despite CesT being abundant in the soluble lysate, the Tir23-80 L49E and Tir490-550 L514E peptide variants were unable to pull-down CesT indicating that disruption of either β-motif prevented capture by CesT (Fig 5A). Consistent with this, T18 fusions of the Tir23-80 L49E and Tir490-550 L514E β-motif variants also showed little to no interaction in the presence of CesT and CesT138 fused with T25 in BACTH assays (Fig 5B, white colonies). However, the wild type T18 fusions of Tir23-80 and Tir490-550 constructs interacted with CesT and CesT138 fused to T25 (Fig 5B, blue colonies). As a control, we tested if the individual Tir23-550 L49E and Tir23-550 L514E variants were stable and could pull-down CesT, as both individual variants still have an intact CesT-binding region. Tir23-550 L49E and Tir23-550 L514E were both able to pull-down CesT suggesting that the mutations did not alter the global stability and structure of Tir, and that the L49E and L514E variants only locally disrupt CesT binding to Tir at the N- and C-terminal CesT-binding regions, respectively (Fig 5C). However, the Tir23-550 L514E variant showed lower levels of CesT pull-down compared to wild-type and L49E Tir23-550. We also tested the Tir23-550 L49E L514E double variant, as we predicted that disrupting both the N- and C-terminal CesT-binding regions would disrupt interaction with CesT. Indeed, the Tir23-550 L49E L514E double variant was significantly impaired for the ability to pull-down CesT (Fig 5C). As a functional readout, we tested whether the individual CesT-binding regions were required for Tir secretion by complementing a Δtir mutant with tir L49E, tir L514E, or the tir L49E L514E double mutant. As a positive control, Tir secretion was restored to wild type levels in the Δtir strain complemented with tir under its native LEE5 promoter (Fig 5D and S4A Fig). The Tir-L49E variant, which retained the functional C-terminal CesT-binding region, also restored Tir secretion to wild-type levels (Fig 5D and S4 Fig). However, the Tir-L514E variant that retains the N-terminal CesT-binding region, but disrupts the C-terminal CesT-binding region, drastically reduced Tir secretion to levels similar to the Tir-L49E L514E double variant (Fig 5D and S4 Fig). To test whether these same phenotypes were observed for the biologically relevant process of effector translocation, we analyzed infected HeLa cells for effector translocation using these same strains. Similar to the secretion assays, complementation of the Δtir strain with tir or tir L49E restored Tir secretion and resulted in the host-cell modification of tyrosine phosphorylation as observed by the upper band in Tir immunoblots (Fig 5E). The Δtir strain complemented with tir L514E or tir L49E L514E significantly reduced host-cell translocation (Fig 5E). Considering that host tyrosine phosphorylation of Tir is important for actin polymerization within the infected cell, we analyzed infected HeLa cells for pedestal formation. Consistent with our phenotypic data, the Δtir strain complemented with empty plasmid was impaired for pedestal formation, whereas the Tir and Tir L49E-complemented strains formed thick actin pedestals underneath microcolonies of bacteria in >90% of all analyzed cells (Fig 5F and 5G). Bacterial cells expressing Tir L514E or Tir L49E L514E were significantly impaired for the formation of actin pedestals (Fig 5F and 5G), which is in agreement with the host translocation results. As a secondary proxy for effector secretion and translocation, we also probed the levels of NleA, as it is highly dependent on free CesT available in the cell. We observed reduced levels of endogenous NleA secretion and host translocation in the Δtir strain complemented with plasmids carrying the tir wt, L49E, and L514E variants (Fig 5D and 5E), likely due to the higher cellular levels of Tir from trans-complementation that would reduce the levels of free CesT. However, the tir L49E L514E double mutant displayed higher levels of NleA secretion and translocation (Fig 5D and 5E), correlating with reduced binding of the Tir L49E L514E mutant to CesT (Fig 5C). Together these data suggest that the C-terminal CesT-binding region of Tir is required for efficient Tir secretion, host translocation, and the formation of actin pedestals.
Complementation studies showed that perturbation of CesT binding to the C-terminal region of Tir impaired Tir secretion and host translocation. Since the second CesT-binding region appears to be unique to Tir among all other CesT cargo, we tested the effect of N- and C-terminal domain truncations of Tir for secretion and translocation efficiency. Truncations of tir were constructed on the chromosome to produce Tir fragments encompassing residues 1–391 (tir NT) and 320–550 (tir CT) (Fig 6A). Tir NT contains the two elements predicted to be required for secretion, a T3SS signal sequence and a CesT-binding region. On the other hand Tir CT lacks a type III secretion signal sequence but contains the novel CesT-binding region in the C-terminal domain. Secretion assays conducted with the tir NT and tir CT-expressing strains revealed that, although Tir NT was present at similar levels as full-length Tir in the bacterial cytosol, Tir NT was secreted at a very low level compared to full-length Tir (Fig 6B). This strain also displayed similar levels of NleA secretion but drastically reduced cytosolic levels of NleA compared to wild type (Fig 6B). These results are consistent with the Tir L514E variant that displayed impaired in vitro secretion (Fig 5D). The tir CT strain did not secrete Tir CT as expected, since it lacks a type 3 secretion signal, but it was present in the whole cell lysate albeit at a very low level compared to wild type Tir. Interestingly the tir CT strain displayed a marked increase in cytosolic and secreted NleA compared to wild type or tir NT strains (Fig 6B and S4B Fig). Next, we tested whether the tir mutant strains were functional for effector translocation into HeLa cells, and the formation of actin pedestals. Similar to the in vitro assays, the tir NT strain translocated lower levels of Tir NT compared to wild type EPEC, but had similar levels of NleA (Fig 6C). In contrast, the tir CT strain was unable to translocate Tir CT as expected, but now had low levels of NleA secretion similar to the Δtir strain (Fig 6C). Other than wild type, none of the strains tested were able to form highly polymerized actin pedestals (Fig 6D and 6E). Taken together, these data indicate that the Tir C-terminal domain is important for Tir secretion and translocation into host cells, and that cellular levels of Tir in the bacterial cytosol affect the production and secretion of NleA.
Since the TGRLISS sequence in Tir was highly conserved between the N- and C-terminal CesT-binding regions (Fig 4C), we conducted a sequence search of this motif in other CesT cargo. We found a similar sequence, which we termed the CesT-extension motif, at the N-terminus of NleH1, NleH2, and EspZ (Fig 7A). The CesT-extension motif was much less conserved or not apparent in other CesT-dependent effectors (Fig 7A), suggesting that only a subset of CesT cargo contain this motif. We constructed disruptive glutamate variants in either the isoleucine or leucine residue present in the CesT-extension motif (residues starred purple in Fig 7A). This conserved isoleucine or leucine residue was chosen because, based on our structure, I38 in Tir32-80 extends into the hydrophobic CesT-binding pocket (purple, Fig 2D) and mutation to glutamate would disrupt the formation of β1’. Tir was tested first, showing that the individual Tir23-550 I38E and Tir23-550 I500E variants were stable and could pull-down CesT similarly to wild type Tir (Fig 7B). The Tir23-550 I38E I500E double variant was also able to pull-down CesT, albeit at slightly lower levels, suggesting that the CesT-extension motif is not obligatory for CesT binding (Fig 7B). To determine if the individual CesT-extension motifs were required for Tir secretion, the Δtir strain was complemented with either tir I38E, tir I500E, or tir I38E I500E and Tir secretion was tested in secretion assays. Tir secretion was restored to wild type levels in the Δtir strain complemented with tir under its native promoter and with the Tir-I38E and Tir-I500E single substitution variants (Fig 7C and S4C Fig). However, the Tir I38E I500E double variant had a slight reduction in Tir secretion compared to the wild type-complemented strain in these secretion assays (Fig 7C and S4C Fig). This is consistent with the reduced levels of CesT observed in the pull-down assays using the Tir I38E I500E double variant, but does not excluded the possibility that other factors contribute to the reduce secretion of the Tir I38E I500E variant. To further our in vitro secretion studies, we also analyzed infected HeLa cells for effector translocation and actin polymerization with the same strains. Complementation of the Δtir strain with tir, tir I38E, and tir I500E, restored Tir translocation and tyrosine phosphorylation (Fig 7D), albeit Tir-I38E had slightly lower Tir levels. The Δtir strain complemented with tir I38E I500E had the lowest levels of translocation and host modification (Fig 7D). Consistent with these results, we observed a significant reduction in actin pedestal formation in the strain expressing the Tir I38E I500E double variant (Fig 7E and 7F). We also probed the levels of NleA and similar to the Tir β-motif variants, we observed reduced levels of NleA secretion and translocation in the tir complemented strains (Fig 7C and 7D). However, the tir I38E I500E double mutant did not show higher levels of NleA secretion and translocation like the tir L49E L514E strain (Fig 7C and 7D), correlating with the pull-down data showing the Tir I38E I500E mutant retains CesT interaction (Fig 7B).
To expand these observations with other effectors, we made the equivalent glutamate substitutions in NleH1 (L28E), NleH2 (L28E), and EspZ (L45E). We also included NleA (V44E) in this analysis because it is a highly translocated effector but contains a very divergent sequence to the CesT-extension motif (Fig 7A). We tested if His6-tagged effectors and their putative CesT-extension motif variants were stable and could pull-down CesT. NleH1, NleH2, EspZ, and their respective glutamate variants were all able to pull-down CesT, however reduced CesT pull-down was observed for NleH2 L28E and EspZ L45E (Fig 8A). Interestingly, we observed little to no CesT in the NleA pull-downs, suggesting that CesT may not act as a chaperone for NleA but is only required to antagonize CsrA repression of the nleA 5’-UTR [27]. To determine if the CesT-extension motif affects secretion of these effectors (excluding NleA), secretion assays were conducted using EPEC carrying a plasmid expressing FLAG-tagged versions of NleH1, NleH2, EspZ, and their glutamate variants (Fig 8B). NleH1 L28E had a slight reduction in secretion and NleH2 L28E had little to no reduction in secretion compared to wild type. EspZ L45E was not detected in either the supernatant or whole cell lysate suggesting the mutation affected overall effector stability, possibly due to reduced CesT binding in the cytosol, which would be consistent with the pull-down data. Finally, we tested if the glutamate mutations in the CesT-extension motif of each effector affected translocation into infected HeLa cells. Under infection conditions, there was reduced translocation of NleH1 L28E, NleH2 L28E, and EspZ L45E compared to the wild type effectors (Fig 8C). Together these data suggest that the presence of a CesT-extension motif, in addition to the canonical β-motif, contribute to cargo recognition by CesT in a subset of effectors.
Tir drives the committal step of intimate attachment between EPEC and the host cell through an extracellular interaction with intimin on the bacterial cell surface [41, 42]. Thus, despite at least 12 effectors having full or partial dependence on CesT for translocation into host cells, Tir is the first effector to be released [22, 23]. This ultimately leads to attaching and effacing lesions from Tir-dependent signaling cascades that cause actin polymerization at the site of attachment, followed by Tir-independent effects on the host cell resulting from the release of secondary effectors [43]. Despite almost two decades of work on Tir and CesT, the mechanism that discriminates Tir secretion over that of other effectors remains unclear. Early transcriptional activation of tir is one possible mechanism to ensure Tir is available first for secretion. However, this does not seem to be the driving mechanism because LEE5 (which contains tir, cesT, and eae) is activated concurrently with LEE2, LEE3, LEE4, and LEE7 approximately 70 minutes after exposure to T3SS inducing conditions [44]. Furthermore, Tir secretion occurs approximately 30 minutes after transcriptional activation [44], suggesting a post-translational mechanism might drive preferential Tir secretion, as a number of other LEE-encoded effectors would also be present in the cytosol that would require discrimination within the cell. The nature in which Tir interacts with CesT is a possible mechanism by which this discrimination occurs. In this work we identified a second CesT-binding region in the C-terminal domain of Tir, and identify a CesT-extension motif, distinct from the known chaperone-binding β-motif, that is present in the CesT-binding regions of Tir and other highly translocated effectors. Our data raise the possibility that the presence of these features contribute to cytosolic discrimination by CesT, however formal assessment of this hypothesis in the context of the bacterial cell remains to be tested.
CesT has been the focus of structural and functional studies since its initial discovery as a Tir-specific chaperone [18]. Following structural determination of EHEC CesT [30], the domain swapped dimer has been a topic of debate as to whether it represents a biologically relevant conformation [45]. Our structural data on Tir32-80-CesT138, and work from others on the CsrA-CesT complex [28], indicates that CesT adopts the same dimer conformation even though it binds different substrates in separate locations. These data, along with solution state structural data from NMR [45], provide evidence that the domain swapped EHEC CesT dimer is most likely an artifact of crystallization, possibly arising from plasticity in the effector-binding region. Interestingly, the structure of the SrcA chaperone from Salmonella also exhibited plasticity in the effector-binding region [15], which suggests this could be a conserved property among multi-cargo chaperones to accommodate binding of multiple effectors.
Previous studies predicted that Tir residues 39–83 contained a degenerate CesT-binding domain [21]. Our Tir-peptide pull-down and structural data show that the minimal CesT-binding region of Tir is residues 35–75. In addition to providing the first structural view of an effector bound to CesT, the structure of the Tir32-80-CesT138 complex was critical in identifying the second CesT-binding region within the carboxy-terminus of Tir (residues 490–550). This chaperone binding arrangement appears unique to Tir, as there is no evidence for a secondary CesT-binding region in other CesT effector cargo. For example, the first 101 residues of Map were sufficient to interact with CesT, whereas the C-terminal 103 residues showed no interaction [20]. Furthermore, the structural data lead to the identification of a putative CesT-extension motif present in the N-terminal region of NleH1, NleH2, EspZ, and in both the amino- and carboxy-terminal regions of Tir. This region was so named because, according to our structure, it appears to extend the β-sheet core of CesT in its cargo-laden state. Mutation of the CesT-extension motif showed that it was most important for efficient host translocation. A commonality among effectors containing the CesT-extension motif (Tir, NleH1, NleH2, and EspZ) is that they are among the most highly translocated effectors among CesT cargo [23]. The presence of this motif in a subset of effectors (and its duplication in the case of Tir) may help to understand how multi-cargo chaperones recognize and possibly discriminate between effectors in the cell.
Efficient Tir-CesT complex formation is likely driven by the presence of two CesT-binding regions in Tir, however additional factors likely contribute to the observed preference for initial Tir secretion [21–23]. One possible contributing factor could be posttranslational modification of the C-terminus of CesT, which contains a site for tyrosine phosphorylation (Y152/153) [25]. A recent study showed that EPEC expressing a CesT Y153F variant exhibited a global increase in effector secretion, whereas EPEC expressing CesT Y152F was attenuated for NleA translocation into HeLa cells [26]. The latter result could be explained by the recent structure of CsrA in complex with CesT, which shows that CesT Y152 forms critical hydrogen bonds along the CsrA binding interface [28]. Therefore, CesT Y152F likely has reduced binding for CsrA and in turn is unable to depress the nleA 5’-UTR. We also observed that NleA had little to no binding of CesT in our pull-down assays. This suggests that the requirement of CesT for NleA secretion dynamics may be indirect and relate more to CsrA antagonism of the nleA 5’-UTR. This would be consistent with other work showing that NleA is only partially dependent on CesT for translocation into host cells [35]. Alternatively, it is possible that NleA or CesT may need post-translational modifications to facilitate interaction, require a third unknown co-chaperone, or NleA follows a chaperone-independent secretion pathway recently reported for a subset of Shigella T3SS effectors [46].
In addition to driving the formation of the Tir-CesT complex, the second CesT-binding region in Tir might stabilize a distinct conformation of the C-terminal domain of Tir that may be required for efficient targeting of the Tir-CesT complex to the T3SS sorting platform. Our data with the CesT-binding region mutants within the C-terminal domain of Tir (L514E variant), along with the chromosomal tir NT mutant (C-terminal domain truncation) support this possibility. For example, the Tir L514E and Tir NT constructs both contain a functional CesT-binding region and a type III secretion signal, but show significantly reduced secretion and translocation efficiency into HeLa cells. Furthermore, deletion of residues 519–524 in the C-terminal domain of EHEC Tir also showed significantly reduced Tir secretion [47]. Interestingly, EHEC Tir residues 519–524 align with EPEC Tir residues 511–516, which overlap with the predicted β-motif in the second CesT-binding region (ie. the Tir L514E mutant we tested). Recently, it was shown that an affinity switch controls substrate secretion hierarchy in the T3SS of EPEC. The SepL-SepD complex engages EscV (translocase) to ensure efficient targeting and secretion of the translocators, while simultaneously inhibiting effector targeting [36]. SepD release from the complex disrupts SepL-EscV crosstalk, leading to equivalent targeting of translocators and effectors for secretion. This is followed by the eventual release of SepL that results in inhibition of translocators and exclusive targeting of late effectors. This study also showed that the Tir-CesT complex had a two-fold increase in affinity for wild-type inner membrane vesicles that contain SepD, SepL, and EscV, over CesT alone [36]. This increased affinity for the Tir-CesT complex is probably due to a SepL-Tir interaction, which is supported by previous pull-down data in EHEC where the C-terminal 48 residues of SepL interact with Tir [37]. This particular Tir interaction site on SepL in EHEC also overlaps with one of the two EscV binding patches observed on EPEC SepL from peptide-binding array data [36]. Considering these studies with our findings, we provide an extended version of the affinity-switch model proposed by Portaliou et al. that includes differential secretion of late effectors (Fig 9). In this model, the SepD-SepL (and likely CesL) complex interacts with EscV and allows for strict translocator secretion (EspA, EspB, EspD) while simultaneously preventing effector secretion. It is also noteworthy that EscP binds SepL in a calcium-dependent manner and also contributes to the blocking of late effectors for secretion [48]. At this point it is plausible that CesT dimers are predominantly loaded with Tir in the cytosol since Tir contains two CesT-binding regions. After SepD dissociation from the EscV membrane complex, the Tir-CesT complex might compete with SepL for EscV binding. This competition could be mediated by the C-terminal domain of Tir leading to strict docking of the Tir-CesT complex to EscV over other effectors. Alternatively, the Tir-CesT complex may compete for EscP binding to SepL, leading to early docking of the complex within the sorting platform. Either possibility may explain preferential Tir release over other effectors, but it remains to be shown experimentally. SepL eventually dissociates from EscV, a process that may be directly influenced by Tir-SepL interactions, leading to the inhibition of translocator secretion and strict targeting of late effectors. Rapid release of Tir would result in the accumulation of free CesT in the cell, which in turn can antagonize CsrA repression of the nleA transcript through CesT-CsrA interactions, and increase binding of other highly translocated effectors such as EspZ, NleH1, and NleH2. Further depletion of the effector pool liberates more CesT, allowing for binding and secretion of lower translocated effectors.
Bacterial strains, plasmids, and primers used in this study are described in Tables A and B in SI Text. Phusion or Phire II polymerase (Thermo Fisher Scientific) were used for all PCR reactions, oligonucleotide primers were synthesized by Sigma, and site-directed mutants were constructed by using the Q5 site-directed mutagenesis kit (NEB) with the mutation encoded in the amplification primer. For protein expression, CesT and the C-terminal truncation construct encoding residues 2–138 (CesT138) were cloned into pET28a using the NheI/XhoI restriction sites. To allow for subsequent sub-cloning of CesT the internal NdeI site was removed by introducing a silent mutation in the coding region of His138 (CAT::CAC). CesT and CesT138 were then sub-cloned into the second multiple cloning site of pCOLADuet-1 using the NdeI/XhoI restriction sites. CesT-FLAG and CesT138-FLAG were also cloned into pCOLADuet-1 using the NdeI/XhoI site with the FLAG-tag encoded within the PCR amplification primer. Tir and EspZ; NleH1 and NleH2; and NleA constructs were cloned using the BamHI/HindIII, BamHI/SalI, and BamHI/NotI sites, respectively, into the first multiple cloning site of pCOLADuet-1 with CesT or CesT138 (and/or their FLAG-tag versions) in the second multiple cloning site for co-expression and pull-down experiments. For T3SS complementation assays, the LEE5 promoter encompassing nucleotides -7 to -323 from the translational start site of tir was cloned into pWSK29 using the XhoI/HindIII restriction sites. Subsequently, tir was cloned into the pWSK29-Ptir plasmid using the HindIII/NotI restriction sites. For effector secretion studies, effectors carrying a C-terminal FLAG-tag encoded in the primer were cloned into pGEN-luxCDABE using the SnaBI/SacI sites. For BACTH experiments, CesT and Tir constructs were cloned into the pKNT25 and/or pUT18C plasmids using the XbaI/SacI sites. All plasmids were verified by sequencing.
Protein fractions were separated by SDS-PAGE and transferred to a polyvinylidene fluoride membrane, and blocked in Tris-buffered saline with 0.1% (w/v) Tween (TBST) containing 5% skim milk. Membranes were then probed using the following primary antibodies: mouse monoclonal anti-Tir (1:2000) for full-length and C-terminal fragments, rat polyclonal anti-Tir from C. rodentium (1:2000) [49] for N-terminal fragments, rat polyclonal anti-NleA from EHEC (1:2000) [50], mouse α-DnaK (Stressgen, 1:5000), mouse α-FLAG M2 (Sigma, 1:5000), mouse α-His6 (GE Healthcare, 1:3000), or goat α-GAPDH (R&D Systems Inc., 1:5000). The blots were then developed using the following secondary antibodies: goat anti-mouse (1:5000, Jackson), goat anti-rat (1:2000, EMD Millipore), or donkey α-goat (Santa Cruz Biotechnology, 1:5000) conjugated to horseradish peroxidase, and imaged using the Clarity Western ECL (BioRad) or SuperSignal West Femto Maximum Sensitivity (ThermoFisher) substrates and a ChemiDoc XRS+ (BioRad).
E. coli BL21-CodonPlus (DE3) cells were transformed with the appropriate co-expression plasmid (pCOLADuet-1 containing N-terminal His6-tagged effector and C-terminal FLAG-tagged CesT), grown overnight in LB media with 50 μg/mL kanamycin at 37°C with shaking, sub-cultured 1:50 in 50 mL LB media with 50 μg/mL kanamycin to an OD600 of ~0.4–0.5, and moved to 30°C. When the cultures reached OD600 of ~0.6–0.7 protein expression was induced by the addition of isopropyl-D-1-thiogalactopyranoside (IPTG) to a final concentration of 0.5 mM. The cells were incubated for 3 h at 30°C, harvested by centrifugation at 5000 g for 10 min, and frozen on dry ice. Cell pellets were thawed and re-suspended in 2 mL of lysis buffer (50 mM Tris-HCl pH 7.5, 300 mM NaCl, 20 mM imidazole, 5% (v/v) glycerol, 2 mM 2-mercaptoethanol). Re-suspended cells were lysed by sonication and cell debris was removed by centrifugation at 16000 g for 30 min. The resulting supernatant was passed over a gravity column containing 0.2 mL Ni-nitrilotriacetic acid (NTA) agarose resin (Qiagen) that was pre-equilibrated with lysis buffer. Bound protein was washed with 100 column volumes of lysis buffer, and eluted with 5 column volumes of lysis buffer with 250 mM imidazole. Soluble lysate and elution fraction samples were mixed with equal parts of 2X SDS-PAGE loading dye and analyzed by SDS-PAGE and western blotting.
The following protocol was used to express and purify CesT, CesT138, and all the Tir-CesT complexes for crystallization. E. coli BL21-CodonPlus (DE3) cells were transformed with the appropriate plasmid, grown overnight in LB media with 50 μg/mL kanamycin at 37°C with shaking, sub-cultured 1:50 into 1 L LB media with 50 μg/mL kanamycin to an OD600 of ~0.4–0.5, and moved to 18°C. When the cultures reached OD600 of ~0.6–0.7 protein expression was induced by the addition of IPTG to a final concentration of 0.5 mM. The cells were incubated overnight at 18°C, harvested by centrifugation at 5000 g for 10 min, and frozen on dry ice. Cell pellets were thawed and re-suspended in 25 mL of lysis buffer (50 mM Tris-HCl pH 7.5, 300 mM NaCl, 10 mM imidazole, 5% (v/v) glycerol, 2 mM 2-mercaptoethanol, and one complete mini protease inhibitor cocktail tablet (Roche)). Re-suspended cells were lysed by sonication and cell debris was removed by centrifugation at 31000 g for 30 min. The resulting supernatant was passed over a gravity column containing 3 mL Ni-NTA agarose resin (Qiagen) that was pre-equilibrated with lysis buffer. Bound protein was washed with 10 column volumes of lysis buffer, 3 column volumes lysis buffer with 20 mM imidazole, and eluted with 5 column volumes of lysis buffer with 250 mM imidazole. The eluted fraction was concentrated using a 10 or 30 kDa cut-off Amicon ultrafiltration device (EMD Millipore) and further purified and buffer exchanged into 20 mM Tris-HCl pH 7.5 and 150 mM NaCl by size exclusion chromatography using a HiLoad 16/60 Superdex 200 prep-grade gel-filtration column (GE Healthcare). The purified constructs were >95% pure as judged by SDS-PAGE and stable for at least 1 week at 4°C.
Purified His6-Tir32-80-CesT138 was concentrated to ~14 mg/mL and screened for crystallization conditions at 22°C using hanging-drop vapour diffusion in 24-well VDXm plates (Hampton Research) and the MCSG 1–4 sparse matrix suites (Anatrace). The best initial crystallization hits were obtained from MCSG-1 condition #17 and MCSG-3 condition #44. Optimized crystals were grown by mixing 2 μL of 14 mg/mL His6-Tir32-80-CesT2-138 with 1.5 μL of precipitant solution (0.1 M Tris-HCl pH 7.5, 0.2 M MgCl2, and 17% (w/v) PEG3350) equilibrated against 500 μL of 1.7 M MgSO4. The crystals took 1–3 weeks to reach maximum size and were frozen without cryoprotection in liquid nitrogen. Diffraction data were collected at a wavelength of 0.98 Å on beam line 08B1-1 at the Canadian Light Source (CLS) (Table 1). The data were indexed and integrated with iMosflm [51] and scaled using SCALA in the CCP4i suite [52]. The structure was determined by molecular replacement with Phenix Phaser [53] using EHEC CesT residues 38–131 (PDB ID: 1K3E) as a search model. The resulting electron density map enabled Phenix AutoBuild [54] to build ~70% of CesT138. The remaining CesT138 residues and the Tir32-80 fragment were built manually in Coot [55] and alternated with refinement using phenix.refine [56]. Translation/Libration/Screw (TLS) groups were used during refinement and determined automatically using the TLSMD web server [57, 58]. Structure figures were generated using PYMOL Molecular Graphics System (DeLano Scientific), and quantitative electrostatics were calculated using PDB2PQR [59, 60] and APBS [61].
Elution fractions from various Tir-CesT co-expression pull-down experiments were concentrated to 100 μL, applied to a Superdex 200 10/300 GL column, and eluted using 20 mM Tris-HCl pH 7.5 and 150 mM NaCl. Protein standards used to calibrate the column were ferritin (440 kDa), conalbumin (75 kDa), ovalbumin (44 kDa), ribonuclease A (13.7 kDa), and aprotinin (6.5 kDa).
E. coli BTH101 cells were co-transformed with the various pKNT25 and pUT18C based plasmids and recovered on LB-Kan50-Amp100 agar plates at 37°C. Single colonies were grown overnight with shaking at 37°C in LB-Kan50-Amp100, and then 20 μL of each sample was spotted onto LB agar plates containing Kan50, Amp100, 0.5 mM IPTG, and 40 μg/mL 5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside (X-gal). Plates were incubated for 24–48 h at 30°C for the development of blue colonies.
Primer pairs with 48 nucleotide homologous tails to escN or tir were used to amplify linear PCR products with pKD3 for generation of the escN or various tir mutants. In-frame marked mutants of EPEC replacing escN residues 9–446 or tir residues 50–319, 392–535, and 17–535 with chloramphenicol acetyltransferase (cat) were constructed using one-step λ-red inactivation with pKD46 and the transformed linear PCR products [62]. The cat cassette was then removed using plasmid pFLP2 and sucrose selection. All tir and escN deletions were verified by sequencing.
Secretion assays were performed similar to those described previously [63]. Standing overnight EPEC cultures grown in LB media (plus 100 μg/mL Amp as needed) at 37°C were sub-cultured 1:40 into 4 mL of pre-warmed Dulbecco’s modified eagle medium (DMEM) plus 2 mM ethylene glycol-bis(β-aminoethylether)-N,N,N',N'-tetraacetic acid (EGTA) in glass tubes. The cultures were incubated standing for 6 h at 37°C in a 5% CO2 incubator (OD600 of 0.7–0.9). The cultures were then harvested by centrifugation at 10000 g for 5 min, and the bacterial pellets were washed once in phosphate-buffered saline (PBS) and re-suspended in 1X SDS-PAGE loading dye (normalized by OD600 as necessary). The culture supernatant was passed through a low-protein binding 0.2 μm filter (Pall), and 1.35 mL aliquots were mixed with 150 μL of ice-cold 100% (w/v) trichloroacetic acid and incubated overnight at 4 °C. The solutions were centrifuged at 16000 g for 30 min, the supernatant was discarded, and the pellet was washed with 1 mL of ice-cold acetone. The washed pellets were centrifuged at 16000 g for 30 min, the pellet was air dried, and then re-suspended in 10 μL 1X SDS-PAGE loading buffer (or normalized by OD600 as necessary). Samples were then analyzed by SDS-PAGE using coomassie blue G250 stain or by western blotting.
HeLa cells (Coombes lab collection) were grown in DMEM + 10% fetal bovine serum at 37°C in a 5% CO2 incubator. Cells were routinely grown in 75 mm2 dishes (VWR) until confluent, and were then seeded at 2.2×106 into 100 mm dishes (Corning) and incubated overnight. Prior to infection the HeLa cells in 100 mm dishes were washed with 10 ml of warm PBS. EPEC cultures were grown overnight standing at 37°C, harvested by centrifugation, and resuspended in DMEM. HeLa cells were then infected with EPEC at a multiplicity of infection of 50:1 for 3 h at 37°C in a 5% CO2 incubator. The cells were washed five times with cold PBS, harvested with a cell scraper, centrifuged at 1000 g for 5 min, and resuspended in 250 μL PBS + 0.5% (v/v) Triton X-100. Cells were then lysed for 30 min on ice with gentle rocking, centrifuged at 10000 g for 5 min, and the following supernatant was mixed with equal parts of 2X SDS-PAGE loading buffer for SDS-PAGE and western blot analysis.
HeLa cells maintained in DMEM + 10% fetal bovine serum were seeded at 1×105 into 24 well tissue culture plates (VWR) containing 12 mm circle micro coverglass slips (VWR) and incubated overnight at 37°C in a 5% CO2 incubator. Prior to EPEC infection the glass slips were washed with 1 ml of warm PBS. EPEC cultures carrying a GFP expression plasmid for visualization (pFPV25.1 for the tir chromosomal domain mutants and pACYC-GFP for the tir complementation strains with various Tir point mutants) were grown standing overnight in LB media at 37°C, harvested by centrifugation, and resuspended in DMEM. HeLa cells were then infected with EPEC at a multiplicity of infection of 50:1 for 3 h at 37°C in a 5% CO2 incubator. Infected cells were washed with PBS (and after each subsequent step), fixed with 4% paraformaldehyde in PBS for 15 min, permeabilized with 0.3% (v/v) Triton X-100 in PBS for 5 min, and blocked with 5% (w/v) bovine serum albumin (BSA) in PBS for 30 min. F-actin was then stained using Alexa Fluor 568 Phalloidin (1:500, ThermoFisher) in PBS containing 1% (w/v) BSA for 60 min. Stained coverslips were washed in PBS and mounted on glass slides using ProLong Gold Antifade Mountant with 4’,6-Diamidino-2-phenylindole dihydrochloride (DAPI) for nuclear staining (Life Technologies) and allowed to sit overnight before sealing with nail polish. Microscopy was performed using a ZEISS Axio Imager 2 with 40X and 100X oil-immersion lenses with laser excitation. Images were captured using a Hamamatsu ORCA-R2 digital CCD camera and exported TIFF files were processed into their individual and composite color channels using ImageJ2 [64]. Quantification of pedestal formation (binding index) was conducted as described previously [65], where the percentage of infected HeLa cells that contained a microcolony of at least five GFP-positive bacteria associated with F-actin condensation were enumerated (co-localization, yellow).
|
10.1371/journal.pntd.0003185 | RNAi Dynamics in Juvenile Fasciola spp. Liver Flukes Reveals the Persistence of Gene Silencing In Vitro | Fasciola spp. liver fluke cause pernicious disease in humans and animals. Whilst current control is unsustainable due to anthelmintic resistance, gene silencing (RNA interference, RNAi) has the potential to contribute to functional validation of new therapeutic targets. The susceptibility of juvenile Fasciola hepatica to double stranded (ds)RNA-induced RNAi has been reported. To exploit this we probe RNAi dynamics, penetrance and persistence with the aim of building a robust platform for reverse genetics in liver fluke. We describe development of standardised RNAi protocols for a commercially-available liver fluke strain (the US Pacific North West Wild Strain), validated via robust transcriptional silencing of seven virulence genes, with in-depth experimental optimisation of three: cathepsin L (FheCatL) and B (FheCatB) cysteine proteases, and a σ-class glutathione transferase (FheσGST).
Robust transcriptional silencing of targets in both F. hepatica and Fasciola gigantica juveniles is achievable following exposure to long (200–320 nt) dsRNAs or 27 nt short interfering (si)RNAs. Although juveniles are highly RNAi-susceptible, they display slower transcript and protein knockdown dynamics than those reported previously. Knockdown was detectable following as little as 4h exposure to trigger (target-dependent) and in all cases silencing persisted for ≥25 days following long dsRNA exposure. Combinatorial silencing of three targets by mixing multiple long dsRNAs was similarly efficient. Despite profound transcriptional suppression, we found a significant time-lag before the occurrence of protein suppression; FheσGST and FheCatL protein suppression were only detectable after 9 and 21 days, respectively.
In spite of marked variation in knockdown dynamics, we find that a transient exposure to long dsRNA or siRNA triggers robust RNAi penetrance and persistence in liver fluke NEJs supporting the development of multiple-throughput phenotypic screens for control target validation. RNAi persistence in fluke encourages in vivo studies on gene function using worms exposed to RNAi-triggers prior to infection.
| RNA interference (RNAi) is a method for selectively silencing (or reducing expression of) mRNA transcripts, an approach which can be used to interrogate the function of genes and proteins, and enables the validation of potential targets for anthelmintic drugs or vaccines, by investigating the impact of silencing a particular gene on parasite survival or behaviour. This study focuses on liver fluke parasites, which cause serious disease in both humans and animals. We have only a handful of drugs with which to treat these infections, to which flukes are developing resistance, and no anti-fluke vaccines have yet been developed. New options for treatment and control of liver fluke parasites are sorely needed, and RNAi is a powerful tool in the development of such treatments. This study developed a set of simple methods for triggering RNAi in juvenile liver fluke, which show that although robust transcriptional suppression can be readily achieved across all targets tested, protein suppression occurs only after a target-specific lag period (likely related to protein half-life), which may require >25 days under current in vitro maintenance conditions. These findings are important for researchers aiming to employ RNAi in investigations of liver fluke biology and target validation.
| Fasciola spp. liver flukes are the causative agents of fascioliasis, or liver fluke disease. Their growing importance as human neglected tropical disease pathogens, alongside their impact on animal health, welfare and global food security, is recognised [1]–[3]. This pathology is compounded by established incidences of field resistance to the handful of available flukicidal drugs in both veterinary and human infections [4], [5]. Although adult fluke are the reproductively active stage and cause physical-, nutritional- and immuno-pathology to the vertebrate host, it is the tissue-penetrating invasive newly-excysted juvenile (NEJ)/juvenile stage Fasciola, that cause significant damage to host tissues during their migration from the gut lumen, through the hepatic parenchyma to the bile ducts. These NEJs are not adequately targeted by existing chemotherapy; triclabendazole is the only flukicide with significant activity against juvenile liver fluke [6]. Novel flukicidal treatment options are urgently required, especially those with activity against juvenile fluke.
Beyond two initial reports of RNA interference (RNAi) [7], [8], there have been no documented accounts of efforts to develop functional genomics tools for the study of liver fluke biology and the validation of new drug and vaccine targets. Traditionally, such studies were hindered by a relative lack of liver fluke nucleotide sequence datasets compared to other parasitic helminths. However, ongoing efforts have seen the primarily adult-derived sequence datasets available through both GenBank and the Wellcome Trust Sanger Institute complemented by the public release of transcriptome datasets for adult specimens of both temperate and tropical liver fluke (F. hepatica and F. gigantica respectively), soon to be complemented by juvenile transcriptome datasets and draft F. hepatica genome sequences [9]–[12]. A suite of functional genomics tools and associated functional assays will be required in order to effectively exploit these sequence resources.
RNAi permits the destruction of target mRNA, and subsequent suppression of target protein, by the introduction of double stranded (ds)RNA trigger molecules of complementary sequence to an mRNA target [13]. In parasitic helminths and other species that are difficult or impossible to transform by available molecular genetic methods, RNAi provides a means of investigating gene function through the relatively simple generation of organisms in which the expression of a target gene has been reduced or ablated. This approach permits both the testing of basic biological hypotheses, and the identification of deleterious phenotypes that might inform drug/vaccine target validation efforts. While parasitic flatworms appear to be broadly amenable to RNAi [14]–[16], the initial development of RNAi methodology in any species requires optimisation of experimental variables, sometimes on a target-by-target basis, validated by measurement of the extent and specificity of transcript and protein knockdown [17]. Schistosoma mansoni has been well served by studies describing extensive experimental optimisations of RNAi methodology [18]–[22], enabling the effective exploitation of functional genomics tools by the associated research community. RNAi can be induced in F. hepatica by either soaking or electroporation [7], [8]. Considering the former method to be less onerous and less technically demanding, and therefore more likely to be employed by the research community in both small- and large-scale assay systems, here we set out to apply dsRNA exposure-induced RNAi to silence a range of virulence gene targets (and proposed vaccine antigen candidates), including cathepsin B and L cysteine proteases, fatty-acid binding protein, leucine aminopeptidase, and μ-, ω-, and σ-class glutathione transferases, in NEJs of a commonly-used, commercially-available F. hepatica isolate (US Pacific North West Wild Strain). Finding that RNAi transcript, protein and phenotype dynamics were quite distinct in this system compared to those reported previously in Oberon and Fairhurst isolates [8], we performed a set of in-depth experimental optimisations in order to characterise the RNAi mechanisms operating in this liver fluke isolate. Focusing on three examples of commonly cited virulence genes and vaccine targets – cathepsin B (FheCatB), cathepsin L (FheCatL) and a sigma-class glutathione transferase (FheσGST), we have systematically investigated the effects of variations in dsRNA trigger concentration, experimental timecourse, and dsRNA soak exposure protocol using both long (200–220 nt) dsRNA and 27 nt short interfering (si)RNA triggers, on knockdown of both target transcript and protein. We show that, although apparently operating in a manner distinct to that reported in existing literature [8], robust, persistent RNAi is eminently achievable in NEJs of the F. hepatica Pacific North West Wild strain, and in the tropical liver fluke F. gigantica, using a simple dsRNA-exposure protocol.
F. hepatica (US Pacific North West wild strain) metacercariae, with outer cyst wall removed, were obtained from Baldwin Aquatics Inc (Monmouth, Oregon, USA), and stored at 4°C until required. F. gigantica metacercariae were obtained from naturally infected wild snails collected in Aligarh, Uttar Pradesh, India, by researchers at Aligarh Muslim University. Outer cyst walls were removed by incubation in a solution of 1% pepsin, 4 mM HCl, for 90 min at 37°C, followed by several washes in distilled water. Metacercariae were then excysted by incubation in 0.6% sodium bicarbonate, 0.45% sodium chloride, 0.4% sodium tauroglycocholate, 0.025 M HCl, 0.4% L-Cysteine, for up to 3 h at 37°C. After approximately 75 min, at 10–15 min intervals, NEJs were transferred to RPMI 1640 without phenol red (Life Technologies), in which they were maintained at 37°C for a maximum of 3 h until transferred to soaking media. F. gigantica cysts were exposed to excystment media for 3 h, after which they were transferred to RPMI for incubation, in which their excystment completed within 18 h. Variations in maintenance/soaking conditions are described below.
Experimental and negative control (dsCTRL) RNAi triggers comprised “long” dsRNA molecules (sized between 200–320 nt) generated by T7 polymerase-driven transcription of single RNA strands (MEGAshortscript T7 Kit, Ambion) from target-specific PCR product templates labelled either sense or antisense with a T7 RNA polymerase promoter sequence (5′- TAATACGACTCACTATAGGGT -3′). Primers used to generate long dsRNA template PCR products were: FheCatB2 [U58000]: forward, 5′- GTTGTCAGCCCTGGATGTTT -3′, reverse, 5′ -GTCCTGGAATATGGCGAAAG- 3′; FheCatL [designed for maximum similarity against alignment of all FheCatL clades on GenBank, see Figure S1]: forward, 5′-TKRTTATGTGACGGAGGTGA-3′, reverse, 5′-GCCBKRTAHGGRTAAK-3′; dsCTRL [bacterial neomycin phosphotransferase, U55762]: forward, 5′- GGTGGAGAGGCTATTCGGCTA -3′, reverse, 5′- CCTTCCCGCTTCAGTGACAA -3′; FheFABP [AJ250098]: forward, 5′- TAAGATTACAACTTTCACATTTGGC -3′, reverse, 5′- GCGGGATGCTCAAAATCCGTC -3′; FheLAP [AY644459]: forward, 5′- GGTAGTGAACTGTCAATTGTTCCG -3′, reverse, 5′- GAGTAGCAGATGTTTGCGTTGC -3′; FheμGST [designed against alignment of M77682, M93434, M77680, M77681, M77679, see Figure S2]: forward, 5′- ACACCCGAGGAACGAGCTCG -3′, reverse, 5′- TCGTAAACCATAAAGTCCACATGG -3′; FheωGST [JX157880]: forward, 5′- GATTGGTATCTGGAGTTATATCCG -3′, reverse, 5′- AATAAAATACCATGCGATTGAGCC -3′; FheσGST [DQ974116]: forward, 5′- AATTCGCCTTCTGCTCACTTGC -3′, reverse, 5′- GTAATACTCTTCGTCTGTTTCACC -3′). Endogenous templates were amplified from F. hepatica (or F. gigantica) cDNA as appropriate; due to high identity between the relevant orthologues, we were able to use F. hepatica primers to amplify FgigσGST and FgigGAPDH from F. gigantica (see ‘quantification of transcript knockdown’ for GAPDH primer sequences). These amplicons were sequence confirmed. Exogenous dsCTRL templates were amplified from an eGFP plasmid (Promega). Complementary single-stranded RNAs were combined and annealed (75°C for 5 min, 19°C for 30 min) before treatment with DNase (Turbo DNase, Life Technologies) and RNase (RNase If, New England Biolabs) and subsequent phenol/chloroform extraction and ethanol precipitation. Pelleted dsRNA was briefly air-dried (∼10 min) at room temperature, resuspended in ≥100 µl DEPC-treated H2O, and stored at −20°C. All dsRNAs were quantified using a NanoDrop ND1000 Spectrophotometer and analysed for the presence of a discrete, correctly sized product on a non-denaturing 1% (w/v) agarose TAE gel. Only dsRNAs that gave a discrete correctly sized band, and 260/280 and 260/230 ratios >1.8 were used in RNAi experiments. Typical dsRNA concentrations following synthesis were in excess of 2000 ng/µl. Verified sizes of dsRNAs were: dsFheCatB: 248 nt; dsFheCatL 246 nt: nt; dsCTRL: 223 nt; dsFABP: 313 nt; dsFheLAP: 234 nt; dsFheμGST: 218 nt; dsFheωGST: 209 nt; dsFheσGST: 223 nt.
“Dicer substrate” 27mer siRNAs to FheσGST were designed and synthesised by Integrated DNA technologies (IDT; www.idtdna.com) (target sites: siGST1: 5′- TTCGAGGACTATCAATTCACAATGGAT -3′; siGST2: 5′- GGGAAACTTAGACGTTATCAAGAATCG -3′; siGST3: 5′- ATTGCTCGATTGCTTGCCAGACAATTC -3′; IDT's off the shelf DS Scrambled Neg siRNA (5′- CTTCCTCTCTTTCTCTCCCTTGTGA -3′) was used as the siCTRL). Upon receipt, siRNAs were resuspended in the supplied buffer to 100 µM concentration, and stored at −80°C in single-use aliquots until required.
NEJs were soaked in solutions of long dsRNA or siRNA dissolved to defined concentrations in RPMI 1640, in 2 ml round-bottomed microcentrifuge tubes. Soaks were handled at least in triplicate alongside triplicate untreated (no dsRNA) and size- and concentration-matched negative control long dsRNA (dsCTRL)- or negative control siRNA (siCTRL)-treated controls. For qPCR experiments 20 NEJs were used per soak, western blot experiments used 50 NEJs per soak. Soaks were performed under one of five variations on a dsRNA exposure protocol (described in Results – Burst exposure to trigger). Unless specified otherwise, NEJs were maintained aseptically in 1 ml RPMI 1640, at 37°C in a 5% CO2 atmosphere. RPMI was replaced at 2–3 day intervals for periods of up to 25 days. Worms were visually assessed for aberrant motility or morphology during media changes. At the end of each soak experiment, worms were either processed for RNA/protein extraction immediately, or snap frozen and stored at −80°C for processing at a later time.
mRNA was extracted from NEJs using the Dynabeads mRNA Direct kit (Life Technologies). We employed the mini extraction protocol as directed by the manufacturer (except that all washes were performed with only 300 µl buffer), and eluted in 12.5 µl Tris-HCl. Eluted mRNA was DNase digested using the Turbo DNA Free kit (Life Technologies) in a total volume of 15 µl, from which 8 µl was added directly to a 20 µl reverse transcription reaction using Applied Biosystems' High Capacity RNA to cDNA Kit, as described by the manufacturer. Following reverse transcription, cDNA was diluted with an equal volume of water before use in qPCR. Triplicate qPCR reactions were performed in 10 µl reaction volumes using the FastStart SYBR Green Master mix (Roche Applied Science), containing primers (FheCatB: forward, 5′-GCACGTACTGTGGTCAGGGGTG-3′, reverse, 5′-GTCCTGGAATATGGCGAAAG-3′; FheCatL: forward, 5′-TKRTTATGTGACGGAGGTGA-3′, reverse, 5′-GTATAGAAGCCAGTCACTTTGGC-3′; FheFABP: forward, 5′- AACAAAATGACTATTAAAATGG -3′, reverse, 5′- GCGGGATGCTCAAAATCCGTC -3′, GAPDH reference amplicon [AY005475]: forward, 5′- GGCTGTGGGCAAAGTCATTC -3′, reverse, 5′- AGATCCACGACGGAAACATCA -3′; FheLAP: forward, 5′- GGTAGTGAACTGTCAATTGTTCCG -3′, reverse, 5′- GGCAGATAGAGTGCATGGTACG -3′; FheμGST: forward, 5′- ACACCCGAGGAACGAGCTCG -3′, reverse, 5′- GGATAAGCATGGAATGCTTGGT -3′; FheωGST: forward, 5′- GATTGGTATCTGGAGTTATATCCG -3′, reverse, 5′- ATACGCAGACGCATTAGCTTCC -3′; FheσGST: forward, 5′- AATTCGCCTTCTGCTCACTTGC -3′, reverse, 5′- TCTTCACACTCACCAATGATACG -3′) at final concentrations of either 200 nM (FheCatB, FheCatL) or 1500 nM (FheσGST), as well as 2 µl cDNA template (or H2O in the case of negative PCR controls). As above, F. hepatica primers were used to amplify FgigσGST and FgigGAPDH from F. gigantica. Amplification was performed on a Qiagen RotorGene Q 5-plex HRM instrument (10 min 95°C; 40 cycles: 10 sec, 95°C, 15 sec, 60°C, 30 sec, 72°C, followed by melt curve analysis of amplicons). Relative expression analysis was performed manually using Pfaffl's Augmented ΔΔCt method [23] (which normalises expression in each sample relative to the untreated control, standardised to a GAPDH reference amplicon), with amplification efficiencies of individual reactions calculated using Real Time PCR Miner software (http://ewindup.info/miner/data_submit.htm; [24]). This analysis method produces a ratio of target transcript change relative to the untreated control (i.e. where a value of 1.0 represents no change) ± SEM. These ratios are plotted in bar graphs in the qPCR figures. Statistical analyses were performed on the ratios produced following relative expression analysis, using one-way ANOVA with Tukey's post test. Statistical significance was determined relative to the effects of negative control treatments (dsCTRL or siCTRL) on target gene expression.
Following soaking/maintenance using Protocol IV (as described in Results – Burst exposure to trigger), 50 NEJs were ground under liquid nitrogen in a round-bottom 2 ml microcentrifuge tube, after which 50 µl radioimmunoprecipitation assay (RIPA) buffer, containing protease inhibitors (Complete Mini, Roche Applied Science) and phosphatase inhibitors (Phosphatase Inhibitor Cocktail 2, Sigma Aldrich), was added. Tubes were then subjected to 3× 20 sec cycles of sonication, and left on ice for 60 min before quantification of protein content by BCA assay (BCA Protein Assay Kit, Pierce). Following addition of 50 µl 2× Novex Tricine SDS Sample Buffer (Life Technologies), and tris(2-carboxyethyl)phosphine (TCEP) to a final concentration of 10 mM, samples were denatured at 95°C for 5 min, and ≤25 µl of each sample (equalised for protein content) loaded into wells of a pre-cast 10–20%, 10 well, 1.0 mm, Tricine Gel (Life Technologies), alongside one lane containing 15 µl SeeBlue Plus2 Prestained Standard (Life Technologies). Gels were run in an Xcell Surelock Mini-Cell apparatus (Life Technologies) at 135 V for 45–60 min in tricine running buffer, before electroblotting onto nitrocellulose membrane in an Xcell Mini-Cell apparatus (Life Technologies), at 35 V, for 2 h. Membranes were washed briefly with TBST (Tris-Buffered Saline prepared from tablets [Sigma Aldrich], containing 0.05% Tween 20) before incubation in blocking buffer (TBST containing 5% Blotting Grade Non-Fat Dry Milk [Biorad]), for 30 min, at room temperature, with rotation. Membranes were then probed overnight at 4°C, with rotation, in blocking buffer containing two primary antisera (one target and one loading control normaliser) (rabbit-anti-σGST 1/15,000; rat-anti-FheCatB2 1/1000; sheep-anti-FheCatL1, 1/2000; rabbit-anti-actin [20]–[33, Sigma Aldrich] 1/400). Following 5×5 min washes in TBST, membranes were probed with appropriate secondary antisera (alkaline-phosphatase conjugated goat-anti-rabbit (Invitrogen), goat-anti-rat (Invitrogen), or donkey-anti-sheep [Sigma Aldrich], diluted 1/1000 in blocking buffer), for 2 h at room temperature. Following 5×5 min washes in TBST, membranes were exposed with NBT/BCIP solution (prepared from tablets, Roche Applied Science). Development, carefully monitored by eye, occurred in as little as 20 sec and was terminated by washing the membrane thoroughly in distilled H2O. Membranes were then dried, scanned, and band intensities quantified by densitometry using ImageJ software (http://rsbweb.nih.gov/ij/). In order to permit relative protein quantitation, band intensities for the proteins of interest were normalised relative to the intensity of the loading control band from the same sample, and then this figure was expressed as a percentage relative to the untreated control ( = 100%) sample. Statistical analyses were performed using one way ANOVA with Tukey's post test. All statistical significances presented in figures are expressed relative to the dsCTRL treatment.
The following accessions refer to GenBank (www.ncbi.nlm.nih.gov). Bacterial neomycin phosphotransferase: U55762; cathepsin B: U58000; cathepsin L: Z22767, EU287918, EU287915, AJ279091, EU191984, AJ279093, EU287914, EU195859, DQ534446, EU287917, EU287916, Z22763, DQ533985, Z22764, EF407948, Z22765, EF611824, U62289, Z22769, AJ279092, AY029229, AB009306, AY519972, AY573569, L33771, AF490984, AY277628, U62288, AY519971, DQ533986, Z22766, L33772, AF271385; fatty acid binding protein: AJ250098; μ-class glutathione transferase: M77682, M93434, M77680, M77681, M77679; ω-class glutathione transferase: JX157880; σ-class glutathione transferase: DQ974116; glyceraldehyde phosphate dehydrogenase: AY005475; leucine aminopeptidase: AY644459.
Although our early experiments were performed using Fasciola saline (FS) maintenance media as described previously [8], we found the viability (assessed as a visual measure of worm motility and morphology; non-motile worms with a visually disrupted tegument were considered dead) of NEJs to be poor over maintenance periods of more than a few days under these conditions (Figure 1). Initial soaks maintained 20 worms in 50 µl FS, in which NEJs survived for no longer than 8 days (50% survival = 5 days). With regular media changes (every 2 to 3 days), maintenance could be extended to a maximum of 11 days (50% survival = 9 days). Maintaining NEJs in a larger volume (1000 µl) improved survival further to a maximum of 13 days (50% survival = 12 days), although in this case NEJ survival was not improved by regular media changes. In order to track protein dynamics occurring on timescales beyond 13 days, we employed RPMI 1640 media, finding that this media supported NEJ survival for up to 27 days when maintained in aseptic conditions under a 5% CO2 atmosphere. All of the data presented below were derived from experiments employing RPMI 1640, although comparative experiments performed in both media across multiple targets over 72 h timescales illustrated no significant difference in RNAi efficacy between FS and RPMI 1640.
This study aimed primarily to identify a set of optimised RNAi experimental conditions that would permit simple and robust gene knockdown in a widely used and commercially available strain of F. hepatica. We describe a set of simple, soaking-based methods that permit rapid, robust and persistent knockdown of target transcripts, and address target-specific aspects of the timecourse of this RNAi response, which is particularly evident in the time lag between transcript and protein knockdown.
Initial experiments maintained NEJs under constant exposure to long dsRNA for periods up to 72 h, using the same dsRNA molecules/sequences, concentrations, and exposure conditions previously reported [8]. These conditions did not, in our hands, trigger aberrant motility phenotypes. Upon analysing transcriptional knockdown dynamics from these soaks, we established that there was no detectable difference in transcript knockdown level between 50 ng/µl and 100 ng/µl dsRNA exposures. All subsequent experiments employed 50 ng/µl dsRNA, unless stated otherwise. Although we were unable to detect the reduction in FheCatL transcript reported by McGonigle et al. after 4 h exposure, simply extending the length of dsRNA exposure did elicit time-dependent suppression of CatL, as well as our other target transcripts after 72 h (FheFABP: 0.03±0.01, P<0.001, n = 3; FheLAP: 0.16±0.10, P<0.01, n = 3; FheμGST: 0.09±0.02, P<0.01, n = 3; FheωGST: 0.38±0.06, P<0.05, n = 3; Figure 2A). Note that these four targets were assayed only by qPCR at the 72 h timepoint. Intermediate time-course exposures, performed to 4, 24 and 72 h (Figure 2B–D), showed that both FheCatL (24 h: 0.29±0.20, P<0.05; n = 4; 72 h: 0.16±0.06, P<0.01, n = 5), and FheσGST (24 h: 0.38±0.05, P<0.05 n = 5; 72 h: 0.02±0.004, P<0.01, n = 4) exhibited significant transcript knockdown from 24 h onwards, while FheCatB displayed a uniquely rapid response, with transcript knockdown detectable after only 4 h exposure to long dsRNA (4 h: 0.40±0.16, P<0.05, n = 6; 24 h: 0.13±0.06, P<0.05, n = 4; 72 h: 0.06±0.02, P<0.05, n = 4). None of these experiments resulted in detectable aberrant phenotypes. The major drawback of this approach was that NEJs maintained under constant exposure to long dsRNA in such small volumes exhibited rapidly declining viability beyond 2–3 days in vitro (Figure 1), which occluded our detection of potential impacts of RNAi on worm behaviour.
Poor survival of NEJs under the constant exposure conditions described above (here and in Figure 3 referred to as Protocol I) was mitigated by either regular changes (every 2–3 days), and/or maintenance in an increased volume (1 ml) of media. Since such approaches would require prohibitively large amounts of RNAi trigger during longer maintenance periods, we investigated the efficacy of short, low volume “burst” exposures to long dsRNA, followed by longer-term incubation in the absence of dsRNA under optimal maintenance conditions. These experiments were performed over a 72 h timecourse (Figure 3). Initial experiments (Protocol II) incorporated a 4 h dsRNA burst exposure, after which dsRNA was removed (or reduced to irrelevant levels) by a series of six washes in 1 ml RPMI 1640, before maintenance for 68 h in 1 ml RPMI 1640. In all cases, target transcript knockdowns had developed to highly significant levels by the end of this maintenance period (FheCatB: 0.09±0.02, P<0.001, n = 4; FheCatL: 0.18±0.08, P<0.001, n = 5; FheσGST: 0.18±0.06, P<0.001, n = 4; Figure 3B–D). Transcript knockdown was triggered even when the burst exposure period was reduced to as little as 30 min (Protocol III), although this knockdown was statistically significant only in the cases of FheCatB and FheσGST (FheCatB: 0.23±0.11, P<0.01, n = 3; FheCatL: 0.79±0.08, P>0.05, n = 3; FheσGST, 0.29±0.03, P<0.01, n = 5). A 15 min burst exposure was ineffective. Since NEJs were effectively removed from dsRNA following the period of burst exposure (and in the cases of FheCatL and FheσGST, knockdown had not been present after 4 h, Figure 2C,D), these findings indicate that dsRNA taken up by NEJs during their burst exposure period was sufficient to trigger the development of knockdown during subsequent incubation in the absence of exogenous dsRNA. In two out of three cases, the highest levels of knockdown were achieved over 72 h using the less onerous method (Protocol IV) of simply adding 1 ml RPMI 1640 for 68 h after the 4 h soak period (i.e. dilution of dsRNA to approximately 2.5 ng/µl) (FheCatB: 0.02±0.003, P<0.001, n = 3; FheCatL: 0.01±0.0003, P<0.001, n = 5; FheσGST: 0.08±0.01, P<0.001, n = 7). To ascertain the importance of the initial high-concentration burst exposure to this response, we also performed soaks in the absence of this step (Protocol V), simply soaking NEJs for 68 h in 2.5 ng/µl long dsRNA. This approach also delivered highly significant levels of transcript knockdown (FheCatB: 0.01±0.003, P<0.001, n = 6; FheCatL: 0.01±0.003, P<0.001, n = 7; FheσGST: 0.12±0.02, P<0.01, n = 3), indicating that the concentration threshold of the NEJ long dsRNA uptake mechanism is ≤2.5 ng/µl. For reasons of efficacy and convenience, all subsequent experiments in this study employed Protocol IV.
In order to establish the applicability of these RNAi methods to the tropical fluke, F. gigantica, we applied long dsRNA complementary to FgigσGST to F. gigantica NEJs using Protocol IV methods. This triggered effective knockdown of FgigσGST transcript (0.19±0.03, P<0.001, n = 5; Figure 3E). These data represent the first demonstration of RNAi in F. gigantica. Note that we did not examine suppression of FgigσGST protein, nor any of the other genes referred to in this study, in F. gigantica.
Triggering highly significant levels of transcript knockdown over the 72 h periods described above did not impact on the survival or behaviour of NEJs maintained in vitro. Supporting this observation, western blot analyses of FheCatB, FheCatL and FheσGST protein performed at 18 h and 72 h confirmed that target protein levels had not changed significantly at these timepoints. Extension of the in vitro maintenance period enabled detection of FheσGST protein suppression after 9 days (61.7±12.43%, p<0.05, n = 4; Figure 4A), which had increased further by 21 days (24.60±2.49%, P<0.001, n = 3; Figure 4A). Suppression of FheCatL protein was not detectable until 21 days (58.97±6.74, P<0.01, n = 3; Figure 4B). FheCatB protein levels were not reduced vs controls at all time points up to and including 21 days (Figure 4C).
PCR analyses showed persistence of highly significant transcript knockdown for all three targets throughout a 25 day period of maintenance in vitro (day 25 FheCatB: 0.03±0.01, P<0.001, n = 5; FheCatL: 0.12±0.04, P<0.001, n = 5; FheσGST: 0.01±0.003, P<0.001, n = 4; Figure 5), suggesting that inadequate target transcript knockdown is probably not responsible for the absence of detectable FheCatB protein suppression at later time points. Notably, even in experiments where suppression of FheCatL or FheσGST protein was detectable, over maintenance periods up to 25 days, we observed no changes in NEJ survival or behaviour compared to time-matched negative controls.
All long dsRNAs tested triggered reproducible, concentration-dependent knockdown of target transcripts. Although our standard soak protocol employed 50 ng/µl long dsRNA, effective transcript knockdown was also achievable using concentrations an order of magnitude lower (5 ng/µl: FheCatB, 0.09±0.01, P<0.01, n = 3; FheCatL, 0.45±0.12, P<0.05, n = 5; FheσGST, 0.20±0.08, P<0.01, n = 4; Figure 6A–C). Statistically-significant transcript suppression relative to control treatments was apparent with as little as 0.05 ng/µl long dsRNA in the case of FheCatB (0.38±0.15, P<0.05, n = 3; Figure 6A). These observations are of practical importance in permitting the possibility of combinatorial RNAi (i.e. the silencing of multiple targets in parallel), an approach that was demonstrated here by combining FheCatB, FheCatL and FheσGST long dsRNAs to a final total concentration of 50 ng/µl ( = each individual dsRNA at 16.7 ng/µl). Under Protocol IV conditions, this method triggered knockdown of all three targets in parallel to a level not significantly different from that achieved in individual soaks (Figure 7A). No phenotypic impacts were observed during the 72 h combinatorial RNAi timecourse, although we did not investigate impacts on protein suppression during these experiments.
In order to establish the efficacy of siRNAs in comparison to long dsRNA, we employed three individual siRNAs complementary to FheσGST (siGST1–3). All three siRNAs triggered highly significant knockdown of FheσGST transcript at 50 ng/µl (siGST1: 0.08±0.003, P<0.0001, n = 3; siGST2: 0.17±0.04, P<0.001, n = 3; siGST3: 0.14±0.03, P<0.001, n = 3; Figure 6D). Titration of the most potent siRNA, siGST1, showed concentration-dependent effects that were significant at a concentration 10-fold more potent than long dsRNA (siGST1 5 ng/µl: 0.34±0.03, P<0.01, n = 5; 0.5 ng/µl: 0.45±0.08, P<0.05, n = 5; Figure 6D).
RNAi off-target impacts were assessed by amplifying non-target amplicons from long dsRNA-treated cDNAs (Figure 7B). While these analyses showed no evidence for off-target knockdown by any of our long dsRNAs, we did observe some up-regulations of transcripts following treatment with non-target long dsRNAs. This phenomenon was only statistically-significant in the case of the impact of FheCatL dsRNA on FheCatB. Further experimentation is needed to fully characterise the consequences, and biological significance, of such up-regulation on target protein levels.
Development of RNAi-based gene silencing methods is pivotal for the effective exploitation of burgeoning database resources for parasitic helminths. Schistosoma spp. trematodes provide arguably the best example of how the effective confluence of genomic resources with powerful RNAi protocols can promote the systems-level understanding of helminth parasite biology [25]. Fasciola spp. liver fluke are increasingly the focus for genome and transcriptome sequencing projects [9]–[12], but have not seen the widespread adoption of RNAi and related tools by the liver fluke research community. As a means of promoting such uptake, this study aimed to develop a standardised set of RNAi protocols applicable to NEJs of a widely used F. hepatica strain. We have shown that transcriptional knockdown can be triggered in seven virulence gene targets, simply by soaking NEJs in a solution of tissue culture media containing long dsRNA, or siRNA, trigger molecules. In three targets subject to in-depth optimisation of variables, transcript knockdown occurred in a concentration-dependent manner that was invariably rapid and persistent. Although target-specific RNAi dynamics were apparent in terms of the rapidity of transcriptional response to dsRNA trigger, these dynamics manifested most notably in the length of the lag phase between transcript knockdown and protein suppression. This lag phase was appreciable in all cases, with FheσGST protein suppression developing only 9 days following exposure to long dsRNA, FheCatL suppression was not evident until 21 days post-exposure. We were unable to detect FheCatB suppression at any point up to and including 21 days. These manipulations did not trigger any measurable changes in NEJ viability or behaviour in vitro. Nevertheless, the protocols described here represent a relatively simple means to achieve targeted transcript and protein knockdowns in the absence of advanced manipulations such as electroporation [7], [26], biolistics [27], transfection reagents [28], or genetic transformation [29], such as have been described in other trematodes. Alongside development of appropriate functional assays, these RNAi protocols should enable gene function analysis even by laboratories with limited molecular expertise, and may lend themselves to multiple-throughput screening approaches to target validation.
Despite the successful knockdowns reported here, our data contrast with those of a previous study [8], which described extremely rapid RNAi dynamics in F. hepatica NEJs of both Oberon and Fairhurst isolates (see [30] for discussion of fluke isolates). In the case of cathepsins B and L, McGonigle et al. [8] demonstrated suppression of both target transcript and protein after just 4 h exposure to long dsRNA, correlating at this time-point with reduced migration through in an in-vitro rat gut penetration assay. Although we attempted initially to replicate the molecular aspects of those experiments using quantitative methods, we employed a commercially available and widely used strain of F. hepatica, the US Pacific North-West Wild strain, instead of Oberon and Fairhurst isolates. After 4 h exposure to long dsRNA, we observed reliable knockdown only of FheCatB transcript (Figure 2B). FheCatL remained refractory to transcript knockdown after 4 h exposure, even over multiple replicates performed over several days using independently prepared batches of long dsRNA. Furthermore, throughout all of the experiments described here, we never observed any of the “erratic locomotion and paralysis” phenotypes reported by McGonigle et al. Even in soaks employing 1 µg/µl FheCatL dsRNA (10 fold higher than that used by McGonigle et al.), US Pacific North West strain NEJs displayed no visibly different behaviour from controls, and no FheCatL transcript knockdown, over a 4 h exposure period. After addressing the relevant experimental variables between our studies, the primary difference that remained was our use of different F. hepatica isolates. Variable RNAi susceptibility has been observed between strains of the nematode, Caenorhabditis elegans [31], where inter-strain differences in somatic RNAi competency correlate with differences in the capacity to mount an anti-viral response. A similar situation could conceivably exist between genetically-distinct liver fluke isolates, related to sequence-level differences in RNAi pathway proteins, although we have been unable to test this hypothesis directly in F. hepatica Oberon and Fairhurst isolates. However, planned draft genome sequences from flukes of diverse genetic backgrounds, and anthelmintic susceptibilities (Steve Paterson, Jane Hodgkinson, personal communication) may shed further light on potential isolate-related differences in RNAi mechanics. While strain/isolate-specific RNAi susceptibilities could explain differences in transcriptional responses to RNAi triggers as discussed here, based on our current knowledge of RNAi mechanisms in other organisms they do not adequately explain the differences in rates of protein suppression observed between this work and that of McGonigle et al. [8]. One may speculate that the proteins in question may exhibit different half-lives in the distinct isolates used in the respective studies, i.e. different post-RNAi protein suppression rates could be due to longer target protein half-life in the Pacific North West strain than in the Oberon/Fairhurst isolates.
The transcript knockdown data described here clearly illustrate that RNAi can be triggered using very simple methodology in NEJs of the US Pacific North West Wild strain of F. hepatica. Burst exposure experiments (Figure 3), and titrations of long dsRNAs and siRNAs (Figure 6), show that even limited temporal exposure to (30 min) or low concentrations (0.05 ng/µl) of long dsRNA are sufficient to trigger the development of transcript knockdown over a 72 h period. Transcript knockdowns also persist at highly significant levels over maintenance periods of at least 25 days – this is reminiscent of S. mansoni, where cathepsin B knockdown was reportedly sustained for at least 40 days following delivery of long dsRNA by electroporation [19]. The transcriptional acquiescence of NEJs to exogenously triggered RNAi suggests the presence of a suite of efficient cellular RNAi mechanisms, presumably involving a rapid and efficient dsRNA uptake mechanism. This mechanism may involve SID (Systemic RNAi Defective) homologues, dsRNA-gated channels responsible for transmembrane-transport of dsRNA molecules, identified originally in C. elegans [32]–[34] with orthologues described in S. mansoni and several other disparate genera [35]–[41]. The development of transcript knockdown over several days following exposure to dsRNA suggests a slow intercellular spread of dsRNA trigger from point of entry, and/or the presence of an RNA-dependent RNA polymerase (RdRP)-like secondary siRNA-based amplification system (although RdRPs have not yet been described in flatworms, our unpublished bioinformatics analyses do suggest the presence of RdRP-like sequences in available fluke datasets).
Throughout this study, no impacts were observed on NEJ viability or behaviour during RNAi experiments. These observations were supported in early experiments by western blots performed at 18 h and 72 h, which revealed no changes in levels of FheCatB, FheCatL or FheσGST protein. Given that some schistosome genes require maintenance periods of up to 14 days post-electroporation before evidence of protein suppression or phenotype [42], we hypothesised that protein suppression could be similarly achieved in our targets simply by maintaining worms for longer in vitro following dsRNA exposure, in order to permit run-down of protein levels. Under such extended maintenance periods, a western blot-based approach detected FheσGST suppression at 9 and 21 days post-exposure, while FheCatL suppression was not detectable until 21 days post-exposure. No suppression of FheCatB protein was detected at any point during the 21-day timecourse, although an apparent upregulation of cathepsin B was detected in our day 9 samples. Although we might assume that these differential dynamics simply reflect the dissimilar run-down rates of functionally distinct proteins, differences in the sequence diversity of these targets might equally impact on our ability to detect changes in protein abundance: FheσGST is a single copy gene for which our long dsRNA had very little sequence identity to other transcripts, and for which we had access to a specific FheσGST-antiserum. In this case, detection of protein suppression is a relatively simple matter, where changes in immunoblot band density can be expected to provide an accurate read-out for changes in protein abundance. In the case of multi-gene cathepsin families, the ability to trigger and detect promiscuous changes in protein abundance may depend on the relative specificity/cross-reactivity of dsRNA and antisera, both within and between members of the target gene family. In the case of cathepsin L, where 78% mean sequence similarity exists between the seven recognised F. hepatica cathepsin L clades, including many regions of identity of >20 nt (consistent with generation of siRNAs by Dicer), long dsRNA targeted to any clade is likely to trigger promiscuous knockdown of the broader gene family. Polyclonal antisera are likely to be similarly promiscuous across the multiple cathepsin L protein clades, hence our detection of FheCatL protein suppression by 21 days post dsRNA exposure. In the case of cathepsin B however, sequence similarity across the ten recognised clades is somewhat lower (mean 53%), and our long dsRNA (in this study, complementary to FheCatB2) does not possess significant stretches of identity with the other clades, reducing the likelihood of promiscuous, broad knockdown of the FheCatB gene family. In contrast, the cathepsin B polyclonal antiserum used here is likely to cross-react with several cathepsin B clades. Under these circumstances, suppression of FheCatB2 alone may not have been detected by our immunoblot methodology unless it accounted for the major proportion of expressed cathepsin protein in the tissue extract (which is unlikely [43]). This issue could be addressed quite simply in future studies by employing a cocktail of dsRNAs representing all of the FheCatB clades expressed in NEJs [44]. The combinatorial RNAi data presented in Figure 7A, showed that at least three targets could be silenced in parallel, supporting the feasibility of such an approach. An alternative to the immunodetection approaches used here is 2D electrophoresis-based sub-proteomics, which has shown promise in detecting RNAi-induced suppression of individual proteins in multi-gene families in other systems (LaCourse et al., 2008 [45]). Another technical consideration is that even where FheCatL protein suppression was detected, our blots showed that it was most apparent in the bands representing the 24.5 KDa mature enzyme and 30 KDa processed intermediate, with little reduction in band density apparent for the 37 KDa preproenzyme (Figure 4). This phenomenon is likely due to the storage of cathepsin L preproproteins as inactive zymogens within secretory vesicles (a location likely to limit proteolytic breakdown), prior to exocytotic release [43], [46]–[49]). This may also have inhibited detection of FheCatB protein suppression, since only the preproprotein form was detected by our cathepsin B antiserum. Alternatively, the apparently long protein half-lives seen here in the face of transcript ablation may simply reflect our inability to provide adequate maintenance conditions for obligate intra-mammalian parasites, which has resulted in abnormal cell division and cellular protein dynamics that artificially inhibit protein turnover. Given the high and sustained levels of transcript knockdown reported here, it seems unlikely that limiting transcript knockdown is responsible for the apparent lack of protein suppression in our experiments. Nevertheless, it is possible that the delivery of dsRNA by other methods such as electroporation, biolistics, transfection reagents, or genetic transformation [18], [19], [27]–[29] may increase transcript knockdown levels, hastening the validity of phenotypic measurements. Varying the mechanism of dsRNA delivery is a factor that could be addressed by future experiments.
Despite detectable suppression of two out of three target proteins, we observed no impact of RNAi on NEJ behaviour or viability. Although these observations contrast with previous reports of paralysis and compromised motility following RNAi of cathepsin B and L in liver fluke [8], they are consistent with silencing of these virulence genes in other organisms. Where RNAi has been used to target cathepsin B, cathepsin L or GSTs in multicellular eukaryotes in an in vitro setting, phenotypic consequences have been reported only under functionally-defined assay settings such as feeding/growth [26], [29], [50], [51], drug susceptibility [52], [53], or in dynamic systems measuring reproductive output or molting [54]–[57]. Under settings where organisms are not subject to functional assay but merely observed in the presence of RNAi (such as the present study), no impacts on survival or motility have been reported following RNAi of cathepsin B, cathepsin L, or GSTs [20], [58]–[62]. The lack of impact on survival/viability following RNAi in vitro is not necessarily a poor reflection on the therapeutic potential of these targets, merely a confirmation of the complexity and subtlety of the host-parasite interface that highlights the need for adequately sensitive assay systems. In vivo assays, where RNAi impacts on worm virulence are assessed via the ability of an RNAi-treated parasite to develop to patency within a host organism, have been employed in helminths [25], [54], [63]–[66]. Although in vitro RNAi results cannot always be recapitulated in vivo [64], the persistent knockdowns achieved in the current study suggest the amenability of F. hepatica NEJ RNAi studies to such an in-vivo infection assay. Efforts to develop in vivo, and tailored in vitro assays for several targets are currently ongoing in our laboratories. These assays will permit the development of RNAi to its full potential in liver fluke.
This study has shown that RNAi protocols, based simply on soaking in long dsRNA or siRNA even for brief periods, are capable of triggering robust, persistent transcript knockdown in liver fluke NEJs. However, notes of caution are offered in terms of the possibility of liver fluke isolate-specific differences in RNAi mechanics, and evidence for variable lag periods between transcript and protein suppression; the latter highlight the need for careful assessment of target dynamics prior to the application and interpretation of phenotypic assays. The simple methods described here should be widely applicable within the liver fluke research community, and should bolster efforts both to investigate liver fluke gene function and identify novel targets for therapy.
|
10.1371/journal.pntd.0007225 | Leptospirosis in sugarcane plantation and fishing communities in Kagera northwestern Tanzania | Leptospirosis is a bacterial zoonotic disease of worldwide importance, though relatively neglected in many African countries including sub Saharan Africa that is among areas with high burden of this disease. The disease is often mistaken for other febrile illnesses such as dengue, malaria, rickettsioses and enteric fever. Leptospirosis is an occupational disease likely to affect people working in environments prone to infestation with rodents which are the primary reservoir hosts of this disease. Some of the populations at risk include: sugarcane plantation workers, wetland farmers, fishermen and abattoir workers. In this study we investigated the prevalence of antibodies against Leptospira among sugarcane plantation and factory workers, fishing communities as well as among rodents and shrews in domestic and peridomestic environments within the study areas.
The study was conducted in Kagera region, northwestern Tanzania and it involved sugarcane plantation workers (cutters and weeders), sugar factory workers and the fishing community at Kagera Sugar Company in Missenyi district and Musira island in Lake Victoria, Kagera, respectively. Blood was collected from consenting human adults, and from rodents and shrews (insectivores) captured live using Sherman traps. Serological detection of leptospiral antibodies in blood serum was carried out by the microscopic agglutination test (MAT).
A total of 455 participants were recruited from the sugarcane plantation (n = 401) and fishing community (n = 54) while 31 rodents and shrews were captured. The overall prevalence of antibodies against Leptospira in human was 15.8%. Sugarcane cutters had higher seroprevalence than other sugar factory workers. Prevalent antibodies against Leptospira serovars in humans were against serovars Lora (6.8%), Sokoine (5.3%), Pomona (2.4%), Hebdomadis (1.1%) and Kenya (0.2%). Detected leptospiral serovars in reservoir hosts were Sokoine (12.5%) and Grippotyphosa (4.2%). Serovar Sokoine was detected both in humans and small mammals.
Leptospirosis is a public health threat affecting populations at risk, such as sugarcane plantation workers and fishing communities. Public awareness targeting risk occupational groups is much needed for mitigation of leptospirosis in the study areas and other vulnerable populations in Tanzania and elsewhere.
| Leptospirosis is caused by a spirochete bacterium of the genus Leptospira. The disease is worldwide distributed, although highly neglected in some parts of the developing world. This study investigated the prevalence of antibodies against Leptospira in sugarcane plantation workers, a fishing community and rodents and shrews in the Kagera region, northwestern Tanzania. Seventy two of the 455 (15.8%) screened participants were seropositive to leptospirosis. The seroprevalence was higher among sugarcane cutters (18.4%) than other plantation workers, and 15.4% of hospitalized patients in the plantation hospital were seropositive. Prevalence of antibodies against Leptospira in the fishing community was 14.8%. Antibodies against Leptospira serovars detected in humans with their respective proportions, in brackets, were Lora (6.8%), Sokoine (5.3%), Pomona (2.4%), Hebdomadis (1.1%) and Kenya (0.2%). In the small mammals, the most detected antibodies were against Leptospira serovar Sokoine (12.5%) and serovar Grippotyphosa (4.2%). These results show that leptospirosis is public health risk requiring attention of the health system as well as the agricultural sector for its management.
| Leptospirosis is a public health concern especially in the tropical and subtropical countries where the environment is optimal for survival of pathogenic leptospires [1]. The annual morbidity and mortality caused by leptospirosis worldwide is estimated to be 14.7 cases per 100,000 population [2]. Globally, Oceania region has the highest disease burden (150.6 cases/100,000 population), South east Asia (55.5), Caribbean (50.6) and East Sub Saharan Africa (25.6) [2, 3]. In Tanzania the annual incidence is 75–102 cases per 100,000 population [4]. Rodents are considered major reservoirs of Leptospira [5] and other wild animals and birds found in wetland areas may also carry and spread leptospires into the surroundings [6]. The disease is associated with certain occupational activities such as rice and sugarcane farming, fishing and fish farming, livestock keeping, handling animal products and water sports [7, 8]. Males are most affected than females contributing to 80% of the total burden [3]. Humans can be infected through contact with urine or other materials from infected animals or contaminated water and soil [9]. In Tanzania, leptospirosis has been reported in patients with non-malaria fevers [10, 11] and in animals including rodents and domestic animals [12–15]. Antibodies against Leptospira have been demonstrated also in freshwater fishes [6] in Tanzania suggesting potential risk to fishermen and people undertaking irrigation activities such as, rice farming and sugarcane plantation. Studies on leptospirosis in these at risk populations are lacking, hence the burden of leptospirosis in fishing communities and sugarcane plantations is not known. Sugarcane plantation and rice farming are important agricultural sectors in Tanzania, which engage permanent and seasonal workers from different parts of the country. Understanding the burden of leptospirosis in these occupational groups could provide baseline information needed for informing policy, especially because the disease is neglected and rarely considered for diagnosis in the health system [16]. In this study we investigated the serological prevalence of leptospirosis in selected risk populations of sugarcane plantation workers and fishing in northwestern Tanzania Also, we identified potential Leptospira serovars circulating in the region, which would serve as important antigens for diagnostic purposes.
This study was conducted in Kagera region northwestern Tanzania at Musira island (S 01° 19.914’, E 031° 49.772’ with elevation of 1120 meter above sea level, and at Kagera sugar company (S 01° 12.807, E 031° 16.510’) with elevation of 1157 meter above sea level. Kagera region receives bi-modal rainfall pattern ranging between 900–2,000 mm per annum, temperatures range between 20°C and 28°C. Kagera region is located along Lake Victoria hence fishing is among major socio-economic activity apart from large scale sugarcane plantation. Kagera Sugar Company is one of the biggest sugarcane plantations in the country. The two study sites (fishing community and Musira and sugarcane plantation community at Kagera sugar company are 76.2 km apart (Fig 1).
The ethical clearance for conducting this study was obtained from the Medical Research Coordinating Committee of the National Institute for Medical Research (NIMR), Certificate No. NIMR/HQ/R.8a/Vol.IX/2453, as well as from the Kilimanjaro Christian Medical University College, Research Ethics and Review Committee (CRERC), Moshi Tanzania (Ref. No. 993). Permission was also sought from local authorities in the study area.
A total of 455 participants were sampled of which 401 (132 females and 269 males) were from sugarcane plantation and 54 (16 females and 38 males were from the Musira fishing island in Lake Victoria. The demographic profile of human participants was as shown in Table 1.
The majority of rodents were Rattus spp. (55%) trapped indoors. Other rodents trapped included forest species (Lophuromys spp.) captured in the bushes near the sugarcane plantation and Arvicanthis spp. found in fallow land near sugarcane fields (Table 2).
Prevalence of human leptospirosis in the two study populations of sugarcane plantation workers and fishing community was 15.8%. Fifty eight of the 317 (18.3%) sugarcane cutters were seropositive compared to 8 out of 54 (14.8%) of the fishing community subjects. Two of the 13 (15.4%) hospitalized patients were seropositive while other participants including office cleaners, petty traders and security guards contributed to 7.0% seropositivity (Table 1).
Antibodies against tested Leptospira were relatively lower in rodents than in humans. The highest titre (1:2560) was observed in two individuals against serovar Pomona (Fig 2).
Prevalence of antibodies against Leptospira among different occupational groups, populations, gender and age groups were compared to determine whether certain groups were at more risk than others. The prevalence of antibodies against Leptospira between male and female participants was 6.4%, which was not statistically significant (p = 0.0800, 95% CI = -0.8658 to 12.6811, χ2 = 3.065, df = 1). The prevalence of antibodies against Leptospira in participants in age group of 18–37 year and 38–57 year old differed by 2.2% which was not statistically significant (p = 0.6206, 95% CI = -5.5629 to 12.5150, χ2 = 0.245, df = 1). There was significant difference in percentage of positive individuals (40%) between participants in age group of 18–37 year and ≥58 year old (40.0%) (p = 0.0003, 95% CI = 13.1776 to 64.4124, χ2 = 12.898, df = 1). Age group of 38–57 yrs and >58 year old also showed significant difference in percentage of positives (37.8%) that was also statistically significant (p = 0.0044, 95% CI = 9.5238 to 62.8984, χ2 = 8.105, df = 1). The prevalence of antileptospiral antibodies between fishing community and sugarcane cutters was 3.6% that is not statistically significant (p = 0.5241, 95% CI = -8.8082 to 12.0938, χ2 = 0.406, df = 1). Comparison of positive rate found in fishing community and unexposed group (others) showed 9.3%, which was not statistically significant (p = 0.0777, 95% CI = -1.2424 to 21.5463, χ2 = 3.111, df = 1). The difference in positive rate between sugarcane cutters and unexposed group (others) was 12.9%, which was statistically significant (p = 0.0068, 95% CI = 4.1963 to 18.6242, χ2 = 7.329, df = 1). Hospitalized participants and unexposed group (others) showed a difference of 9.9% that was not statistically significant (p = 0.1999, 95% CI = -3.6319 to 36.9580, χ2 = 1.643, df = 1). Hospitalized participants and hospital staff also showed a difference of 1.1% that was not significant (p = 0.9490, 95% CI = -37.5454 to 30.4008, χ2 = 0.004, df = 1).
This study shows high prevalence of antibodies against Leptospira in humans involved in sugar production and fishing in the Kagera region, northwestern Tanzania. Leptospirosis in rodents and shrews captured in the areas is also reported.
Findings suggests that sugarcane plantation workers especially sugarcane cutters and fishing communities are potentially at risk. A prevalence of 15.8% was found in sugarcane plantation workers, with cane cutters having the higher prevalence of 18.4%, followed by other plantation workers and hospitalized patients. Prevalence of anti-leptospiral antibodies was also high (14.8%) in fishermen and other individuals living on the Musira island, which is a fishing island. This suggests that fishing communities can get leptospirosis following contact with water contaminated with urine of the reservoir hosts. The prevalence of human leptospirosis in sugarcane plantation workers reported in this study (18.4%) is lower than that reported in sugarcane plantation workers in central America (59%) [19] but higher than the 0.7% prevalence reported from Trinidad and Tobago [23].
Prevalence of antibodies against Leptospira among different occupational groups, populations, gender and age groups showed variations suggesting that individuals belonging to certain groups and occupation groups have different levels of risk of contracting leptospirosis. For example, while there was no significant difference in the prevalence of leptospiral antibodies between male and females despite that the study had more males than females due to the nature of the occupation of the study populations, there was a significant difference in prevalence of antibodies against Leptospira found in participants in two age groups of 18–37; 38–57 year old versus participants with age above 58 year old. Findings show that participants with over 58 year old have significantly higher proportion of antibodies against Leptospira than those with age below 58 year old (i.e. 18–37; 38–57 year old). This could be probably associated with potential prolonged exposure to risk environment such as sugarcane cutting for many years than newer entrants. The fishing community and sugarcane plantations appear to have similar risk levels since the prevalence of antibodies in these two populations was not statistically significant. However, comparison of fishing community and sugarcane cutters considered risk populations with unexposed groups consisting of participants engaged with less risk activities such as office work, security and petty traders show that fishing community does not differ to the unexposed group while sugarcane cutters show more risk than unexposed group. This can be explained with fact that fishing community included the general population of the fishing island including school pupils and other residents likely to have various levels of risk of contracting leptospirosis while sugarcane cutters consisted a uniform group of individuals engaged with same activity of cutting sugar hence likely to have same level of risk higher than the general population.
The prevalent antibodies against Leptospira serovars found in humans were against Leptospira interrogans serovar Lora (6.8%), L. kirschneri serovar Sokoine (5.3%) and slightly Leptospira interrogans serovar Pomona (2.4%). Leptospira interrogans serovar Hebdomadis and L. borgpetersenii serovar Kenya were least found with prevalence of 1.1% and 0.2%, respectively. Leptospira kirschneri serovar Sokoine and L. kirschneri serovar Grippotyphosa were frequently found in both humans and animals as previously reported [8, 15] in agro-pastoralists communities living in Katavi-Rukwa ecosystem [8] indicating a wider distribution of leptospirosis in Tanzania.
These findings shows that the roof rat (Rattus spp.) is an important reservoir of leptospirosis in Kagera region as demonstrated by high positivity rate among the house rats collected in different localities in the study areas. Comparison of positive rates found in the roof rats and an insectivore showed no statistically significant difference due to small sample size of rodents and shrews collected. A larger sample size estimated for this study was not achieved due to seasonal variations in rodent populations hence suggesting further sampling to enhance robust determination of the major reservoir of Leptospira in this region. Antibodies against L. serovar Kenya, Lora, Pomona and Hebdomadis were not detected in rodents nor insectivores. The rats were seropositive against L. kirschneri serovars Sokoine and L. kirschneri serovar Grippotyphosa. Rodents had lower antibody titres (1:20–1:40) than humans in which higher titres up to 1:2560 were determined by MAT which is the gold standard test for leptospirosis diagnosis [9, 24]. High antibody titres against Leptospira serovars detected in humans suggest the existence of recent infections.
The predominant circulating Leptospira serovars which antibodies against was detected in humans, namely Leptospira interrogans serovar Lora, L. kirschneri serovar Sokoine, L. interrogans serovar Pomona, L. interrogans serovar Hebdomadis and L. borgpetersenii serovar Kenya have been previously reported in humans, rodents and domestic animals [10, 15, 25]. Leptospira kirschneri serovar Sokoine was mainly found in both humans and animals in Tanzania whereas L. interrogans serovar Grippotyphosa was mainly detected in the reservoir hosts. Leptospira interrogans serovar Lora was not detected in rodents, indicating potential diversity of sources of human infection. It is known that certain Leptospira serovars demonstrate host-specificity and might be absent in certain rodent species [15]. Further investigations are needed to establish the source or reservoir hosts of serovar Lora in the study areas and to determine whether the plantation workers who also come from outside Kagera had leptospirosis exposure prior to their recruitment at the sugarcane plantation. This could be achieved by including leptospirosis screening during general health examinations performed before recruiting cane cutters.
The observed high prevalence of leptospirosis in the fishing community corroborate previous report of high seropositivity/leptospiral antibodies in freshwater fishes and thus potential risk of leptospirosis in fishing communities and in people working in the fishing industry [15, 19, 23]. It is recommended that leptospirosis control should include rodent management, and public awareness. Furthermore, leptospirosis screening should be introduced in risk occupational groups in Tanzania and elsewhere where the disease is neglected [16]. Detection of leptospiral antibodies in hospitalized patients during this study indicates further the importance of considering leptospirosis among febrile illnesses that are non-malarial. The prevalence of 15.4% of leptospirosis in hospitalized patients corroborates previous reports from northern Tanzania and Morogoro among hospitalized patients with febrile illness [10, 11, 26]. This further emphasizes the need to include leptospirosis in differential diagnosis of febrile illnesses.
Further surveillance studies are needed to isolate and characterize the disease causative Leptospira serovars beyond serological surveillance. These should include cross agglutination absorption test, and molecular typing [25]. The major limitations of this study were failure to isolate the causal agent, which would have enabled its characterization. Similarly, future studies should include larger populations of potential reservoirs.
Leptospirosis is a public health threat in sugarcane plantation workers and the fishing communities. Preventive measures are needed to mitigate risks of leptospirosis. These should include rodent control, public awareness and screening for leptospirosis in individuals with non-malarial fevers [16] and vulnerable occupational groups such as sugarcane cutters. Leptospira serovars Lora, Sokoine, Pomona, Hebdomadis, Kenya and Grippotyphosa should be included as antigens for broader leptospirosis screening in humans and animals from this region.
|
10.1371/journal.pbio.0050271 | Conformational Motion of the ABC Transporter MsbA Induced by ATP Hydrolysis | We measured the amplitude of conformational motion in the ATP-binding cassette (ABC) transporter MsbA upon lipopolysaccharide (LPS) binding and following ATP turnover by pulse double electron-electron resonance and fluorescence homotransfer. The distance constraints from both methods reveal large-scale movement of opposite signs in the periplasmic and cytoplasmic part of the transporter upon ATP hydrolysis. LPS induces distinct structural changes that are inhibited by trapping of the transporter in an ATP post-hydrolysis intermediate. The formation of this intermediate involves a 33-Å distance change between the two ABCs, which is consistent with a dimerization-dissociation cycle during transport that leads to their substantial separation in the absence of nucleotides. Our results suggest that ATP-powered transport entails LPS sequestering into the open cytoplasmic chamber prior to its translocation by alternating access of the chamber, made possible by 10–20-Å conformational changes.
| Clinical multidrug resistance in the treatment of bacterial and fungal infections and cancer chemotherapy can result from the expression of pumps that extrude toxic molecules from the cell. A subclass of these pumps—ATP-binding cassette (ABC) transporters—use energy from ATP to remove a wide range of molecules. MsbA is a conserved ABC transporter from Gram-negative bacteria with sequence similarity to human multi-drug ABC transporters. MsbA flips the building block of the outer membrane, lipid A, across the inner membrane. The input of ATP energy occurs in two dedicated nucleotide-binding domains (NBDs), whose configuration in intact transporters is controversial. We determined the amplitude of MsbA conformational motion that couples energy expenditure to substrate movement across the membrane. Using molecular probes introduced into the protein sequence, we found that ATP hydrolysis fuels a relative motion of the NBDs close to 30 Å. The movement of the NBDs is coupled to reorientation of the chamber, which binds the lipid substrate from cytoplasmic-facing to extracellular-facing through large amplitude motion on either side of the transporters. In addition to revealing the structural mechanics of transport, these results challenge current models deduced from studies of substrate-specific ABC importers that envisions the two NBDs in contact throughout the ATP hydrolysis cycle.
| ATP-binding cassette (ABC) transporters harness the free energy of ATP hydrolysis to power the thermodynamically unfavorable trafficking of a wide spectrum of substrates in and out of the cell [1–3]. Cooperative ATP binding and hydrolysis occur in a molecular motor composed of two ATP-binding and hydrolysis cassettes (ABCs), also referred to as nucleotide-binding domains (NBDs) [4]. ATP binds at the interface of an NBD dimer sandwiched between the Walker A motif of one subunit and the signature motif of the symmetry-related subunit [5–7]. Two transmembrane domains encode the determinants of substrate binding and provide a passageway across the bilayer.
In Gram-negative bacteria, the transport of lipid A from its site of synthesis across the inner membrane is critically dependent on the expression of the ABC transporter MsbA. Loss of MsbA activity inhibits growth and is associated with the accumulation of lipid A in the cytoplasmic leaflet of the inner membrane [8–10]. MsbA has sequence similarity to a subclass of ABC transporters that is linked to the development of multidrug resistance in microorganisms and cancer through the extrusion of structurally dissimilar molecules [11,12]. Polyspecifity appears to be a common property of ABC efflux systems, whereas importers are substrate specific, often requiring a dedicated high-affinity binding protein for substrate delivery [1].
The molecular organization of the four domains of ABC transporters was gleaned from crystal structures of a number of importers as well as the bacterial multidrug efflux system Sav1866 [13–15]. The structures confirm the canonical interface observed in isolated NBD dimers [4,16], identify structural elements in the cytoplasmic side that mediate communication between the NBDs and the transmembrane domains, and define the likely pathway of substrate transport and putative gates that control substrate access. Initial structures of MsbA proved incompatible with biochemical and structural data and were subsequently retracted owing to an analysis error [17–20]. A large body of biochemical studies including cryo–electron microscopy (EM) analysis [21], cross-linking studies of P-glycoprotein (P-gp) [22–25], and kinetic and thermodynamic analysis of P-gp and LmrA substrate transport cycles [2,26] have delineated many aspects of the transport mechanism. They collectively demonstrate that the energy input of ATP binding and/or hydrolysis is transduced to the mechanical work of an inward-outward–facing cycle of the substrate binding sites [27].
Structural studies of MsbA by site-directed spin labeling [28] and electron spin resonance (ESR) spectroscopy suggested that in liposomes, MsbA undergoes substantial conformational changes upon ATP binding and hydrolysis [29]. Reporting on the accessibilities and relative proximities of three transmembrane helices—2, 5, and 6—and adjacent regions of the intracellular domain and periplasmic loops, the spin labels revealed the presence of an asymmetric, water-exposed chamber that is open to the cytoplasm in the absence of nucleotides. ATP binding or hydrolysis occludes the chamber to the cytoplasm and increases hydration in the periplasmic side along an alternating access model [29]. Accessibility changes are accompanied by opposite proximity changes on the cytoplasmic and periplasmic sides of the transporter, although the amplitude of these movements was not determined. Conformational changes induced by either ATP binding or by the formation of a ADP/Vanadate (Vi) post-hydrolysis intermediate are of similar sign and magnitude, suggesting that the ATP binding provides the power stroke for transport as previously reported for P-gp [21]. The ESR constraints indicate that apo-MsbA samples conformations that depart from the reported crystal structures in helix topology and the extent of opening on the extracellular side [29]. In particular, these experiments indicated that helix 6 is shielded from direct exposure to the bilayer and is likely packed at the dimer interface as concluded from cross-linking studies of P-gp and subsequently confirmed by the Sav1866 structure [13,30].
Recent crystal structures have captured importers in inward-facing and outward-facing conformations and presented a model of the amplitude and extent of the underlying structural changes [15]. A central theme of this model is the limited rearrangement of the NBD dimer interface upon ATP binding and hydrolysis. Whether this model applies for efflux ABC transporters is yet to be determined. The Sav1866 structure [13] corresponds to a post-hydrolysis intermediate; thus it does not define the amplitude of the movement associated with chamber reorientation nor does it address the critical question of whether the two NBD domains remain closely packed during the ATPase cycle. In addition, the crystal structures were all determined in detergent micelles and thus may be influenced by the absence of lipids in addition to conformational selectivity imposed by crystal lattice forces [15,31].
To further map the conformational changes in the transport cycle of MsbA, we used pulse dipolar ESR spectroscopy [32–36] and fluorescence homotransfer [37–39] to obtain a set of critical distance constraints that monitor the relative separation of the transmembrane domains, the dimer interface, and the packing of the NBDs in detergent micelles and in the native-like environment of lipid bilayers. The distances were determined following the addition of a putative substrate, LPS, and in the high-energy post ATP-hydrolysis intermediate in the presence and absence of LPS. The change in the distance constraints provides evidence of LPS interaction with the transporter and establishes the sign and amplitude of the conformational changes on the cytoplasmic and periplamic sides following ATP hydrolysis.
Spin and fluorescence labels were attached to single cysteines in each MsbA monomer, resulting in the introduction of two symmetry-related probes in the functional dimeric unit. Distances between the probes were determined by ESR and fluorescence spectroscopies in detergent micelles. For a selected set, distances between spin labels were also measured in liposomes to establish the correspondence with conformational changes in detergent micelles. The mutants selected for this study form dimers as demonstrated by their retention times on size-exclusion chromatography. Spin-labeled mutants were shown previously to turn over ATP with rates ranging from 20%–100% of that of the wild type (WT) and at least 10-fold higher than the Vi-inhibited WT [29].
Three nucleotide-bound intermediates of MsbA can be stably populated: ATP-bound, ADP-bound, and ADP/Vi-inhibited. The latter is a high-energy post-hydrolysis intermediate often referred to as the transition state of ATP hydrolysis [40]. Previous site-directed spin labeling data indicated that ATP binding and hydrolysis are associated with similar overall conformational changes in MsbA [29]. Therefore, here we focus on the ADP/Vi intermediate.
The sites for distance measurements were selected in regions that report changes in the local environment of spin labels [29] following the formation of the ADP/Vi intermediate. These include the intracellular or cytoplasmic sides (IL in the notation of Dawson and Locher [13]) of helices 2, 5, and 6, where an overall reduction in accessibility to nickel-ethylenediaminediacetic acid (NiEDDA) was observed, and the periplasmic loop 2 (ECL1), where increased water accessibility accompanies ATP binding and hydrolysis.
Double electron-electron resonance (DEER) data in liposomes and detergent micelles were obtained in the apo state and following ATP hydrolysis and Vi trapping to form a high-energy ADP/Vi intermediate. Figure 1A shows a representative set of data, and Figure 1B shows distance distributions calculated as previously described [41]. The length of data records displayed in Figure 1A was selected to optimize the trade-off between the signal-to-noise ratio and attainable range and resolution of distances [34,42]. We observed distinct dipolar oscillations in the DEER signal for all sites (except 61) in the post-hydrolysis intermediate. They were less distinct in the apo state, and aperiodic decays for the loop site 61 are typical of a wide distribution of distances, pointing to a range of protein conformations. All signals provide accurate distances, as illustrated in Figure 1 and in Figures S1–S5. The difference in appearance of the signal indicates that there is a distinct protein conformation in the post-hydrolysis intermediate in contrast to a wider distribution in the apo intermediate.
Average distances, reported in Table 1, reveal substantial reconfiguration in all three domains of MsbA. ATP hydrolysis fuels a closing motion in the intracellular domain of approximately 10–20 Å. The location of spin labels in three helices suggests that the distance changes reflect a concerted conformational rearrangement in the transmembrane segment. Site 103 in the intracellular part of helix 2 is particularly instructive, because the attached spin labels do not undergo changes in the motional state or in collision frequency with NiEDDA [29]. This indicates that there is no significant change in the local environment of the spin label, and the distance change at this site thus reflects a rigid body type movement of the backbone. An even larger distance change is reported at the NBD interface where spin labels at sites 539 move closer by almost 30 Å (we note that the spin label at site 539 is confined such that the uncertainty in distance is 0.3 Å). In contrast, an opening movement of 10 Å occurs between spin labels at site 61 in ECL1. Taken together, these distance changes suggest that alternating access of the chamber is induced by substantial movements on both sides of the transporter. We emphasize that the results for detergent micelles and liposomes given in Table 1 are sufficiently close to justify the use of (detergent) micelles in the study of MsbA. This was not at all obvious at the outset of the present study and highlights the advantage of pulse dipolar spectroscopy in allowing the measurements of distances in both environments.
A similar pattern of distance changes in the three regions of MsbA is reported by fluorescein probes undergoing homotransfer [37,39] (Table 2). The probes were introduced in the same general locations as the spin labels, although the exact residues were adjusted to minimize perturbation by their larger molar volume, which affected reactivity and compromised stoichiometric labeling at sites such as 301. Unlike DEER, homotransfer is measured at ambient temperatures where the transporter samples all the conformers that are accessible. However, the widths of distance distributions in the apo state as well as for spin labels in the ECL1 loop support a significant range of conformations trapped in frozen samples. Comparison of Tables 1 and 2 shows a general agreement in the sign of the distance changes at both sides of MsbA. The absolute distances between fluorescein labels in liquid solution and spin labels in frozen media are in reasonable agreement to the extent imposed by the difference in the reporter groups, and the average nature of the distances calculated from steady-state fluorescence anisotropy. Zou et al. carried out a systematic comparison of distances calculated by the two methods and suggested that the primary factors accounting for the differences are the extension of the linking arm and the tendency of either probe to undergo specific interactions with neighboring main and side chains [39].
We compared the distances measured in the ADP/Vi intermediate to the crystal structure of Sav1866. Figure 2 maps the sites of spin labeling onto the crystal structure of Sav1866 [13]. Distances were calculated by modeling the spin label side chain into the protein structure (e.g., Figure S6). At solvent-exposed sites, the spin label is not expected to have a preferred orientation relative to the backbone and can be represented as a cone projected along the Cα-Cβ bond with a 7-Å distance between the Cα and the nitroxide oxygen [43], although exclusions due to the tertiary contacts leading to “unusual” rotamers are possible [44,45]. In particular, at site 99, which is equivalent to 103 in MsbA, the spin labels point in opposite directions, and the predicted distance between the nitroxide oxygens is 50 Å, in close agreement with the experimental distance of 47 Å. A 47–53 Å distance range is obtained by modeling the g+ g+ g+ rotamer of the spin label visualized in a number of crystal structures [44].
At buried sites, repacking due to steric constraints between the main-chain and side-chain atoms biases the dihedral angles along the linking arm and results in a net orientation of the spin label that cannot be easily modeled. However, even in these cases, the deviations of the measured distance from the alpha carbon separation can be rationalized by the expected projection of the spin labels (Figure 2B). To illustrate this effect, we sampled the range of distances between spin labels at site 244 (248 in MsbA) by changing the torsion angles around the Cα-Cβ and Cβ-Sγ bonds. Whereas the sampling grid was selected to represent the extreme distances, it was not exhaustive and we did not attempt to assess the relative energies of spin labeled MsbA for every set of torsion angles. We found the modeled distance between the spin labels to vary from 14 to 24 Å (Table 1), which is shorter than the distance between the corresponding alpha carbons as expected from projecting the spin label along the Cα-Cβ vector (Figure 2B), but well in line with the experimental data in Figure 1B. At site 297 (301 in MsbA), the range of distances between the spin labels is 25 to 34 Å which is larger than the alpha carbon separation reflecting a relative outward projection of the spin labels (Figure 2B). Finally, we can model the spin labels at site 536 (539 in MsbA) in a conformation with no steric overlap to yield a separation of 25 Å that agrees well with the measured distance (Table 1).
Similar to multidrug ABC transporters, a spectrum of potential MsbA substrates has been identified based on stimulation of ATP hydrolysis [46,47]. Among them is the substrate lipid A and a number of its processed derivatives including Ra LPS. The latter was used in the crystallization of MsbA [48] and was visualized bound to the external surface of the molecule. Ra LPS is relatively more soluble than lipid A, hence its use is more practical for spectroscopic analysis.
LPS at a concentration used for stimulation of ATP hydrolysis [46] increases the distance between fluorescein labels at all sites, although the effect on the NBDs appears marginal (Table 2). The distance increase has the shape of an apparent binding isotherm in the lower range of LPS concentrations (Figure 3A). The effect is detergent-dependent: LPS titration in undecyl-maltoside, which has a higher critical micelle concentration (CMC) relative to α-ddm (dodecyl maltoside), shifts the curve to lower concentrations, and exposes an inflection point above which the distance increases monotonically (Figure 3A Figure S7). Given that LPS forms micelles at submicromolar concentrations, the inflection point may reflect a major structural transition as LPS progressively assumes the role of a solvent.
If LPS is acting as a substrate, then it is expected that ATP binding and/or hydrolysis will reduce its affinity [21,25]. Indeed, the LPS-induced distance increases are inhibited by prior formation of the ADP/Vi intermediate and the concomitant closure of the chamber (Figure 3A), which is consistent with a model where the LPS molecule or its head group interacts with residues at the cytoplasmic end of the chamber.
To further characterize the modes of LPS interaction with MsbA, we probed the structural changes induced by excess LPS concentration (1–10 mM) relative to the initial detergent. LPS induces the appearance of a mobile component in the ESR spectrum at all sites explored and leads to a new population of transporters with longer distances (Figure 3B). At the periplasmic site 57, the addition of LPS reduces the amplitude of dipolar splittings (arrow in Figure 3B), which reflects spin labels separated by less than 10 Å, implying an increase in distance.
An effective distance increase between residues 301 (Figure 3C) and 61 (Figure S8) for an LPS concentration of 1.5 mM (which is already in excess, as compared to a protein concentration of ∼100 μM) is also detected by DEER. The increase in distance at site 61 is more pronounced than for site 301, with virtually no change for site 248 (unpublished data). This may indicate preferential binding of LPS at the ECL1 loop region that is always accessible due to the spherical shape of micelles. In addition, the same sign of the distance change at sites 301 and 61 is not in line with the opposite signs of comparably large changes produced in opening or closing of MsbA by ATP hydrolysis, but is more indicative of a different structural change. Taken together, the data suggest that at these concentrations, LPS induces an increase in monomer separation in the transmembrane domain.
At an LPS concentration of 5 mM, there is no further increase in distances, but the pattern of DEER signals clearly changes and can be interpreted in two ways. The first, and most likely, explanation is that it reflects the presence of a second component with a much longer distance (>70 Å); this could imply the complete dissociation of the dimer. The second, less likely, explanation is that at excessive concentrations of LPS, lipids bridge two transporters at their periplasmic side (as was observed in the crystal structure of MsbA-ADP/Vi in presence of LPS in high concentration), leading to a more rapid decay of the DEER signal. But this case technically is more difficult to reconcile with the nearly uniform pattern of signal change for sites 61, 248, and 301 located at progressively larger distance from spin-labels residing on the suggested second dimer. A more detailed DEER study would be instrumental in establishing the precise nature of the mode of LPS interaction with MsbA
Hydrolysis of ATP subsequent to LPS addition resets the distance constraints close to those of the ADP/Vi as long as the LPS concentration is in the range of the binding isotherm (Figure 3A). The ensemble averaging of the homotransfer by steady-state anisotropy detection implies that the partial distance recovery may reflect a population of transporters that did not turn over ATP; i.e., have a separation similar to that obtained by LPS addition. This interpretation is reinforced by the multi-component nature of the ESR lineshape and its partial recovery at higher LPS concentrations. Notable is the decrease in the population of labels with dipolar coupling at sites 248 and 307 (unpublished data) if LPS is added before ATP and Vi are added (Figure 2C). In contrast, the ESR spectra of the preformed ADP/Vi intermediate are unchanged after the addition of up to 10 mM LPS, as illustrated in Figure 3B for site 248 (compare black and blue traces). Thus, high concentrations of LPS relative to the detergent result in substantial structural reorganization, which may reflect a solvent effect rather than the specific interaction of a substrate.
The nucleotide-bound structure of Sav1866 [13] has demonstrated that the post-hydrolysis conformation of ABC efflux transporters has an open chamber to the extracellular or periplasmic side [13]. What is less clear is the orientation of the chamber in the absence of nucleotides and whether the two NBDs undergo cycles of dimerization/dissociation [4,6]. Crystallographic analysis of isolated NBDs led to a model wherein ATP binding is required for NBD dimerization whereas its hydrolysis favors dissociation. The ATP-switch model proposes that the NBDs cycle between open and closed dimer conformations upon ATP binding although without dissociation or major reorientation relative to the transmembrane domain [27]. Crystallographic analysis of intact ABC importers has been particularly supportive of a limited separation between the NBD dimer during transport [14,15]. The structures suggest that the alternating access can be accommodated with little change in overall transporter architecture [14,15]. In this model, the two NBDs, which are independent subunits, are in contact throughout the transport cycle with relative movement confined to the P-loop and the signature motif. In the ADP-bound structure of Sav1866, the packing of the NBDs was interpreted as challenging dimerization/dissociation models [13].
Our distance constraints obtained in detergent and liposomes provide a scale for the movements in MsbA that mediate the cycling of chamber accessibility reported by spin labels in the intracellular regions, ECL2, and along helices 2, 5, and 6 [29]. In conjunction with previously reported distance changes at site 57 [29], the data from sites 60 and 61 strongly indicate that the extracellular side of the transporter undergoes opening motion of large amplitude following ATP hydrolysis. This distance is well beyond the uncertainties imposed by possible label conformations. Spin labels at site 57 show distinct dipolar splitting in the continuous wave ESR [29] spectrum, implying closer proximity of ECL1 in the apo conformation relative to the nucleotide-bound structure of Sav1866. Similarly, the closing of the chamber in the cytoplasmic side that leads to reduction in NiEDDA accessibility upon ATP binding occurs through large movements, although in the opposite direction compared to the periplasmic side. While an outward/inward cycling of chamber accessibility can be accommodated in the constant contact model [15], the maximum predicted separation between the MsbA monomers is limited. The uniformly large magnitude of the distance changes reported here by two independent probes cannot be easily interpreted by this model. The variations in the amplitude of the distance changes between different sites in the cytoplasmic domain suggest that the relative movement is not a simple relative translation between the two MsbA monomers as noted by Dawson and Locher and schematically illustrated in Figure 4 [13].
More importantly, the distance change between the NBDs cannot be reconciled with constant contact models. The packing of the NBDs in the apo BtuCD structure represents their maximal separation during the cycle [14]. The structure predicts a 13-Å distance at the α carbons of 539, which is not consistent with our measured distance in apo MsbA. Similarly, comparison of the isolated NBDs of MalK in different nucleotide states reveals a limited association/dissociation cycle and predicts a distance change close to 4Å at the α carbon at the equivalent residue to 539 [6]. The scale of the experimentally measured distance change implies a significantly larger separation of the dissociated NBDs in MsbA as depicted in Figure 4.
It is possible that dissociation of the NBD dimer is a property of efflux ABC transporters reflecting a more substantial reconfiguration of the substrate chamber and the need to accommodate bulky substrates such as lipid A. In the Sav1866 structure, two helices in the ICD (IL) of one monomer contact the NBD of the opposite monomer, an interaction not observed in the BtuCD transporter. If these contacts were to be disrupted due to the opening in the ICD region as implied by our data (sites 103 and 248), this may destabilize the NBD dimer enough to allow its complete dissociation.
Our results provide direct structural insight into the interaction of LPS with MsbA. It is clear that the addition of LPS leads to structural rearrangements even at low concentrations. However, as with all lipophilic substrates, the results can be easily confounded by changes in the properties of the micelles and liposomes. In the case of MsbA, high concentrations of LPS induce pronounced structural rearrangements possibly reflecting a solvent-like effect. At all sites explored here, high concentrations of LPS increased mobility of spin labels. We are carrying out a systematic analysis of LPS effects on MsbA structure in detergents and liposomes (P Zou HS Mchaourab, unpublished observations).
When analyzed in the context of previous crystallographic and biochemical studies, our results are consistent with the initiation of the transport cycle by substrate binding to an open chamber at the cytoplasmic side of the transporter. After ATP binding, large-amplitude motion is required to form the ABC dimer, consistent with the two domains having significant conformational entropy in the apo intermediate and not being in contact throughout the cycle. In addition to the absolute distances reported in Table 1, this configuration is also supported by large NiEDDA accessibilities of residues on the cytoplasmic side in the apo intermediate [29]. Furthermore, the reanalyzed structure of apo MsbA shows a large open chamber and the two NBDs are separated by about 50 Å (G Chang, personal communication). Because the cellular concentration of ATP exceeds the Km of MsbA, the open conformation of the apo intermediate is expected to be transiently populated. Current models of ATP hydrolysis propose that ATP turnover resets the conformation of the substrate binding site to high affinity [29]. Indeed, the accessibility profiles of spin labels in the ADP-bound and apo intermediates (J Dong and HS Mchaourab, unpublished results) are similar implying that the open state may be populated after ATP turnover and release of inorganic phosphate but before rebinding of ATP. In conjunction with previous accessibility data [29], our results support a model where a reversal of the chamber polarity gradient through an alternating access mechanism makes the two orientations of the substrate headgroup relative to the transporter energetically equivalent and initiates substrate translocation.
MsbA mutants were expressed and purified as previously described [29]. Briefly, Escherichia coli BL21(DE3) harboring the mutant plasmids were grown in minimal media and the protein expression was induced at 30°C. MsbA was extracted using α-ddm and purified by a two step nickel-affinity and size-exclusion chromatographies. The mutants were labeled with either the MTSSL (1-oxyl-2,2,5,5-tetramethylpyrrolinyl-3-methyl)-methanethiosulfonate spin label or MTS-fluorescein (2-[(5-fluoreceinyl) aminocarbonyl] ethyl methanethiosulfunate) (Toronto Research Chemicals; http://www.trc-canada.com/) [39]. Both fluorescein and spin-labeled mutants have retention times similar to the WT on the superdex 200 size-exclusion chromatography column. The spin-labeled mutants reported here were previously shown to turn over ATP at rates 20%–100% of the WT [29]. Reconstitution into liposomes was carried out as previously described [29] except that the lipid to protein molar ratio was adjusted to 2000/1.
For homotransfer, two samples were prepared for each mutant. Stoichiometrically labeled samples were prepared by addition of 10-fold molar excess of fluorescein twice over a period of 4–6 h. The reaction was allowed to proceed overnight at 4 °C. Underlabeled samples were prepared by adding 0.2 moles of fluorescein per mole of MsbA followed by addition of a 5-fold molar excess of a diamagnetic analog of the MTSSL (Toronto Research Chemical) to block the unreacted cysteines [49]. Labeling efficiencies were determined by comparing the absorbance at 280 nm to that at 492 nm. As shown previously, this ratio can be used to confirm labeling efficiency [39]. All stoichiometrically labeled samples had a 0.5 absorbance ratio.
Liposomes samples for spectroscopic analysis were in a 50 mM Hepes, 50 mM NaCl, pH 7.5 buffer. Ra LPS (Sigma-Aldrich; http://www.sigmaaldrich.com) was dissolved into the same buffer containing the appropriate amount of detergent. Typically, MsbA mutants were incubated with LPS at 37 °C for 15 min before or after formation of the ADP/Vi intermediate. The ADP/Vi intermediate was trapped by addition of 1 mM Vi following addition of ATP solutions containing 5 mM MgCl2 and incubated for 20 min at 37 °C.
Distance measurements were carried out either on a home-built spectrometer at the National Biomedical Center for Advanced Electron Spin Resonance Technology (ACERT) facility at Cornell University, which operates at 17.3 GHz, or on a Bruker 580 pulsed ESR spectrometer, which operates at 9.36 GHz, using DEER with a standard four-pulse protocol [35] in both cases. For detergent samples, glycerol was added to yield 30% w/w prior to cooling. All experiments were carried out at 50–80 K. DEER signals were analyzed by the Tikhonov regularization and maximum entropy methods (MEM) [41,50] to determine average distances and distributions in distance, P(r), as illustrated in Figures S1–S5. The error in the distance was conservatively estimated by taking half of the P(r) width at 0.7 of the height.
Samples were analyzed on a steady state T-format fluorometer (Photon Technology International; http://www.pti-nj.com/). For each mutant, we collected steady-state anisotropy for a stoichiometrically labeled sample as well as an underlabeled sample. The latter serves as a reference wherein the steady-state anisotropy reflects the intrinsic reorientation of the probe. The extent of labeling was parametrized using the ratio of absorbance at 280 and 492 nm and compared to that expected based on labeling of T4L where the absolute extinction coefficient is available.
The fluorescence anisotropy, r, was measured by comparing the polarization of the emitted light to the polarization of the excitation light according to the equation:
where Ivv and Ivh refer to the amplitude of fluorescence emission parallel and perpendicular to the plane of excitation light, respectively. The G-factor was determined for each sample to correct for bias in each channel.
Distances were calculated using an expression derived by Runnels and Scarlata [38,39]. We have calibrated this method using T4 lysozyme as a model protein system and demonstrated the correspondence of distances determined between spin labels and those determined by homotransfer [39].
|
10.1371/journal.pntd.0003058 | Treatment of Tungiasis with Dimeticone: A Proof-of-Principle Study in Rural Kenya | Tungiasis (sand flea disease) is a neglected tropical disease, prevalent in resource-poor communities in South America and sub-Saharan Africa. It is caused by an inflammatory response against penetrated female sand fleas (Tunga penetrans) embedded in the skin of the host. Although associated with debilitating acute and chronic morbidity, there is no proven effective drug treatment. By consequence patients attempt to remove embedded sand fleas with non-sterile sharp instruments, such as safety pins, a procedure that represents a health threat by itself. In this proof-of-principle study we compared the topical application of a mixture of two dimeticones of low viscosity (NYDA) to the topical application of a 0.05% solution of KMnO4 in 47 school children in an endemic area in rural Kenya. The efficacy of the treatment was assessed during a follow up period of seven days using viability signs of the embedded parasites, alterations in the natural development of lesion morphology and the degree of local inflammation as outcome measures. Seven days after treatment, in the dimeticone group 78% (95% CI 67–86%) of the parasites had lost all signs of viability as compared to 39% (95% CI 28–52%) in the KMnO4 group (p<0.001). In the dimeticone group 90% (95% CI 80–95%) of the penetrated sand fleas showed an abnormal development already after 5 days, compared to 53% (95% CI 40–66%; p<0.001) in the KMnO4 group. Seven days after treatment, signs of local skin inflammation had significantly decreased in the dimeticone group (p<0.001). This study identified the topical application of dimeticones of low viscosity (NYDA) as an effective means to kill embedded sand fleas. In view of the efficacy and safety of the topical treatment with dimeticone, the mechanical extraction of embedded sand fleas using hazardous instruments is no longer warranted.
| Tungiasis (sand flea disease), a parasitic skin disease, causes important morbidity, and eventually leads to mutilation of the feet. Hitherto, the only effective treatment is the surgical extraction of embedded sand fleas. In the endemic areas this is done using inappropriate sharp instruments and causes more harm than good. We identified the three last abdominal segments of Tunga penetrans which protrude through the skin and through which the parasite breathes, defecates, and expels eggs - as an Achilles heel of embedded sand fleas. In a proof-of-principle study we investigated whether this Achilles heel is vulnerable to dimeticone with a low viscosity and a high creeping property. We randomized the left and the right feet to either receive a topical application of KMnO4 (the standard treatment in Kenya) or of dimeticone. The major outcome measure was the absence of viability signs of the treated sand fleas. The study shows that the topical application of a mixture of two dimeticones (NYDA) effectively kills embedded sand fleas within seven days. Since dimeticones are considered to be wholly non-toxic and are not expensive the new treatment could become a means to control tungiasis-associated morbidity on the population level.
| Tungiasis (sand flea disease) is a neglected tropical disease frequent in South America, The Caribbean and in sub-Saharan Africa. [1], [2], [3]. It is prevalent in resource-poor rural and urban communities, where animal reservoirs are present and people live in poverty [2], [4], [5], [6], [7], [8]. In the last decade, tungiasis has re-emerged in East Africa in epidemic dimensions [9]. In 2010, Ahadi Kenya Trust, a non-governmental organization, reported several hundred thousand cases of tungiasis in Kenya alone, of which the majority were children [4], [10], [11].
Sand flea disease is the result of an intense inflammatory response against penetrated sand fleas embedded in the skin of the host. The mechanisms underlying the inflammation are complex and only partially understood [11], [12], [13]. Immediately after a successful penetration the female sand flea starts to hypertrophy reaching the size of a pea after 10 days [14]. Through its abdominal rear cone the parasite remains in contact with the environment [14]. The tiny opening in the skin (250 to 500 µm) is needed for copulation with male sand fleas, breathing, defecation and expelling eggs [14]. After expulsion of all eggs the female sand flea dies in situ and is discarded from the epidermis by tissue repair mechanisms [14].
Although by its nature a self-limiting infection, tungiasis is actually a debilitating disease in endemic areas [15]. Sequels are common and are related to repeated and severe infection. They include acute and chronic inflammation of toes, deformation and loss of toe nails, fissures and lymphoedema [11].
Bacterial super-infection is almost invariably present [13]. It increases the inflammation and leads to intense pain [16]. If embedded sand fleas are removed by using inappropriate sharp instruments, severe mutilation of the feet may develop including deep ulcers, gangrene and loss of toes [15]. Septicaemia has also been described [17] and tetanus is a known deadly sequel in non-vaccinated individuals [18].
Hitherto, the only effective treatment is the surgical extraction of embedded sand fleas under sterile conditions in medical facilities. However, in the endemic areas patients do not have access to appropriately equipped health centers and therefore use any kind of sharp instruments (safety pins, sewing needles, hair pins, sharpened pieces of wood, etc.) to remove embedded sand fleas. Attempts to remove the embedded parasites by using a sharp instrument, invariably causes a (micro) hemorrhage [9]. As the same instrument is frequently used to remove embedded sand fleas from different persons, this procedure increases the risk of the transmission of blood-borne pathogens, such as hepatitis B and C virus [19].
In an act of desperation, patients may apply toxic substances to the skin with the intention of killing the embedded parasites. In Brazil and Madagascar, for instance, kerosene, used petrol, and insecticides are used [9], [20]. In rural Uganda, a crop pesticide used in tomato cultivation is applied (H. Feldmeier, unpublished observation 2013).
In the absence of safe and effective treatment options, Ahadi Kenya Trust recommends to bath the feet in a 0.05% solution of potassium permanganate (KMnO4) for 10 minutes [10]. However, the efficacy of this approach is not known. In Brazil several antihelminthic compounds, including ivermectin, have been tested, but none proved to be a really effective [21].
Dimeticones are silicone oils of low viscosity with a low surface tension and excellent creeping properties. They are highly effective against head lice [22]. The substance creeps into the tracheae of head lice and leads to lethal asphyxia within one minute [23]. The mode of action is purely physical. Dimeticones are biochemically inert and are not absorbed when applied to the skin or swallowed [24]. They are neither carcinogenic nor teratogenic and are considered wholly non-toxic [24].
Previous observation in rats infested with T. penetrans showed that if a drop of a solution of two dimeticones of low viscosity (NYDA) was applied on top of the protruding rear cone of an embedded sand flea, the parasite rapidly lost signs of viability (H. Feldmeier, unpublished observation 2011). Based on this observation we decided to investigate the efficacy of the dimeticone for the treatment of tungiasis in a proof-of-principle study in rural Kenya. The results show that wetting the skin of the feet with dimeticones with low viscosity effectively kills embedded sand fleas and reduces tungiasis-associated inflammation within seven days.
The study was performed in Gatundu North District, central Kenya, approximately one hour north of Nairobi. Tungiasis is endemic in this region. People live in small hamlets in houses made of wood or bricks. Families earn their living from subsistence farming. Most households possess animals, dogs, chicken and pigs. The animals live on the compound or are brought back to it in the night. Living conditions are generally very poor.
The study participants were school children aged five to sixteen years enrolled at the public Kiamwangi Primary School and Ikuma Primary School, which are situated five km to each other. The classrooms consist of simple houses without a solid floor. Both schools have a limited access to water, so that the schoolyards and rooms cannot be cleaned regularly. Most pupils wore worn-out sandals or walked barefoot. The study was carried out between January 10 and February 17, 2012. This period coincides with the high transmission period of T. penetrans.
To allow comparison between the new approach (the application of the dimeticone) and the local reference procedure (bathing feet in a 0.05% solution of KMnO4), one foot was bathed in the KMnO4 solution for 10 minutes and to the other foot the dimeticone was applied three times during this period (see below). Since bathing a foot in a 0.05% KMnO4 solution changes the color of the skin into dark purple, neither the patient nor the examiner were blinded with regard to the treatment applied.
Individuals, aged ≥5 years, with at least one lesion in stage IIa – IIIa of the Fortaleza classification on each foot were eligible [14]. In IIa the sand flea is already completely embedded in the skin of the host and has started to hypertrophy [14]. Lesions in stage IIIa correspond to a fully developed parasite with a characteristic watchglass-like appearance. In this stage the female sand flea starts to expel eggs [14]. In stage IIIb egg expulsion stops, thereafter the sand flea dies and the lesion changes into stage IV: the lesion becomes crusted, viability signs become rare and eventually completely disappear [14]. Hence, sand fleas in stage IIa – IIIa are most suitable to assess viability and alterations in the normal development of the parasites.
The inclusion criterion for an eligible lesion was the presence of at least 2 out of 4 viability signs at the baseline examination: expulsion of eggs, excretion of a faecal thread, excretion of faecal liquid or pulsations/contractions of the parasite. Viability signs were determined using a handheld digital video microscope (eScope iTEZ, Hongkong, China) (see supplementary electronic material 1).
When several eligible lesions were present on one foot only those (at most three) were selected for evaluation that allowed a clear discernment of the developmental stage of the embedded parasite and a quantification of the inflammatory response around the lesion. Hence, lesions occurring in cluster and lesions which the patient had attempted to manipulate were excluded. Other exclusion criteria were: Presence of gross inflammation, abscess or ascending lymphangitis or lymphedema on either foot. Children with such complications of tungiasis were referred to the nearest health facility for treatment.
For practical reasons we decided to treat always the same foot with dimeticone and KMnO4, respectively. At the beginning of the study a coin was tossed for randomizing the two treatments. This resulted in application of the dimeticone to the left foot and of KMnO4 to the right foot. Children were informed not to manipulate the lesions during the next seven days.
Before each examination the feet of the participants were washed properly with water and soap and dried with a clean towel. Then, the left foot was wetted with NYDA up to the ankle three times within 10 minutes. In the interval, the foot was kept in an upright position to allow surplus dimeticone to evaporate. Simultaneously, the right foot was put into a bucket containing a 0.05% KMnO4 solution, and remained there for 10 minutes. After sun drying the right foot, vaseline was applied to compensate the desiccation of the skin caused by KMnO4. The immersion of the foot in 0.05% KMnO4 for 10 minutes and the subsequent oiling with vaseline is the standard procedure applied by Ahadi Kenya Trust. After treatment the children were allowed to continue their daily activities.
The lesions were monitored daily for viability signs and the abnormal development of the embedded parasite for a total of seven days. One week reflects the period of normal development of a sand flea from stage IIa to stage IIIa [14]. Thereafter, it looses its characteristic watchglass-like appearance, but does not increase in size anymore [14]. Hence, abnormalities in development are difficult to be detected.
In order to detect a change of tungiasis-associated inflammation an inflammation score was developed. In addition to the classic signs of local inflammation (erythema, oedema and warmness) the score included the presence of suppuration, ulcers and fissures as well as itching and pain. The inflammation score ranged from 0 to 27 points [25].
In total, 48 participants were recruited and 47 were randomized. The flow diagram is shown in Figure 1.
Two major outcome measures were defined. First, the proportion of viable embedded sand fleas which lost viability signs after seven days of follow-up. An embedded sand flea was considered to be dead when none of the four viability signs (expulsion of egg, excretion of faecal thread, excretion of faecal liquid, pulsations/contractions) was detected during 15 minutes of observation with the digital handhold video microscope on two consecutive follow-up examinations. Videos were recorded and reviewed in the evening of the examination day (see supplementary electronic material 1). Second, the proportion of embedded sand fleas in which the normal development was interrupted. We defined a development as abnormal, when the lesion did not change its size within two consecutive follow ups and/or morphological abnormalities developed, e.g. discoloring or desiccation of the abdominal rear cone [14].
A secondary outcome measure was the intensity of local inflammation, as assessed semi-quantitatively by the inflammation score. The observation units for all outcome measures were single sand flea lesions.
The sample size calculation was based on the following assumptions: with a level of confidence set at 95% together with a power of 90% assuming equal number of lesions in treatment and control group, 45 lesions in each group were needed to determine a difference of 35% in the major outcome measure between the two treatments assuming a 40% effect of the standard treatment.
Fisher's exact test was used to compare proportions. General estimation equations were used to analyze the evolution of the inflammation score during the observation period.
The study was approved by the Ethics Committee of the Ministry of Health, Nairobi (MMS/ADM/3/8/Vol 111), and was registered at Controlled-trials.com (ISRCTN: 91405042). The study was performed in accordance with the ethical standards of the Ethics Committee of the Ministry of Health, and with the Declaration of Helsinki as amended 2013 by the World Medical Association. Informed written consent was obtained from the guardians of the participants in English before starting the study. For ethical reasons no controls were included. During the study, food was provided free of charge to the participants. At the end of the study, any remaining viable sand fleas were removed under sterile conditions and the wounds were dressed following standard procedures. All patients received a new pair of closed solid shoes.
The baseline characteristics of the feet of the 47 participants are summarized in Table 1. None of the variables differed significantly between the two feet. In the NYDA group, 88 lesions were included in the study, in the KMnO4 group 82.
Table 2 shows the efficacy of treatment based on the disappearance of viability signs. Already three days after application of dimeticone 50% of the parasites lost all viability signs (efficacy = 50%), whereas the efficacy in the KMnO4 group was 14% (p<0.001). At day 7 the efficacy was 78% (95% CI 67–86%) after treatment with dimeticone and 39% (95% CI 28–52%) after treatment with KMnO4 (p<0.001); a difference of 39% (95% CI 23–54%). In the dimeticone group, lesions in an early stage of development lost viability signs more often than lesions in later stages (efficacy = 88% (95% CI 75–95%) versus 65% (95% CI 47–79%) at day 7 (p = 0.01)). In the KMnO4 group, there was no difference between lesions in early and later stages of development.
The effect of treatment on the morphological development of the lesions is shown in Table 3. Already after 5 days in the dimeticone group 90% (95% CI 80–95%) of sand flea lesions showed an abnormal development as compared to 53% (95% CI 40–66%) (p<0.001) in the KMnO4 group.
Figure 2A–C and 3A–C show the macroscopic development of lesions after the treatment with dimeticone or KMnO4, respectively. Figure 4A–D and 5A–D depict the microscopic development of lesions after treatment as seen through the digital handhold video microscope.
In the dimeticone group the inflammation score decreased from a median of 6.0 at baseline to a median of 4.75 at day 7. In contrast, in the KMnO4 group, the inflammation score increased (median 4.5 versus 5.0). Both differences were significant (p<0.0001 and p = 0.009, respectively).
During the study period three sand fleas were extracted by the participants or their caregiver in the NYDA group and 11 in the KMnO4 group.
Tungiasis, a wide spread neglected tropical disease, is prevalent in resource-poor rural and urban communities, where animal reservoirs are present and people live in poverty [2], [4], [5], [6], [7], [8]. Elimination of sand flea disease is not possible as long as the precarious living conditions, which are characteristic of the endemic areas, prevail and animal reservoirs exist.
Taking into consideration the high prevalence of tungiasis, the absence of appropriate infrastructure in the endemic areas and the health hazards associated with the traditional treatment, there is an urgent need for a safe and effective drug treatment. Recently, dimeticones have emerged as highly effective chemicals against ectoparasites such as head lice [26]. Since dimeticones have a purely physical mode of action and are considered to be non-toxic, they have become the standard treatment of pediculosis capitis in Europe [22].
We considered the last abdominal segments of an embedded sand flea, which protrude through the skin by forming a miniature cone and through which the parasite breathes, defecates and excretes eggs, as an Achilles heel, which can be targeted by dimeticone. Since the opening leading to internal organs measures less than 1 mm, we decided to use a combination of two dimeticones of very low viscosity with a low surface tension and excellent creeping properties (NYDA) [23].
We defined a set of viability signs of embedded sand fleas detectable through a handhold digital video microscope. We used the presence of viability signs as the major outcome measure and compared the efficacy of a 0.05% solution of KMnO4 – the standard treatment used in mass campaigns in Kenya – to wetting the foot with dimeticone three times during a period of 10 minutes. The observation period was limited to seven days, since a certain number of embedded sand fleas will die even without any intervention during this period [14].
After 7 days, 78% of the lesions did not show any sign of viability in the dimeticone group, whereas the proportion was 39% in the KMnO4 group. True efficacy of a 0.05% solution of KMnO4 alone may be lower since KMnO4 is a disinfectant and has no insecticidal properties. It is unlikely that KMnO4 diluted in water will creep into vital organs of embedded sand fleas through the parasite's abdominal cone. Presumably, the observed effect in the KMnO4 treated lesions was due to the vaseline which was applied to the skin for cosmetic reasons (because bathing the feet in KMnO4 makes the skin rough and cracked). Applied on the skin, vaseline rapidly turns into oil, particularly in hot climate countries. Liquid fatty acids of the vaseline may thereby creep into the abdominal rear cone and suffocate the parasite.
Interestingly, the efficacy of dimeticone to kill embedded sand fleas depended on the stage of development: parasites being in an early stage of development were more susceptible than those who had already fully developed (efficacy = 88% versus 66%). This is plausible, since embedded sand fleas increase their size by a factor of approximately 2000 within 6–7 days during the development from stage IIa to stage IIIa [14]. Such a rapid growth requires an intense metabolism, which in turn needs constant supply of oxygen. During the early stages of development supply of oxygen might be at a critical limit. This makes the parasite vulnerable for suffocating compounds such as low-viscosity.
Since it is important to kill sand fleas as soon as they have penetrated in order to prevent the development of clinical pathology [16], the enhanced effect of dimeticone on early developmental stages is an additional advantage. The early death of the embedded parasite will also prevent the expulsion of eggs – which starts about one week after penetration – and, thereby, may have an impact on transmission.
92% of the embedded fleas treated with dimeticone showed an abnormal development. This could indicate that no (or fewer) eggs are produced and released into the environment. Hence, if applied on the population level, treatment with dimeticones could have even an impact on the off-host cycle of the parasite, possibly resulting in lower attack rates over time.
In the dimeticone group, the inflammation score started to decrease after 3 days and became significantly lower after 7 days, whereas in the KMnO4 group the inflammation slightly increased. It is conceivable that the resolution of inflammation reflects the rapid death of the parasites. Previous studies have shown that tungiasis-associated inflammation comes to a halt and tissue repair mechanism begins, when the parasites are dead [25], [27].
Another indicator of the efficacy of the dimeticone was that in the course of the study 11 sand fleas were extracted from the feet treated with KMnO4 by the patients themselves, whereas in the NYDA treated feet only 3 sand fleas were removed. Similarly, when the study participants were asked at the end of the study about their satisfaction, only 10 participants preferred KMnO4, but 37 preferred the dimeticone. Children also disliked that KMnO4 colored the skin into deep purple for a few days which led to teasing in school (Figure 6).
This study on the treatment of a neglected parasitic disease is particularly in the sense that an Achilles heel of the parasite was identified first and then a compound was identified that is able to target the vulnerable body part. The abdominal cone which protrudes through the skin and through which the parasite breathes, defecates, excretes liquids and expels eggs was considered to be an ideal target for a dimeticone with a low viscosity and excellent creeping properties.
Although this was a proof-of-principle study with a small number of units of observations, it can be concluded that the topical application of a mixture of two dimeticones (NYDA) comprises a promising approach to treat sand flea disease. The treatment can be performed by the patient himself with minimal input from the health sector. Hence, surgical extraction with all its associated complications is no longer warrantable. After the sand flea has died in situ, the inflammation resolved. Importantly, future resistance of the parasites against dimeticone treatment is highly unlikely to evolve, since the drug acts only physically.
|
10.1371/journal.ppat.1002974 | Transcription Factor Amr1 Induces Melanin Biosynthesis and Suppresses Virulence in Alternaria brassicicola | Alternaria brassicicola is a successful saprophyte and necrotrophic plant pathogen. Several A. brassicicola genes have been characterized as affecting pathogenesis of Brassica species. To study regulatory mechanisms of pathogenesis, we mined 421 genes in silico encoding putative transcription factors in a machine-annotated, draft genome sequence of A. brassicicola. In this study, targeted gene disruption mutants for 117 of the transcription factor genes were produced and screened. Three of these genes were associated with pathogenesis. Disruption mutants of one gene (AbPacC) were nonpathogenic and another gene (AbVf8) caused lesions less than half the diameter of wild-type lesions. Unexpectedly, mutants of the third gene, Amr1, caused lesions with a two-fold larger diameter than the wild type and complementation mutants. Amr1 is a homolog of Cmr1, a transcription factor that regulates melanin biosynthesis in several fungi. We created gene deletion mutants of Δamr1 and characterized their phenotypes. The Δamr1 mutants used pectin as a carbon source more efficiently than the wild type, were melanin-deficient, and more sensitive to UV light and glucanase digestion. The AMR1 protein was localized in the nuclei of hyphae and in highly melanized conidia during the late stage of plant pathogenesis. RNA-seq analysis revealed that three genes in the melanin biosynthesis pathway, along with the deleted Amr1 gene, were expressed at low levels in the mutants. In contrast, many hydrolytic enzyme-coding genes were expressed at higher levels in the mutants than in the wild type during pathogenesis. The results of this study suggested that a gene important for survival in nature negatively affected virulence, probably by a less efficient use of plant cell-wall materials. We speculate that the functions of the Amr1 gene are important to the success of A. brassicicola as a competitive saprophyte and plant parasite.
| Induction of cell wall-degrading enzymes, secretion of toxins, enforcement of fungal cell wall architecture, and detoxification of host defense molecules are important or essential for the fungal infection of plants. Genes important for each of these functions have been identified in various fungal species. Understanding how these genes are coordinately regulated and how many regulators are involved in the process is a challenge. We recently discovered a transcription factor gene, AbVf19, which positively regulates hydrolytic enzyme-coding genes. Here, we report on another transcription factor gene, Amr1, which negatively regulates a subset of these genes during late-stage pathogenesis and positively regulates melanin biosynthesis during conidiogenesis. This study adds another dimension to the complex regulation and overall importance of hydrolytic enzyme genes during plant pathogenesis by necrotrophic fungi. In addition, this study provides an example on the evolutionary implication of virulence in the necrotrophic fungus, A. brassicicola. One transcription factor essential for long-term survival because of its role in melanin biosynthesis is used instead to suppress virulence. We speculate that the suppressive functions of Amr1 contribute to the specialized adaptation of A. brassicicola as an efficient and successful facultative parasite.
| Alternaria brassicicola is the causal agent of black spot disease of cultivated brassicas (e.g. cabbage, canola, mustard). Pathogenesis in necrotrophic fungi is generally described as a two-step process: killing host cells directly (necrosis) or inducing programmed cell death with toxins, then decomposing host tissues with cell wall-degrading enzymes. The importance of toxins in disease development by other necrotrophs has been clearly demonstrated [1], [2], [3], [4]. Several A. alternata pathotypes produce toxic, host-specific secondary metabolites that are essential for pathogenicity [5], [6], [7], [8]. Our knowledge of toxins produced by A. brassicicola is currently limited. In recent studies, Brassicicolin A emerged as the most selective phytotoxic metabolite produced in liquid cultures of A. brassicicola [9]. The genes responsible for biosynthesis of this toxin, however, have yet to be determined. Diterpenoid toxins called brassicenes have also been described from this fungus and linked to gene clusters in the A. brassicicola genome [10]. In addition, a cluster of five genes responsible for synthesis of the secondary metabolite depudecin, a histone deacetylation inhibitor was recently reported [11]. Mutation of the five genes abolished depudecin synthesis and caused only a small (10%) reduction in virulence compared to wild-type A. brassicicola. A weak protein toxin was also reported [12], [13] but the gene or genes responsible for its production have yet to be identified.
In addition to toxins, several A. brassicicola genes have been linked to pathogenesis [14], [15], [16], [17], [18], [19], [20]. These genes are involved in iron uptake, cell wall integrity, peroxisome-mediated redox homeostasis, hyphal fusion, hydrolytic enzymes, and signal transduction. All mutants of the pathogenesis-related genes showed either a reduction in virulence or loss of pathogenicity. For example, mutants of a mitogen-activated protein (MAP) kinase gene, Amk1, and its downstream transcription factor-coding gene, AbSte12, were nonpathogenic. The mutants of either gene also showed slow vegetative growth and impaired conidium maturation.
Bioinformatics analysis of the genome sequence of A. brassicicola identified many candidate genes important for pathogenesis. They included several genes that are essential for melanin synthesis. Alternaria species produce 1,8-dihydroxynaphthalene (1,8-DHN) melanin, like various other filamentous fungi [21], [22], [23]. It is heavily concentrated in the primary cell walls of conidia and in their septa [24]. Melanin is a ubiquitous pigment that plays an important role in protecting fungi from the damaging effects of environmental stress and so may be considered an advantageous adaptation. It increases the tolerance of fungi to UV irradiation [25], [26], [27], enzymatic lysis [28], and extreme temperatures [26], [29]. Melanin is also required for the mechanical penetration of host plants by other phytopathogenic fungi, such as Magnaporthe grisea and Colletotrichum lagenarium [30].
Synthesis of 1,8-DHN melanin requires polyketide synthase (Pks1 in Bipolaris oryzae and Pks18 in Cochliobolus heterostrophus), T4HN reductase (Brn2), scytalone dehydratase (Scd1), and T3HN reductase (Brn1 or Thr1) genes identified in B. oryzae [31], Col. lagenarium [32], and Coc.heterostrophus [21], [33]. Expression of these genes is mainly regulated by the Cmr1 transcription factor gene in Col. lagenarium. Cmr1 homologs are regulators of the primary melanin biosynthesis pathway in several plant pathogenic fungi but virulence is not affected by the gene mutation in M. grisea, Col. lagenarium, or Coc. heterostrophus [21], [22].
For all pathogenesis-associated genes in A. brassicicola studied to date, their mutants have been either nonpathogenic, or less virulent than the wild type. In this study, we determined that mutants of the Cmr1 homolog in A. brassicicola (Amr1) were melanin-deficient but more virulent than wild-type A. brassicicola. We defined virulence as quantitative host plant damage as measured by lesion diameter. The increase in virulence of the Δamr1 mutants suggested that the loss of the gene was beneficial to pathogenesis. We tested three research questions: whether melanin is important for pathogenesis, if mutants efficiently neutralize host defense chemicals, or if the mutants more efficiently utilize plant cell wall-associated materials such as pectin. Our study provides an example of a transcription factor gene that is important for survival in nature negatively regulates virulence in a necrotrophic plant pathogen. We speculate that the negative regulation of virulence by the Amr1 gene contributed to A. brassicicola's efficiency as a facultative plant pathogen while retaining characteristics of a robust saprophyte.
The genome sequence of A. brassicicola has been determined by Washington University School of Medicine, Genome Sequencing Center, and is publicly available from DDBJ/EMBL/GenBank under accession ACIW00000000. In addition, the annotated genome is available in the context of other sequenced Dothideomycetes genomes through the interactive JGI fungal portal MycoCosm [34] at http://jgi.doe.gov/Abrassicicola. The A. brassicicola genome size is approximately 31.9 Mb, smaller than the average genome size of other dothideomycete genomes [35]. There are 10, 688 predicted genes in the assembled genome. We predicted 421 genes encoding putative transcription factors in the draft genome of A. brassicicola using Pfam searches [36]. These transcription factors excluded the core proteins necessary for the formation of transcription initiation complexes and RNA polymerases. We classified the transcription factors in 13 categories based on their functional domains (Table 1). Members of the largest group (221 of 421) contain at least one zinc-finger domain. The second largest group (72 of 421) contains fungal-specific transcription factor domains.
We previously created targeted deletion mutants for 22 transcription factor and signaling related-genes and reported the association of AbSte12, AbPro1, and AbVf19 transcription factors with pathogenesis [15], [20]. In this present study, we generated targeted gene disruption mutants for an additional 117 transcription factor genes (Figure S1). All mutants corresponding to 109 of the 117 genes caused disease symptoms on host plants with lesion sizes similar to those caused by the wild type in multiple pathogenicity assays. Mutants of eight of 117 genes showed unusual changes in virulence. Among the eight mutants, five mutants also grew very slowly on nutrient rich medium. These mutants either failed to cause disease symptoms or their lesions did not expand beyond the initial infection site (Table S1). Gene disruption mutants of the AbPacC transcription factor (abpacc) were nonpathogenic (Table 2), unlike the severely reduced virulence in the loss-of-function mutants of its homologs in Col. acutatum [37] and Sclerotinia sclerotiorum [38] or no changes in Ustilago maydis [39]. Mutants corresponding to two other genes (Amr1and AbVf8) grew normally on nutrient rich media, were novel factors associated with pathogenesis, and are described further in this study.
In contrast to the loss-of-function mutants of other virulence genes studied thus far in this pathosystem, Amr1 disruption mutants (amr1) caused larger lesions than wild-type A. brassicicola when green cabbage (Brassica oleracea) leaves were inoculated with the same number of conidia (data not shown, see below for detailed study). The predicted Amr1 gene encodes a protein of 997 amino acids with two C2H2-zinc finger motifs and one fungal-specific Zn(II) binuclear motif. Of the 997 predicted amino acids, 869 (87%) were identical (E-value = 0) to the transcription factor Cmr1 in the taxonomically closely related dothideomycete fungus, Coc. heterostrophus [21]. We named this Cmr1 homolog in A. brassicicola, Amr1 (Alternaria melanin regulation, GenBank accession number: JF487829). Only one copy of the gene was present in the draft genome sequence of A. brassicicola (http://jgi.doe.gov/Abrassicicola).
In order to confirm our results with gene disruption mutants and to perform additional studies, deletion mutants of Amr1 (Δamr1) were produced by replacing the coding region with either a Hygromycin B (HygB) resistance cassette, or a green fluorescent protein (GFP) coding sequence [40] plus a HygB resistance cassette. Southern hybridization confirmed that the Amr1 gene was absent from nine melanin-deficient transformants but still present in the eight transformants producing melanin (Figure 1). We replaced the Amr1 coding region with a single copy of the HygB resistance cassette in Δamr1-4 and three Δamr1:Amr1p-GFP mutants (d4, D1, D2, and D3 in Figure 1). The replacement construct containing the GFP and HygB resistance cassette was designed so the GFP gene would be regulated by native promoter elements of the Amr1 gene in the replacement mutants (Δamr1:Amr1p-GFP) (Figure 1B). We complemented Δamr1-5 mutant with either the wild-type allele or a chimeric construct with the TrpC promoter to constitutively express the wild-type Amr1 gene.
Because the increase in virulence in A. brassicicola was unusual and unprecedented, we performed additional pathogenicity assays with four strains of gene deletion mutants (Δamr1-1, Δamr1-4, Δamr1-5, and Δamr1:Amr1p-GFP). All lesions caused by the mutants were 30–80% larger in diameter than the wild type in assays with 5- to-7-week-old plants (Table 3). Generally, wild-type A. brassicicola caused larger lesions on 5-week old plants than on 7-week-old plants and on older leaves compared to younger leaves. Interestingly, all Δamr1-type mutants produced consistent lesion sizes regardless of plant age. Thus, the relative size of lesions caused by the mutants compared to the wild type was greater on 7-week-old plants than on 5-week-old plants and on younger leaves than on older leaves (Figure 2A and Table 3). The difference in lesion size was similar between assays on whole plants and assays on detached leaves, so for convenience we performed a subsequent assay on the detached leaves of 8-week-old plants. In this assay, we compared lesion sizes caused by a gene deletion mutant (Δamr1), one strain from each of the two complemented mutants (Δamr1:Amr1 and Δamr1:TrpCp-Amr1), and the wild type. Lesions caused by the deletion mutant (Δamr1) were 2-fold greater on average than those made by the wild type or both complemented mutants (Δamr1:Amr1 and Δamr1:TrpCp-Amr1) (Figure 2E and Table 4).
We performed additional pathogenicity assays using as few as ∼300 spores to test the effect of low inoculum concentration on pathogenesis (Figure 2 B, C). Lesions caused by ∼440 wild-type conidia were small and some stopped expanding, while lesion diameters caused by ∼400 Δamr1:Amr1p-GFP mutant conidia were on average ∼3-fold larger (p<0.0001, df = 29, two tailed t-test) and continued to expand throughout the infection process. Inoculation with approximately ∼330 wild-type conidia produced black spots, indicating successful penetration but a failure to colonize host tissue. Lesion diameters created by the same concentration of mutant conidia, however, were on average ∼5-fold larger (p<0.0001, df = 29, two tailed t-test) and continued to expand during the five-day experiments. This result demonstrated that the Δamr1 mutants required fewer conidia than the wild type for successful infection.
In order to support the notion that AMR1 is a transcription factor, we monitored the localization of AMR1 protein using a mutant strain expressing an AMR1-GFP fusion protein. The AMR1-GFP fusion protein was primarily located in the nuclei of highly melanized conidia and aerial hyphae (Figure 3A). AMR1 protein expression in conidia was consistent with its proposed regulatory role in melanin biosynthesis during conidiogenesis in B. oryzae, Coc. heterostrophus, Col. lagenarium, and M. grisea [21], [22], [41]. We could not detect the fluorescent fusion protein in hyphae growing on PDA (data not shown) or at 24 hours post-inoculation (hpi) in planta during the early stages of pathogenesis (Figure 3B). In contrast, the fusion proteins were detected in the nuclei of hyphae actively invading host tissues during the late stage of infection (Figure 3C). As a negative control, expression and localization of GFP were also monitored in a promoter-tagged mutant (Δamr1:Amr1p-GFP) in which GFP expression was regulated by the native Amr1 promoter (Figure 3 D–F). GFP in the promoter-tagged mutant was distributed throughout the cytoplasm of all fungal tissues, including conidia. The GFP continued to be expressed at moderate levels in the hyphae of the Δamr1:Amr1p-GFP mutant, even during the early stages of plant infection when AMR1-GFP protein was undetectable (Figure 3 B, E).
All strains of verified gene disruption and deletion mutants (amr1, Δamr1, and Δamr1:Amr1p-GFP) produced melanin-deficient conidia. The color of the mutant colonies, however, was orange by 3 days post-inoculation (dpi) when grown on a PDA medium (inset in Figure 4 A, B). The orange color became denser with increased conidia production, or when the culture was exposed to light. The pigment was not accumulated in fungal tissues but was secreted into the medium. Droplets of pink pigment were visible at the tips of many conidial chains on PDA and on plant surfaces by 7 dpi (Figure 4E). The pigment was also secreted during the culture in liquid medium (Figure S3A). The pigment did not have phytotoxic effects on host plants (Figure S3B). Furthermore, addition of pink pigment to inocula of wild-type conidia did not affect lesion size (Figure S3C). The wild-type allele of the Amr1 gene, controlled by its own promoter or by the TrpC promoter, restored the ability of the Δamr1 mutants to accumulate melanin (Figure 4 C, D) and pink pigment was no longer visible. Conidia produced by the mutant complemented with TrpCp-Amr1 constructs were slightly lighter than the taupe-colored wild-type conidia. None of the complemented mutants secreted pink pigment to the medium.
Because melanin plays an important role in protecting fungi against various types of stress, we examined the effect of UV irradiation, glucanase, heat, salt, and osmolites on Δamr1 mutants. UV irradiation at 80 mJ had a marginal effect on conidial germination and hyphal growth of wild-type A. brassicicola (Figure 4F, upper panel). However, 40 mJ of UV irradiation severely affected conidial germination and hyphal growth of the mutants, and 60 mJ almost eliminated germination of the Δamr1mutant (Figure 4F, lower panel). It took more than three hours for β-glucanase to digest the melanized cell walls of wild-type A. brassicicola spores, but less than one hour for it to digest the cell walls of the melanin-deficient Δamr1 mutant (data not shown).
Germination and vegetative growth were not significantly different between the wild type and Δamr1 mutants under high-temperature conditions or in the presence of NaCl, H2O2, or sorbitol. The colony size was comparable for both the Δamr1 and Δamr1:Amr1p-GFP mutants, a complemented mutant, and the wild type on the same media (Figure 4G).
We produced gene disruption mutants of the melanin-associated polyketide synthase (Pks) homolog in A. brassicicola (AbPks7). The Pks uses acetyl coenzyme A or malonyl coenzyme A as a precursor and produces a melanin-intermediate product, 1,3,6,8-tetrahydroxynaphthalene (T4HN). The abpks7 disruption mutants produced melanin-deficient conidia (Figure S4 A–C). These melanin-deficient mutants caused lesions comparable to the wild type on their host plant, B. oleracea (Figure 4D), further demonstrating that melanin is not a virulence-associated factor in this fungus.
Carbohydrate-active enzymes, such as glycoside hydrolases, polysaccharide lyases, and carbohydrate esterases are slightly expanded in A. brassicicola as has been found in other plant pathogenic Dothideomycetes [35]. The enzymes encoded by these genes are thought to be involved in the breakdown of complex carbohydrates in the cell walls of host plants. We indirectly evaluated the role of Amr1 in regulating genes encoding putative cell wall-degrading enzymes by comparing the vegetative growth (dry weight of mycelium) of the Δamr1-1 and Δamr1-4 mutants to the wild type on various carbon sources. A minimal broth medium was supplemented with either 1% glucose, or the common cell wall polysaccharides α-cellulose, xylan, or pectin. Vegetative growth of the Δamr1-1 and Δamr1-4 mutants was the same as the wild type in the presence of glucose, xylan, lignin, and α-cellulose (data not shown). However, growth of these mutants was greater than the wild type in the presence of citrus pectin (Figure 5 A, C). To verify this observation, we tested the effects of pectin on the growth of the Δamr1-3, Δamr1-5, and Δamr1:Amr1p-GFP mutants. These mutants also showed a significant increase in vegetative growth over the wild type in the presence of pectin (Figure 5 B, D). Vegetative growth was up to 86% greater than the wild type when mutants were cultured longer than 46 hours (Figure 5 C, D). There was no significant difference in growth between the wild type and the mutants in the presence of glucose. Because all Δamr1 mutants showed similar levels of melanin deficiency, pink pigment secretion, and pectin utilization, we used Δamr1-4 in subsequent gene expression profiling studies.
We used RNA-seq approach to determine the subset of genes regulated by Amr1. Using the RNA-seq approach, gene expression profiles were obtained and then compared between a Δamr1 mutant and the wild type during the late stage of infection on green cabbage. At this time, the AMR1 protein was localized in the nuclei of infecting hyphae. A total of 6.17×107 reads were produced for the wild type and 6.94×107 for the Δamr1 mutant. Of these, 4.14×107 (67.1%) tags for the wild type and 4.78×107 (68.9%) tags for the mutant were mapped to the genome of A. brassicicola. Of A. brassicicola's 10,688 predicted genes, 101 and 163 genes were significantly (p<0.05) up- or down-regulated more than 2-fold, respectively, in the Δamr1 mutant compared to the wild type (Table S2). This respectively represents 0.95% and 1.53% of the genes in the A. brassicicola genome. Functional categories that were overrepresented in the up-regulated genes included putative cell-wall depolymerization enzymes (Table S3). They included 21 glycoside hydrolases, two pectate lyases, and one pectin esterase (Table S2). In addition, two cutinases and one lipase enzyme were also up-regulated in the mutant. One of the two cutinases, CutAb, was up-regulated >32-fold. The necrosis-inducing protein gene (3.1-fold), two P450 genes (2.8- and 2.6-fold), and the homolog of the peptidase aegerolysin (3.3-fold) were also up regulated in the mutant. 23 of 101 up-regulated genes in Δamr1 mutants were expressed at a lower level in the previously characterized Δabvf19 mutants and eleven of the shared genes between the two mutants were glycoside hydrolases (Table 2 and Figure 6). Genes in the melanin synthesis pathway were among the down-regulated genes. As expected, the expression level of Amr1 was close to zero (background noise level) in the Δamr1mutant during the late stage of infection, compared to a high level of induction in the wild type (Table S2). Expression levels of other structural genes in the melanin synthesis pathway—Brn2, Scd1, and Brn1 (Eliahu et al., 2007; Tsuji et al., 2000)—were 8- to 30-fold lower in the mutant than in the wild type. Differential expression between the wild type and Δamr1 mutant was verified by quantitative real time PCR (qRT-PCR) for seven differentially and three similarly expressed genes (Table S2). AbPks7 was annotated as three separate genes and their expression in the mutants was reduced about four-fold (Table S2).
We examined the expression of the nine genes at two additional time points when Amr1 was expressed at a low level. As expected, the Amr1 gene and its three downstream genes in the melanin synthesis pathway (Brn2, Scd1, and Brn1) were not expressed in the Δamr1:Amr1p-GFP mutant under any of the tested conditions (Figure 7B). They were expressed at a low level in the wild type at 40 and 64 hpi, when Amr1 expression was low. When the Amr1 transcripts were slightly increased at 88 hpi, however, the three genes were highly induced in the wild type (Figure 7A). The expression level of Amr1 was extremely low in the wild type compared to the expression of GFP in the Δamr1:Amr1p-GFP mutant. This may explain why the gene expression level of GFP is high in the Δamr1:Amr1p-GFP mutant, but AMR1-GFP fusion protein was not detected during early infection (Figure 3 B, E). Moreover, this could have been due to the absence of feedback inhibition from Amr1. It may also have been caused by the increased stability of GFP mRNA, its protein, or both as compared to Amr1. The high stability of GFP protein has previously been reported (reviewed in [40]).
The lipase and cellobiohydrolase Cbh7 (AB06252.1) genes were weakly expressed in the wild type from early pathogenesis until 64 hpi; during this time the Amr1 expression level was low in the wild type. There was no expression of these two genes during saprophytic growth in an axenic medium by either the wild type or the mutant (Figure 7 A, B). At 88 hpi, the time when Amr1 was induced in the wild type, Cbh7 and lipase induction were 4.3- and 4.7-times greater respectively (p<0.001, t-test, df = 2) in the Δamr1:Amr1p-GFP mutant than in the wild type (Figure 7D). The glycoside hydrolases and chymotrypsin genes were not expressed in either the wild type or the mutant during saprophytic growth in glucose yeast extract broth (GYEB) (Figure 7 GY24h). Expression of the chymotrypsin gene was twice as high in the mutant at 64 hpi as in the wild type, but there was little difference in chymotrypsin levels at 88 hpi (Figure 7C). In summary, transcript levels of putative hydrolytic enzyme genes, including lipase and Cbh7, were moderately higher with statistical significance in the mutant than in the wild type during the colonization of host plants (Figure 7). This coincided with the time when the Amr1 gene was induced in the wild type.
Another virulence factor identified in these studies was the putative transcription factor AbVf8. The AbVf8 gene showed no sequence similarities to known pathogenesis-associated genes in any fungus. AbVf8 in A. brassicicola and its homologs in other fungi (e.g., in Pyrenophora spp. or Phaeosphaeria nodorum) were annotated as either a hypothetical protein or a predicted transcription factor. The diameters of lesions caused by disruption mutants of AbVf8 (abvf8) were 50–80% smaller than those produced by the wild type. Vegetative growth rates and the formation of conidia were comparable to the wild type. Predicted domains in the protein sequence of the AbVf8 genes were somewhat informative regarding their functions. The predicted protein encoded by AbVf8 contained a putative SET domain. SET domains are thought to be an epigenetic regulator of gene expression during development and modulate chromatin structure.
Because it was a novel transcription factor associated with pathogenesis and had a possible association with epigenetic regulation, we decided to confirm the results of pathogenicity assays using additional mutants. We made gene deletion mutants (Δabvf8) by replacing the coding region with a HygB resistance cassette. Southern hybridization confirmed that the AbVf8 gene was replaced by a single copy of the HygB resistance cassette in nine transformants (Figure S2). One transformant had an ectopic insertion of multiple copies of the HygB resistance cassette. Pathogenicity assays using Δabvf8-2, Δabvf8-3, and an ectopic insertion mutant produced results consistent to those we produced with the gene disruption mutants. Both deletion mutants caused lesions with diameters 55% smaller than lesions caused by the wild type or the ectopic insertion mutant (Table 2). We complemented the Δabvf8-2 mutant with the wild-type allele of the AbVf8 gene. Complemented mutants restored its ability to produce lesions comparable to the wild type (data not shown). We will further characterize the functions of the AbVf8 gene in the future.
Of 138 A. brassicicola transcription factor genes with zinc finger domains and other regulatory genes analyzed to date in this and previous studies, only six genes were strongly associated with pathogenesis. These six genes were AbSte12, AbPacC, AbVf19, Amr1, AbVf8, and AbPro1 (Table 1). This result was comparable to the comprehensive knockout study of 693 putative transcription factors in Fusarium graminearum, where 62 genes affected virulence [42]. AbSte12 is a homolog of the Ste12 gene with mutants that are either nonpathogenic or reduced in virulence in several fungi (reviewed in [43]). Ste12 is regulated by a MAP kinase, Fus3/Kss1. Mutants of either homologous gene in A. brassicicola showed defective conidial development and were nonpathogenic [14], [15]. The AbPacC gene has a single copy in the genome sequence of A. brassicicola and shows a high sequence similarity (E = 0.0) with the PacC gene in other fungi. PacC is the transcription factor in the pH regulation pathway of Aspergillus nidulans [44]. It also controls the expression of fumonisin toxins in F. verticillioides [45] and endopolygalacturonase and oxalic acid in S. sclerotiorum [38], [46]. Mutation of its homologous genes reduced virulence in S. sclerotiorum and Col. acutatum and increased virulence in F. oxysporum [37], [38], [47]. AbVf19 regulates a suite of genes that are important for decomposing and utilizing plant materials during the late stage of plant infection [20]. We have yet to characterize the functions of AbVf8 and AbPro1. Regardless of their functions, however, all mutants studied to dates except Amr1 were nonpathogenic or showed a reduction in virulence.
Melanin is a ubiquitous pigment that plays an important role in protecting fungi from the damaging effects of environmental stress. It accumulates in the cell walls of hyphae and conidia during the late stationary phase of mycelial growth. It also forms under stressful conditions, including ultraviolet irradiation, a hyperosmotic environment, nutrient deprivation, or an accumulation of toxic wastes during in vitro culture. Over-expression of a Cmr1 ortholog in Bipolaris oryzae (Bmr1) caused continuous induction of the three downstream genes in the melanin synthesis pathway and increased melanization of the colonies [41]. The loss of function of Cmr1, its homologs, or their downstream genes in other fungi, resulted in melanin deficiency [21], [22], [25], [48]. We identified a transcription factor in this study, Amr1, which regulates the melanin synthesis pathway in A. brassicicola. The melanin synthesis-associated structural genes Brn1 and Pks are located together in a 30 Kb region and their organization and orientation is the same as in the closely related fungus Coc. heterostrophus [21]. Interestingly, as reported here melanin was not associated with virulence in A. brassicicola as has been reported in other fungi [21]. However, based upon our data there appears to be a linkage between melanin biosynthesis and a reduction in virulence, perhaps due to a lifestyle switch from pathogenesis to reproduction.
Amr1 gene knockout mutants, both gene disruption (amr1) and gene deletion (Δamr1 and Δamr1:Amr1p-GFP), were easily recognized by the lack of melanin in their conidia, orange-colored colonies, and the secretion of a pink pigment (Figure 4). These phenotypes were similar to those found in mutants of its homologs in other fungi [21], [22], [41]. The Δamr1mutants were more susceptible to UV light and enzyme digestion (Figure 4). Unexpectedly, the Amr1 gene knockout mutants were more virulent than the wild type, producing larger lesions (Table 3), and required fewer conidia for successful infection (Figure 2). Mutation of Amr1 homologs in the phytopathogenic fungi M. grisea and Col. lagenarium produced melanin-deficient conidia. The appressoria of these mutants, however, remained melanized suggesting an alternate regulatory mechanism and pathogenicity was not affected [22]. Melanin deficiency caused by a mutation of other genes in the melanin biosynthesis pathway was associated with a loss of pathogenicity in M. grisea and Col. lagenarium [49], [50], [51] and a decrease in virulence in the opportunistic human fungal pathogens Aspergillus fumigatus and Waggiella dermatitidis [52], [53]. In most other plant-pathogenic fungi, melanin deficiency usually had little effect on pathogenesis under controlled laboratory conditions [25], [31], [48], [54]. Melanin deficient mutants of Coc. heterostrophus, however, were nonpathogenic under field conditions possibly due to increased sensitivity to UV light outdoors [54]. In summary, the increased virulence of the Δamr1 mutants was unexpected and unique among melanin-deficient mutants of pathogenic fungi.
The pink pigment secreted by Δamr1 mutants was neither toxic to host plants nor beneficial for the wild type A. brasscicola to infect host plants (Figure S3). Furthermore, melanin-deficient mutants of another gene, abpks7, showed no changes in virulence (Figure S4). These data suggested that neither the presence of pink pigment nor the lack of melanin is associated with the increased virulence of the Δamr1 mutants investigated in this study. Additionally, there was no difference in germination or vegetative growth between the Δamr1 mutants and the wild type in the presence of various chemicals (Figure 4G). However, the mutants were more susceptible than wild type to UV light (Figure 4F) and the cell wall-degrading enzyme, glucanase. These data suggested that the mutants' ability to respond to stress was not associated with their increased virulence, since plant glucanases are known to be defense enzymes.
The ability of the mutants to outgrow the wild type when citrus pectin was the major carbon source (Figure 5) suggested that the mutants were more efficient in utilizing the pectin component of plant cell walls and middle lamellae. The expression profile comparisons identified 30 hydrolytic enzyme genes expressed moderately more in the mutant than in wild type A. brassicicola during pathogenesis (Table S2). However, they included only two pectate lyase genes and one pectin esterase gene. Other genes upregulated in the mutant compared to the wild type included genes coding for two cutinases, a lipolytic enzyme and 20 other various glycoside hydrolases (GH). Interestingly, 6 of the 22 genes in the GH61-gene family were expressed at higher levels in the Δamr1 mutant. The GH61 family might have a role in degrading cellulose, lignocellulose, chitin, or other polysaccharides [55]. The GH61 family may be involved in digesting pectins whose side chains are structurally diverse in constituting sugars and glycosidic linkages. They may digest other cell-wall components, such as cellulose and hemicelluloses along with other glycoside hydrolases. It was not possible in this study to evaluate whether the Δamr1 mutants made better use of these cell wall materials because A. brassicicola grew poorly in the presence of α-cellulose, beechwood-xylan, and lignin. Our uninformative growth data was likely due to the non-brassica source of cell wall materials. Nonetheless, the gene expression and growth assay data indirectly support the importance of cell wall-degrading enzymes and possibly those involved in pectin digestion, since pectic polysaccharides constitute about one-third of the cell wall components and most of the middle lamella in dicotyledonous plants.
Induction of the Amr1 gene in A. brassicicola did not occur until 64 hpi (Figure 7). Suppression of the Amr1 gene might be important during pathogenesis when energy is allocated to colonization of host tissues and overcoming host defense mechanisms, rather than producing reproductive dispersal structures like conidia. In the wild type, the transcripts of hydrolytic enzyme genes were slightly increased compared to the actin transcripts, as previously reported [14], [56], [57]. In addition, when conidiation was initiated the induction of Amr1 was accompanied by a sharp increase in transcripts of its downstream genes in the melanin synthesis pathway. There was also a small increase in the transcripts of hydrolytic enzyme-coding genes in the wild type. However, many of these genes, mostly expressing glycoside hydrolases, a few cutinases, and pectate digestion enzymes, were expressed at higher levels in the Δamr1-4 and Δamr1:Amr1p-GFP mutants, which did not express Amr1 (Figure 7D). This suggests that Amr1 suppressed genes predicted to be involved in the digestion of lipids and carbohydrates. This occurred during late pathogenesis (88 hpi) when they were still needed to some extent for either killing [58] or digesting host tissue. Whether the Amr1 gene is involved in the regulation of most hydrolytic enzymes sequentially induced in A. brassicicola during pathogenesis has yet to be investigated [24], [59]. Other genes encoding a necrosis-inducing protein, two P450s, the aegerolysin peptidase homolog, and other hypothetical proteins might have contributed to the increased virulence of the Δamr1 mutant and warrant further investigation. Interestingly, 24 genes expressed at a higher level in Δamr1 were expressed at lower levels in the Δabvf19 deletion mutants that displayed a reduction in virulence (Figure 6C), suggesting opposite roles of the two transcription factor genes in the regulation of pathogenesis-associated genes. This finding may aid in the selection of specific targets for future analyses of their functions in pathogenesis.
The melanin-deficient mutant phenotype and expression pattern of the Amr1, Brn1, Brn2, and Scd1 genes in A. brassicicola suggested that the function of the Amr1 gene was to regulate melanin synthesis during conidiogenesis. This function is highly conserved in other fungi [21], [22], [41]. In addition to regulating the melanin biosynthesis pathway, Amr1 negatively regulated virulence. _ENREF_1The evolution of virulence has been a subject of extensive theoretical analysis. The “trade-off hypothesis” [60] produced a number of models showing a contested relationship between virulence and transmissibility. It challenges the hypothesis that strong virulence can seriously damage or kill host plants and threaten pathogen survival, so parasites will ultimately evolve to be less virulent. However, analyses of ecological and epidemiological factors related to virulence has produced inconsistent results [61]. Therefore, a modified trade-off theory predicted that strong parasites became moderately virulent to their hosts over time, but did not completely lose virulence [62]_ENREF_4. Our current work identified a transcription factor suppressing virulence but enhancing the ability of a fungus to preserve itself in space and time via melanin synthesis, primarily in conidia. This study provides an example of a transcription factor gene important for survival in nature that plays a critical role in negatively regulating virulence in the necrotrophic plant pathogen, A. brassicicola. The suppression of virulence we observed is unique and might have been acquired by this fungus, as this phenotype has not been observed in loss-of-function mutants of Amr1 homologs in other species of plant pathogenic fungi. An ancestral fungal strain of A. brassicicola without the ability to suppress virulence by the activity of Amr1 could have been a more virulent pathogen. Negative regulation of virulence is rare but exists in other phytopathogenic fungi. For example, a phosphatase enzyme in the MAP kinase signaling pathway negatively regulates virulence in Ustilago maydis [63]. Currently, A. brassicicola is a prolific saprophyte and necrotroph, efficiently colonizing susceptible, weakened, or dead host plants. Its pathogenicity, however, is inhibited on more resistant host plants. We speculate that the suppressive functions of Amr1 contribute to the specialized adaptation of A. brassicicola as an efficient and successful facultative parasite.
We used the facultative plant pathogen Alternaria brassicicola (Schweinitz, Wiltshire) (ATCC96836) in this study. Growth and maintenance of the fungus and its transformation, nucleic acid isolation, mutant purification, and mutant verification by Southern hybridization were performed as described previously [15]. The wild-type fungus and each of the mutant strains created during this study were purified by two rounds of single-spore isolation. The cultures were maintained as glycerol stock in separate tubes, with one tube used for each assay.
The genome sequence and gene predictions for A. brassicicola ATCC 96836 were obtained from http://genome.wustl.edu/genomes/view/alternaria_brassicicola/ and are available from DDBJ/EMBL/GenBank under accession ACIW00000000. In addition, the annotated genome is available in the context of other Dothideomycetes through the interactive JGI fungal portal MycoCosm [34] at http://jgi.doe.gov/Abrassicicola.
The partially assembled genome sequence was annotated using an EnsEMBL annotation pipeline [64]. The HMM search results of FGENESH predicted peptides against PFAM and TIGRFAM databases were stored in a SQL database. A total of 251 PFAM accession numbers and 13 keywords descriptions were used to manually search for the putative transcription factors using SQL queries. Subsequently, the predicted amino acid, predicted cDNA, and genomic DNA sequences were manually retrieved from the database.
Four primers for targeted gene disruption and six primers for targeted gene deletion were synthesized for each gene (Table S4). We followed methods described previously for all steps from producing transformation constructs to verifying purified mutants [15], [65]. Briefly, constructs of up to 12 genes were independently transformed at a time. The transformed protoplasts were plated on a molten regeneration medium (1 M sucrose, 0.5% yeast extract, 0.5% casamino acids, and 1% agar) for 24 hours, followed by an overlay with the same volume of PDA containing 30 ug/ml HygB. On the seventh day after the transformation, four transformants of each gene were transferred to PDA plates containing 30 ug/ml HygB and cultured for five days. The transformants were purified by two rounds of single-spore isolation and verified by PCR using two sets of verification primers for each gene (Table S5).
All transformation constructs described in this work were produced by a double-jointed PCR method [66] with modifications described previously [15]. We made Amr1 deletion mutants by replacing the gene with a HygB resistance cassette. The replacement construct was produced with the following three sets of primers (Figure 8). The primer set P1 and P2, and the primer set P5 and P6 were used to respectively amplify the 5′ and 3′ flanking regions of the targeted locus. Another set of primers, P3 and P4, was used to amplify the HygB-selectable marker gene cassette (1,436 bp) from pCB1636. In order to create mutants expressing green fluorescence under the control of the Amr1 gene promoter, the Amr1 promoter region and 3′ flanking region were amplified with P7 and P8, and P5 and P6, respectively. Another set of primers, P9 and P4 was used to amplify the 2,423 bp that covered the coding regions of the GFP and the HygB resistance cassette. All primers used Amr1gene are listed in Table S6.
The Δamr1-5 mutant was complemented with either the wild-type Amr1 allele with its native promoter, or a chimeric construct of an Amr1 allele under the control of a constitutive promoter. This promoter was derived from a TrpC gene (TrpCp) [67]. We used two primers, P1 and P6, to make constructs and reintroduce wild-type Amr1 into the Δamr1 mutant. These primers amplified the 6,196 bp wild-type allele of the Amr1 gene using A. brassicicola genomic DNA as a template. The PCR product included a 1,342 bp 5′ flanking region, 3,282 bp complete coding region, and a 1,570 bp 3′ flanking region. Separately, P18 and P19 were used to amplify a 2,226 bp-long nourseothricin-resistant cassette as a selectable marker gene, using a pNR vector as the template [68]. These two products (8 µg of Amr1 and 7 µg of NTC cassette) were mixed to transform the Δamr1 mutant.
To produce constitutive Amr1 expression mutants, three sets of primers were used to replace the 1,840 bp 5′ upstream to the predicted start codon with the TrpC promoter, TrpCp. We used two primers, P10 and P11, to amplify the 943 bp constitutive promoter region that includes a 152 bp partial coding sequence of NAT, 458 bp nonfunctional ToxA promoter, and a 333 bp-long functional TrpC promoter. Another set of promoters, P12 and P6, were used to amplify the 4,852 bp wild-type allele that spans the 3,282 bp coding region from the start codon to the stop codon of the Amr1 gene and the 1,570 bp 3′ flanking region of the gene. The PCR products were mixed and diluted 10 times. The final mixed PCR products were used as template DNA to create TrpCp-Amr1 gene constructs using two primers, P7 and P6. A total of 8 µg of the TrpCp-AMR1 chimeric gene constructs and 7 µg of the 2,226 bp-long nourseothricin-resistant cassette were transformed into the Δamr1 mutant. Among the nourseothricin-resistant transformants, two clones of the former (Δamr1:Amr1) and four clones of the latter (Δamr1:TrpCp-Amr1) were purified by two rounds of single-spore isolation in the presence of nourseothricin. Complementation of the mutant by a single copy of the Amr1 gene was confirmed by Southern blot hybridization. All mutant clones were tested for purity on PDA by checking their growth patterns, growth rates, and uniformity of colony colors (sectoring) in the presence and absence of selectable markers.
In order to create mutants expressing green fluorescence protein fused to the C-terminal of AMR1, the Amr1 coding region (1,209 bp) and 3′ flanking region (192 bp) were amplified with P13 and P14, and P5 and P15, respectively. Another set of primers, P16 and P17, was used to amplify the 2,384 bp that covered the coding regions of the GFP and the HygB resistance cassette. The final transformation constructs were produced by PCR amplification from the mixture of the PCR products using P1′ and 6′z9219-3R.
Pathogenicity assays were performed with modifications as described previously using detached leaves harvested from 5- to 8-week-old plants or on the leaves of whole plants [15]. Commercially available seeds (Jonny's, Winslow, ME) of green cabbage (Brassica oleracea) or Arabidopsis thaliana were planted and grown under the same conditions for each pathogenicity assay. We grew green cabbage under 14 hours light 10 hours dark cycle and Arabidopsis under 10 hours light 14 hours dark cycle. Plants of similar heights and with similar-sized leaves were selected for each assay. Leaves were detached and placed in mini-moist chambers and randomly arranged on a laboratory bench for most assays. For the pathogenicity assays on whole plants, potted plants were placed in a semi-transparent plastic trough with adequate water. The troughs and plants were sealed with plastic wrap after inoculation to keep the relative humidity close to 100%. The increased virulence of each mutant was calculated using the formula (∑(Dmi-Dwi)/∑(Dwi))×100, where Dwi was the lesion diameter created by the wild-type for the ith sample and Dmi was the lesion diameter produced by the mutant for the ith sample (Table 2). We also analyzed lesion sizes among the wild type and various mutants (Tables 2). Lesion sizes were subjected to 5×4 (genotype×leaf position (leaf age)) two-way analysis of variance (ANOVA) using the general linear model (GLM) procedure in the Statistical Analysis System (SAS Institute, Cary, NC). Means were separated by the Waller-Duncan k-ratio (k = 100) t-test, or an f-test, whichever was appropriate. The Waller-Duncan test was chosen to minimize the Bayesian risk of the additive loss function.
Infected plant tissues were trimmed with a razor blade, placed on microscope slides, and covered with Gold Seal cover glasses. Confocal images were acquired using a 633 C-Apochromat (numerical aperture 1.2) water-immersion objective lens and an Olympus Fluoview 1000 Laser Scanning Confocal System on a IX-81 inverted microscope. Spectra for fungal tissues expressing standard green fluorescence and for plant cells emitting autofluorescence were collected by simultaneous 488 nm and 543 nm excitation using 30 mW argon and 1 mW helium∶neon lasers, respectively. The standard GFP spectrum was collected through 488 nm excitation using a 20 nm window from 505 to 525 nm. Plant tissue, including chloroplasts, was visualized using 543 nm excitation with a 560 nm-long pass filter. Images of fungal tissue grown in nutrient media were captured with 488 nm excitation and DIC-transmitted light. All fluorescent images were composed of multiple layers acquired with the Confocal System.
Each fungal strain from glycerol stocks was inoculated on PDA with an appropriate selectable agent and grown in the dark for 5 days at 25°C. In order to test for sensitivity to osmotic stress and oxygen radicals, wild-type and mutant conidia were pipetted onto PDA containing 2% (w/v) or 4% (w/v) NaCl, 0.6 M or 1.2 M sorbitol, or 2.5 mM or 5 mM H2O2. Mutant and wild-type strains were also cultured at 28, 30, and 33°C to measure the effect of temperature on their growth. Colony diameters were measured 4 days post-inoculation (dpi). To study the effect of UV light on germination, 1,000 spores were pipetted onto PDA, irradiated at 20, 40, 60, and 80 millijoules in a UV crosslinker (Agilent Technologies, Inc., CA), and incubated at 25°C for 24 hours. All experiments were conducted three times.
Flasks (250 ml) containing a 50 ml broth of 0.5% (NH4)2SO4, 0.05% yeast extract, 0.15% KH2PO4, 0.06% MgSO4, 0.06% CaCl2, 0.0005% FeSO4 7H2O, 0.00016% MnSO4 H2O, 0.00014% ZnSO4 7H2O, and 0.00037% CoCl2 were supplemented with either 1% glucose or 1% of α-cellulose (cat #, C8002-1KG), or xylan (cat #, X4252-100G), lignin (cat # 471003-100G), or citrus pectin (cat #, P9135-500G) purchased from Sigma (St. Louis, MO). Each flask was inoculated with 4–6×105 conidia of either Δamr1-1, Δamr1-3, Δamr1-4, Δamr1-5, Δamr1:Amr1p-GFP, or the wild type. The flasks were incubated in the dark at 25°C with continuous agitation at 100 rpm. The flasks were shaken vigorously by hand several times during the first eight hours to prevent the conidia from sticking to the flask. Mycelia were harvested at 4 dpi, washed with distilled water, and dried at 70°C overnight. The increased biomass of each mutant and the wild type was calculated using the formula (∑(Wmi-Wwi)/∑(Wwi))×100, where Wwi was the dry weight of the wild type for the ith sample and Wmi was the weight of the mutant for the ith sample.
To study the regulatory roles of Amr1 in A. brassicicola, we compared gene expression profiles between the wild type and a Δamr1 mutant at 88 hours post-inoculation (hpi), a late stage of infection. We inoculated nine detached leaves harvested from three plants with conidia from the Δamr1-4 mutant and the wild type. Tissue samples containing both host plant tissue and fungal mycelium were harvested. Three biological replicates were produced for both the mutant and the wild type. Total RNA was purified from the tissue using an RNeasy kit (Quiagen, Palo Alto, CA). We used 4 µg of total RNA for each RNA sample to construct strand-specific sequencing libraries with a TruSeq RNA Sample Prep Kit (Illumina, San Diego, CA). Each library was constructed with unique index primers and all were run in a single lane. They were later decoupled using index primer sequences.
Sequence tags were mapped to the genome sequence of A. brassicicola using the programs TopHat 1.3.1 [69] and Bowtie 0.12.7 [70]. Default settings were used, except the segment length was set at 25 nucleotides and the number of allowed segment mismatches was set at 1 nucleotide. Additionally, intron length was designated as a minimum of 10 nucleotides and a maximum of 400 nucleotides. The program Cuffdiff, version 1.0.3, which is part of Cufflinks [71], was used to identify reads overlapping with previously predicted genes. The expression levels of each predicted gene were determined and normalized by the mapped Fragments Per Kilobase of exon model per Million (FPKM). Differentially expressed genes between the wild type and the mutant were determined by comparing FPKMs from three biological replicates for both the wild type and the mutant. We also applied a cutoff of at least a two-fold change in expression value for differential expression. The bias correction method was used while running Cuffdiff [72]. Custom scripts were written in Python to analyze the data.
Custom scripts were developed in Python and R to analyze over- and under-representation of functional annotation terms in sets of differentially regulated genes using the Fisher Exact test. The Benjamini-Hochberg correction was used to correct for multiple testing using a p-value of <0.05.
Full-length sequences of the three downstream genes in the melanin synthesis pathway regulated by Amr1 were identified in the A. brassicicola genome by Blast search. Primer sequences for each gene are listed in Table S6. The Δamr1:Amr1p-GFP mutant and the wild type were used for qRT-PCR with three biological replicates of infected plant tissues. Each biological sample was collected from three or four leaves and total RNA was extracted using an RNeasy kit with DNAse digestion according to the manufacturer's protocol (Qiagen, Valencia, CA). Two micrograms of total RNA were transcribed to cDNA in a final volume of 20 µl using 50 ng of random pentamers and 200 ng of poly(T)20N with Superscript III (Invitrogen, Carlsbad, CA). Each cDNA was diluted 1∶20. Subsequent qRT-PCR reactions were performed in a 20 µl volume containing 120 nM of each primer, 1 µl of diluted cDNA, and 10 µl of FastStart SYBRGreen Master (Roche, Mannheim, Germany). Each reaction was run in a Biorad I-cycler (Bio-Rad, Hercules, CA, USA) as described previously (Cho et al, 2006). Relative amounts of the transcript of each gene were calculated as 2−ΔCt using a threshold cycle (Ct), where ΔCt = (Ct,genei –Ct,actin). Fold changes of each gene between the wild type and Δamr1 strain were calculated as 2−ΔΔCt, where ΔΔCt = (Ct,genei –Ct,actin)Δamr1 – (Ct,genei –Ct,actin)wild type.
RNA-seq data: GEO Series Accession No. GSE36781 http://www.ncbi.nlm.nih.gov/geo/info/linking.html
Amr1 (Alternaria melanin regulation) GenBank accession number: JF487829
|
10.1371/journal.pgen.1007498 | Cyclin G and the Polycomb Repressive complexes PRC1 and PR-DUB cooperate for developmental stability | In Drosophila, ubiquitous expression of a short Cyclin G isoform generates extreme developmental noise estimated by fluctuating asymmetry (FA), providing a model to tackle developmental stability. This transcriptional cyclin interacts with chromatin regulators of the Enhancer of Trithorax and Polycomb (ETP) and Polycomb families. This led us to investigate the importance of these interactions in developmental stability. Deregulation of Cyclin G highlights an organ intrinsic control of developmental noise, linked to the ETP-interacting domain, and enhanced by mutations in genes encoding members of the Polycomb Repressive complexes PRC1 and PR-DUB. Deep-sequencing of wing imaginal discs deregulating CycG reveals that high developmental noise correlates with up-regulation of genes involved in translation and down-regulation of genes involved in energy production. Most Cyclin G direct transcriptional targets are also direct targets of PRC1 and RNAPolII in the developing wing. Altogether, our results suggest that Cyclin G, PRC1 and PR-DUB cooperate for developmental stability.
| During development, the part of stochasticity inherent to biological processes induces noise. In animals with bilateral symmetry, developmental noise can be estimated by the variance in a population of the difference between the left and the right sides of individuals, the so-called fluctuating asymmetry (FA). The genetic bases of developmental stability, i.e. the control of developmental noise, are still unsolved. A large amount of data converges towards the importance of hubs–i.e. the most connected genes–in gene networks. Deregulating Cyclin G, a protein involved in transcription and in the cell cycle, induces high FA providing a unique system to address the genetic bases of developmental stability. Using this system, we show the existence of an organ intrinsic control of developmental noise linked to the interaction between Cyclin G and chromatin regulators. We also show that Cyclin G-induced FA correlates with the activation of genes involved in translation and the repression of genes involved in energy production. We suggest that fine-tuned control of these key functions is important for developmental stability
| Developmental stability has been described as the set of processes that buffer disruption of developmental trajectories for a given genotype within a particular environment [1]. In other words, developmental stability compensates the random stochastic variation of processes at play during development. Many mechanisms working from the molecular to the whole organism levels contribute to developmental stability [2]. For example, chaperones, such as heat-shock proteins, participate in developmental stability by protecting misfolded proteins from denaturation in a large variety of processes [3–5]. In Drosophila, adjustment of cell growth to cell proliferation is essential to developmental stability by allowing to achieve a consistent organ size (e.g. wing size) in spite of variation in cell size or cell number [6,7].
Developmental noise, the “sum” of the stochastic part of each developmental process, can be observed macroscopically for morphological traits. In bilaterians, quantification of departure from perfect symmetry, the so-called fluctuating asymmetry (FA), is the most commonly used parameter to estimate developmental noise [8,9]. Indeed, the two sides of bilaterally symmetrical traits are influenced by the same genotype and environmental conditions, and differences between them are thus only due to developmental noise. The use of FA as an estimator of developmental noise makes analysis of the mechanistic and genetic bases of developmental stability compatible with custom genetic and molecular approaches of developmental biology.
The genetic bases of developmental stability remain unclear. Thus, its evolutionary role is subject to many speculations (for reviews see [7,10,11]). Experiments showing the role of Hsp90 in buffering genetic variation led to the idea that developmental stability could be ensured by specific genes [12–15]. On the other hand, both theory and experiments show that complex genetic networks can become intrinsically robust to perturbations, notably through negative and positive feedbacks, suggesting that the topology of gene networks is of paramount importance for developmental stability [16]. Several authors have further suggested that hubs, i.e. the most connected genes in these networks, might be particularly important for developmental stability [17,18].
In Drosophila, mutants for dILP8 and Lgr3 involved in the control of systemic growth, have been reported to display high FA as compared to wild type flies, indicating that these genes are important for developmental stability [19–23]. Two studies have scanned the Drosophila genome for regions involved in developmental stability [24,25]. Several deletions increased FA but genes responsible for this effect inside the deletions were not identified. Nevertheless, these studies confirm that the determinism of developmental stability could be polygenic, as suggested by Quantitative Trait Loci analyses in mouse ([11] and references therein). Together, these data reinforce the idea that developmental stability depends on gene networks.
We have shown that the gene Cyclin G (CycG) of D. melanogaster, which encodes a protein involved in the cell cycle, is important for developmental stability [6,26,27]. Indeed, ubiquitous expression of a short Cyclin G version lacking the C-terminal PEST-rich domain (CycGΔP) generates a very high FA in several organs, notably in the wing. Interestingly, FA induced by CycGΔP expression is associated with loss of correlation between cell size and cell number, suggesting that the noisy process would somehow be connected to cell cycle related cell growth [6]. Hence, CycG deregulation provides a convenient sensitized system to tackle the impact of cell growth on developmental stability.
CycG encodes a transcriptional cyclin and interacts with genes of the Polycomb-group (PcG), trithorax-group genes (trxG) and Enhancer of Trithorax and Polycomb (ETP) families [28]. These genes encode evolutionary conserved proteins assembled into large multimeric complexes that bind chromatin. They ensure maintenance of gene expression patterns during development (for a recent review see [29]). PcG genes are involved in long-term gene repression, whereas trxG genes maintain gene activation and counteract PcG action. ETP genes encode co-factors of both trxG and PcG genes, and behave alternatively as repressors or activators of target genes (for a review see [30]). More recently, we discovered that CycG behaves as an Enhancer of Polycomb regarding homeotic gene regulation suggesting that it is involved in the silencing of these genes [31]. Importantly, Cyclin G physically interacts with the ETP proteins Additional Sex Comb (Asx) and Corto via its N-terminal ETP-interacting domain, and co-localizes with them on polytene chromosomes at many sites. Hence, Cyclin G and these ETPs might share many transcriptional targets and might in particular control cell growth via epigenetic regulation of genes involved in growth pathways.
Here, we investigate in depth the role of CycG in developmental stability. We first show that localized expression of CycGΔP in wing imaginal discs is necessary to induce high FA of adult wings. Furthermore, this organ-autonomous effect increases when the ETP-interacting domain of Cyclin G is removed. We show that several mutations for PcG or ETP genes, notably those encoding members of the PRC1 and PR-DUB complexes, substantially increase CycG-induced FA. Next, we report analysis of the transcriptome of wing imaginal discs expressing CycGΔP by RNA-seq and find that transcriptional deregulation of genes involved in translation and energy production correlates with high FA of adult wings. By ChIP-seq, we identify Cyclin G binding sites on the whole genome in wing imaginal discs. Strikingly, we observe a significant overlap with genes also bound by Asx, by the Polycomb Repressive complex PRC1, and by RNAPolII in the same tissue. We identify a sub-network of 222 genes centred on Cyclin G showing simultaneous up-regulation of genes involved in translation and down-regulation of genes involved in mitochondrial activity and metabolism. Taken together, our data suggest that Cyclin G and the Polycomb complexes PRC1 and PR-DUB cooperate in sustaining developmental stability. Coordinated regulation of genes involved in translation and energy production might be important for developmental stability.
We previously reported that expression of CycG deleted of the PEST-rich C-terminal domain (amino-acids 541 to 566) (CycGΔP)—a domain potentially involved in degradation of the protein [26,27]—under control of ubiquitous drivers (da-Gal4 or Actin-Gal4), generated extremely high FA, notably in wings [6]. The strength of this effect was unprecedented in any system or trait. Expression of CycGΔP thus provides a unique tool to investigate developmental stability in depth. To determine whether wing FA was due to local or systemic expression of CycGΔP, we tested a panel of Gal4 drivers specific for wing imaginal discs or neurons. A brain circuit relaying information for bilateral growth synchronization was recently identified [22]. It notably involves a pair of neurons expressing the dILP8 receptor Lgr3 that connects with the insulin-producing cells (IPCs) and the prothoracicotropic hormone (PTTH) neurons. This circuit was particularly appropriate to test the existence of a remote effect of CycGΔP expression in generating high FA in the wing. Expression of CycGΔP in this neuronal circuit (using dilp3-, NPF-, pdf-, per-, phm-, ptth and R19B09-Gal4 drivers) did not increase FA of adult wings (Fig 1 and S1 and S2 Tables). Furthermore, expression of CycGΔP in cells of the future wing hinge using the ts-Gal4 driver did not affect wing FA either. By contrast, expressing CycGΔP with 5 different wing pouch drivers (nub-, omb-, rn-, sd- and vg-Gal4) induced high wing FA. We thus concluded that CycGΔP-induced wing FA was due to an intrinsic response of the growing wing tissue.
The 566 amino-acid Cyclin G protein exhibits three structured domains: the ETP-interacting domain (amino-acids 1 to 130) that physically interacts with the ETPs Corto and Asx, a cyclin domain (amino-acids 287 to 360) that presents high similarity with the cyclin domain of vertebrate G-type cyclins, and a PEST-rich domain (amino-acids 541 to 566) [28,31]. To test whether the interaction with ETPs, and thus transcriptional regulation by Cyclin G, could be important to control FA, we generated new transgenic lines enabling to express different versions of the CycG cDNA: CycGFL (encoding the full-length protein), CycGΔE (encoding an ETP-interacting domain deleted protein), CycGΔP (encoding a PEST domain deleted protein), and CycGΔEΔP (encoding an ETP-interacting plus PEST domain deleted protein) (Fig 2A). In order to compare the amounts of FA induced, all transgenes were integrated at the same site using the PhiC31 integrase system. Globally, the different fusion proteins were expressed at the same level (S1 Fig). Contrarily to da-Gal4, the wing drivers used above induced not only high FA but also few ectopic veins or small notches that prevented to accurately measure wing centroid size. We then used da-Gal4 to ubiquitously drive expression of the transgenic lines and focus on the FA phenotype. We confirmed that expression of CycGΔP induced very high FA as compared to + and da-Gal4/+ controls. Furthermore, expression of CycGFL also significantly increased FA, although to a much lesser extent. Interestingly, expression of either CycGΔE or CycGΔEΔP significantly increased FA as compared to CycGFL or CycGΔP, respectively (Fig 2B and 2C, S3 and S4 Tables). These results show that the ETP interacting domain tends to limit CycGΔP-induced FA and suggest that the interaction between Cyclin G and chromatin regulators sustains developmental stability.
We next addressed genetic interactions between CycG and PcG or ETP alleles (Table 1) for developmental stability. FA of flies heterozygous for PcG and ETP loss of function alleles was not significantly different from that of control flies. However, when combined with da-Gal4, UAS-CycGΔP, many of these mutations significantly increased wing FA as compared to da-Gal4, UAS-CycGΔP flies (Fig 3, S5 and S6 Tables). This was the case for alleles of the PRC1 and PR-DUB encoding genes Sex comb extra (Sce1, Sce33M2 and SceKO4), calypso (caly1 and caly2), Sex comb on midleg (ScmD1), Polycomb (Pc1), and polyhomeotic (ph-p410 and ph-d401ph-p602). No modification of CycGΔP-induced FA was observed with the Psc1 allele. However, this allele has been described as a complex mutation with loss and gain of function features [32]. Opposite effects were observed for alleles of the ETPs Asx and corto. Asx22P4 increased CycGΔP FA whereas AsxXF23 decreased it. AsxXF23 behaves as a null allele but has not been molecularly characterized [33], whereas the Asx22P4 allele does not produce any protein and thus reflects the effect of ASX loss [34]. Similarly, the cortoL1 allele increased CycGΔP-induced FA whereas the corto420 allele had no effect. To characterize these alleles, we combined them with the Df(3R)6-7 deficiency that uncovers the corto locus, amplified the region by PCR and sequenced. The corto420 allele corresponds to a substitution of 14,209 nucleotides starting at position -59 upstream of the corto Transcriptional Start Site (TSS) by a 30-nucleotide sequence. Hence, this allele does not produce any truncated protein. By contrast, cortoL1 corresponds to a C towards T substitution that introduces a stop codon at position +73 downstream the TSS, generating a 24 amino-acid polypeptide. cortoL1 might then behave as a dominant-negative mutation. Lastly, no modification of CycGΔP-induced FA was observed for E(z)63 and esc21.
Interestingly, Asx and caly encode proteins of the PR-DUB complex whereas Pc, ph, Sce and Scm encode proteins of PRC1, and E(z) and esc encode proteins of PRC2. Taken together, these results indicate that Cyclin G interacts with the Polycomb complexes PRC1 and PR-DUB, but not with PRC2, for developmental stability.
Cyclin G binds polytene chromosomes at many sites and co-localizes extensively with Ph and Asx suggesting a potential interaction with PRC1 and PR-DUB on chromatin [28,31]. Sce and caly encode antagonistic enzymes of the PRC1 and PR-DUB complexes, respectively. Sce, aka dRing, ubiquitinates histone H2A on lysine 118 (H2AK118ub) whereas Calypso, aka dBap1, deubiquitinates the same H2A residue [34,35]. To investigate whether Cyclin G was related to these ubiquitin ligase/deubiquitinase activities, we immunostained polytene chromosomes from w1118 larvae with anti-Cyclin G and anti-human H2AK119ub antibodies (homologous to Drosophila H2AK118ub) [36,37]. Cyclin G and H2AK118ub co-localized extensively on chromosome arms suggesting that Cyclin G transcriptional activity might somehow be connected to this histone mark (Fig 4A). Wing imaginal discs presented a uniform pattern of H2AK118ub. When either CycGΔP or CycGΔEΔP was expressed in the posterior compartment of wing imaginal discs using the en-Gal4 driver, the global amount of H2AK118ub was not markedly modified (Fig 4B and 4C). Similarly, clones expressing CycGΔP or CycGΔEΔP showed the same amount of H2AK118ub than control GFP clones (Fig 4D, 4E and 4F). We thus concluded that high FA was not related to a global perturbation of H2AK118 ubiquitination.
Cyclin G controls transcription of the homeotic gene Abdominal-B and more specifically behaves as an Enhancer of PcG gene for the regulation of homeotic gene expression [31,38]. However, the high number of Cyclin G binding sites on polytene chromosomes suggests that it has many other transcriptional targets. We thus hypothesized that CycGΔP-induced FA might be related to the deregulation of Cyclin G transcriptional targets. To further address the role of Cyclin G in transcriptional regulation, we deep-sequenced transcripts from da-Gal4, UAS-CycGΔP/+ wing imaginal discs. Considering that the Gal4 transactivator might unspecifically interfere with transcription of some genes, we deep-sequenced da-Gal4/+ wing imaginal disc transcripts as negative control.
Sequence reads were aligned with the D. melanogaster genome to generate global gene expression profiles. With an adjusted p-value threshold of 0.05, we retrieved 530 genes whose expression was significantly different between da-Gal4, UAS-CycGΔP/+ and the da-Gal4/+ control (S7 Table). Surprisingly, expression of CycG was only weakly induced in da-Gal4, UAS-CycGΔP/+ imaginal discs (1.3 fold). To test the hypothesis that Cyclin G could, directly or not, regulate its own repression, we quantified expression of the endogenous CycG gene by RT-qPCR using primers located in the 3’UTR that were not present in the transgene. Indeed, expression of endogenous CycG was significantly decreased upon CycGΔP induction, suggesting that Cyclin G repressed its own expression (Fig 5A and S8 Table). Among the 530 genes deregulated in da-Gal4, UAS-CycGΔP/+ imaginal discs, 216 were up-regulated and 314 down-regulated. Up-regulated genes were enriched in the Gene Ontology categories cytoplasmic translation and translational initiation whereas down-regulated genes were enriched in the category mitochondrial respiratory chain complex (Fig 5B and S9 Table). By RT-qPCR, we verified that several ribosomal protein genes (RpL15, RpL7 and Rack1) were over-expressed in da-Gal4, UAS-CycGΔP/+ imaginal discs (Fig 5C and S10 Table). In conclusion, CycGΔP-induced FA correlates with activation of genes involved in translation and repression of genes involved in energy production.
We next sought to determine the direct transcriptional targets of Cyclin G by ChIP-seq. To do this, we took advantage of the transgenic +/ UAS-CycGΔP line in which Cyclin G was fused to a Myc tag. We performed ChIP experiments with anti-Myc antibodies using chromatin from +/ UAS-CycGΔP; da-Gal4/+ wing imaginal discs or da-Gal4/+ wing imaginal discs (mock ChIP). 3363 significant peaks were identified (IDR < 0.05) in +/ UAS-CycGΔP; da-Gal4/+ wing imaginal discs. Among these peaks, 1045 were located on a subset of 889 genes, most of them corresponding to TSS (Fig 6A and 6B, and S11 and S12 Tables). We could not formally exclude that there were differences in Myc-Cyclin G binding and endogenous Cyclin G binding. However, the increase in CycG mRNA being low (1.3 fold), we assumed that the midly over-expressed exogenous CycGΔP would respect the binding pattern of the endogenous protein. Snapshots of some TSS bound genes (RPL7, RPL5, Rack1, CycG) are shown on Fig 7. ChIP-qPCR analysis of these four genes confirmed that Cyclin G peaked on their TSS and decreased on gene body (Fig 6C and S13 Table). As endogenous CycG was down-regulated when CycGΔP was expressed and Cyclin G bound its own TSS, this confirmed that Cyclin G represses its own promoter.
The 889 Cyclin G-bound genes were enriched in GO categories cytoplasmic translation and protein phosphorylation (Fig 6D). Comparison of the 530 genes deregulated in wing imaginal discs expressing CycGΔP with these 889 genes showed that only 62 genes were both deregulated (39 up- and 23 down-regulated) and bound by Cyclin G (S14 Table), suggesting that most of the effects on gene expression were indirect. Strikingly, the 39 up-regulated genes were significantly enriched in the GO category translation (GO:0002181~cytoplasmic translation, 14 genes, enrichment score: 12.31, adjusted p-value 2.07E-16) and the 23 down-regulated ones in the GO category cytochrome-c oxidase activity (GO:0004129~cytochrome-c oxidase activity, 3 genes, enrichment score: 2.01, adjusted p-value 4.40E-02).
Using published datasets, we analysed the correlation between regions bound by Cyclin G in +/UAS-CycGΔP; da-Gal/+ imaginal discs and those enriched in PRC1, or PR-DUB components, RNAPolII, or H3K27me3 in wild type wing imaginal discs (S15 Table). Cyclin G-bound regions were significantly exclusive from H3K27me3, corroborating previous results [31]. The same comparisons were performed gene-wise and gave the same results (Fig 8). Importantly, 78% of Cyclin G-bound genes were also bound by RNAPolII and Pc. We cannot exclude that Cyclin G might co-localize with PRC1 in some cells and with RNAPolII in others. Alternatively, Cyclin G-bound genes might be simultaneously bound by PRC1 and RNAPolII. Considering RNAPolII as a proxy for transcriptional activity, and given that PRC1 has the ability to block transcriptional initiation [39], these genes would be most probably paused. Cyclin G also shared many target genes with Asx but, though Asx and Calypso belong to the PR-DUB complex, Cyclin G did not share binding sites with Calypso. This indicates either that the interaction between Cyclin G and Asx destabilizes the PR-DUB complex or that it takes place outside PR-DUB.
These genome-wide analyses indicate that Cyclin G coordinates the expression of genes involved in translation and energy production. However, only a few Cyclin G-bound genes were deregulated in da-Gal4, UAS-CycGΔP/+ imaginal discs. To better understand how Cyclin G orchestrates target gene expression, we developed a systems biology approach. We first built an interactome based on genes expressed in control da-Gal4/+ wing imaginal discs (with a cutoff of 10 reads). Edges corresponding to protein-protein interactions (PPI) and transcription factor-gene interactions (PDI) were integrated into this interactome through DroID [40]. The resulting wing imaginal disc interactome, further called the WID network, was composed of 9,966 nodes (proteins or genes) connected via 56,133 edges (interactions) (WID.xmml). We then examined the position of Cyclin G in this network. Betweenness centrality—i.e. the total number of non-redundant shortest paths going through a certain node–is a measure of centrality in a network [41]. A node with a high betweenness centrality could control the flow of information across the network [42]. With 8.32E-03, Cyclin G had one of the highest value of betweenness centrality of the network, ranking at the 30th position among the 9,966 nodes. This suggests that Cyclin G represents a hub in the WID network.
In order to isolate a connected component of the WID network that showed significant expression change when CycGΔP is expressed, we introduced the expression matrix describing expression of the 530 significantly deregulated genes in the WID network. We next used JactiveModules to identify sub-networks of co-deregulated genes [43]. A significant sub-network of 222 nodes and 1069 edges centred on Cyclin G was isolated (Z score 48.53). This sub-network was laid out according to functional categories (Fig 9, CycG_subnetwork.xmml). Four modules composed of genes respectively involved in transcription, mitochondrial activity, translation, and metabolism, were found to be highly connected to Cyclin G. Strikingly, the “translation” module was mainly composed of genes up-regulated in da-Gal4, UAS-CycGΔP/+ wing imaginal discs. On the contrary, the “mitochondrion” and “metabolism” modules were mainly composed of genes down-regulated in da-Gal4, UAS-CycGΔP/+ wing imaginal discs. Interestingly, Cyclin G-bound genes in this sub-network were enriched in genes bound by the PRC1 proteins Pc, Ph and Psc, as well as by RNAPolII, and to a lesser extent by Asx (Fig 10).
The CycG gene of D. melanogaster encodes a cyclin involved in transcriptional control, cell growth and the cell cycle [26,28,38]. Mild overexpression of Cyclin G induces high fluctuating asymmetry (FA), notably of wings, providing a unique tool to investigate the genetic bases of developmental stability [6,7]. Cyclin G interacts physically with two chromatin regulators of the Enhancers of Trithorax and Polycomb family (ETP), and genetically with Polycomb-group (PcG) genes [31]. This prompted us to examine the role of these interactions in developmental stability and to investigate deeply the function of Cyclin G in transcriptional regulation.
CycG-induced wing FA only occured when the deregulation was local, i.e. in wing imaginal discs. Although we cannot exclude that Cyclin G induces expression of a systemic factor that is released into the haemolymph, our observations suggest that CycG maintains developmental stability through an organ-autonomous mechanism which would not involve the Dilp8/Lgr3 pathway. Many Cyclin G targets in the wing imaginal discs are also bound by PRC1, by Asx and by RNAPolII, but are not enriched in H3K27me3. In agreement, mutations in PRC1 and PR-DUB, but not in PRC2, strongly increase CycG-induced FA. We did not observe any significant overlap between Cyclin G-bound genes and binding sites for Calypso, the second component of PR-DUB. Yet, caly mutations also strongly increase CycG-induced FA. Thus, the role of PR-DUB in this context remains to be clarified. PRC1 and PR-DUB contain antagonistic enzymes (Sce/dRing and Calypso) that respectively ubiquitinates and deubiquitinates H2AK118. However, the global level of H2AK118 ubiquitination is not modified in tissues where Cyclin G isoforms are overexpressed suggesting that this epigenetic mark is not involved in developmental stability. Cyclin G targets strikingly remind the neo-PRC1 targets described in [44]. Indeed, PRC1 components are redeployed during development and control these neo-PRC1 targets that are robustly transcribed, not enriched in H2AK118ub, and on which PRC1 is recruited independently of PRC2 [44]. It was proposed that PRC1 limits the expression of these neo-PRC1 genes that are mainly involved in cell proliferation, cell signaling and polarity, thus explaining its tumor suppressor role [44]. Hence, Cyclin G might participate with PRC1 and PR-DUB in the control of these neo-PRC1 genes and this might be important for developmental stability.
Drosophila Cyclin G and the two vertebrate G-type cyclins, CCNG1 and CCNG2 exhibit a complex relationship to growth, on the one hand promoting it, [45–48] and on the other hand, slowing down or even blocking the cell cycle [26,49–52]. Accordingly, we found that Cyclin G controls a small regulatory sub-network connecting genes involved in metabolism, mitochondrial activity and translation. Notably, many genes involved in basic metabolism, such as Gapdh1, Gapdh2 or Jafrac1, are down-regulated in the CycGΔP context, which also agrees with the small mean size of CycGΔP flies, organs and cells. A large scale analysis of the Drosophila wing imaginal disc proteome has recently shown that wing size correlates with some basic metabolic functions, positively with glucose metabolism and negatively with mitochondrial activity, but not with ribosome biogenesis [53]. However, in CycGΔP flies while mitochondrial genes are negatively regulated, ribosomal biogenesis genes are simultaneously positively regulated. Although transcriptome variations are probably not a direct image of proteome variations, our data suggest that robustness of wing size correlates with the fine-tuning of these key functions relative to each other. Corroborating our study, a mutant for the AAA mitochondrial protease FTSH4 in Arabidospsis thaliana displays high variability of sepal size and shape associated with early ROS production [54]. Furthermore, a recent analysis of human mesothelioma cells also points to a role of BAP1 and the PR-DUB complex in mitochondrial function and ROS homeostasis [55]. It will be important in the future to understand how epigenetic control of genes involved in mitochondrial activity and control of growth impact developmental stability and how deregulation of these processes might lead to cancer.
The pPMW-attB plasmid was built as follows: the Gateway vector pPMW [56] was linearized by digestion with NsiI; the attB sequence was amplified from pUASTattB [57] (using primers attB-NsiIF and attB-NsiIR (S16 Table) and the PCR product was digested with NsiI; the digested PCR product and the linearized plasmid were ligated and sequenced. This plasmid was deposited at Addgene (plasmid # 61814).
The full-length CycG cDNA (CycGFL, encoding the 566 amino-acid protein) was amplified from S2 cell cDNAs using primers CycGnF and CycGnR. cDNAs encoding truncated forms of Cyclin G (CycGΔP, Cyclin G deleted of the putative PEST domain corresponding to amino-acids 542 to 566; CycGΔE, Cyclin G deleted of the ETP-interacting domain corresponding to amino-acids 1 to 130; CycGΔEΔP, Cyclin G deleted of both domains) were amplified from the full-length CycG cDNA using primers CycGnF and CycG541R, CycG130F and CycGnR, and CycG130F and CycG541R, respectively (S16 Table). The PCR products were cloned into pENTR/D-TOPO (Invitrogen), transferred into pPMW-attB and the resulting plasmids pPMW-attB-CycGFL, pPMW-attB-CycGΔP, pPMW-attB-CycGΔE, pPMW-attB-CycGΔEΔP were sequenced.
Flies were raised on standard yeast-cornmeal medium at 25°C.
UAS-Myc-CycG transgenic lines were obtained by PhiC31-integrase mediated insertion into strain y1M{vas-int.Dm}ZH-2Aw*;M{3xP3-RFP.attP'}ZH-51C (stock BL-24482). Plasmids pPMW-attB-CycGFL, pPMW-attB-CycGΔP, pPMW-attB-CycGΔE and pPMW-attB-CycGΔEΔP were injected into embryos, G0 adults were back-crossed to yw, and G1 transformants were crossed to yw again to obtain G2 transformants (BestGene Inc.). Transformants were individually crossed with yw; Sp/CyO, and the curly wing siblings were crossed with each other. Homozygous transgenic lines were then obtained by crossing 5 females and 5 males. The resulting lines were named UAS-CycGFL, UAS -CycGΔP, UAS-CycGΔE and UAS -CycGΔEΔP.
Gal4 drivers used were daughterless-Gal4 (da-Gal4), engrailed-Gal4 (en-Gal4, nubbin-Gal4 (nub-Gal4), optomotor-blind-Gal4 (omb-Gal4), rotund-Gal4 (rn-Gal4), scalloped-Gal4 (sd-Gal4), teashirt-Gal4 (tsh-Gal4), vestigial-Gal4 (vg-Gal4) (from the Bloomington Drosophila stock center), and Insulin-like peptide 3-Gal4 (dILP3-Gal4), neuropeptide F-Gal4 (NPF-Gal4), Pigment-dispersing factor-Gal4 (Pdf-Gal4), period-Gal4 (per-Gal4), phantom-Gal4 (phm-Gal4), Prothoracicotropic hormone-Gal4 (Ptth-Gal4), R10B09-Gal4 (kind gifts from Dr M. Dominguez’s lab).
To generate clones, the strain hs-flp; tub>stop>Gal4, UAS-GFP/CyO (a kind gift of Dr. M. Gho) was crossed with the UAS-CycG strains. After 24h of egg-laying, embryos were allowed to develop 24h. Then, they were heat-shocked at 37°C during 1h, allowed to develop 24h more, and heat-shocked a second time at 37°C during 1h.
The da-Gal4, UAS-CycGΔP third chromosome, obtained by recombination of da-Gal4 with the original UAS-CycGΔP transgene (RCG76), was used to test genetic interactions between CycG and several PcG or ETP mutations [31]. PcG and ETP alleles used are described in Table 1.
For FA analyses, five replicate crosses were performed for each genotype, wherein 6 females carrying a Gal4 driver were mated with 5 males carrying a CycG transgene. Parents were transferred into a new vial every 48h (three times) then discarded. Thirty females were sampled from the total offspring of the desired genotype. For genetic interactions with PcG or ETP mutants, crosses were performed similarly except that 6 PcG or ETP mutant females were mated either with 5 da-Gal4, UAS-CycGΔP males, or with 5 da-Gal4 males as control.
Right and left wings of 30 sampled females were mounted on slides, dorsal side up, in Hoyer’s medium. Slides were scanned with a Hamamatsu Nanozoomer Digital Slide scanner. Wing pictures were exported into tif format using NDP.view. All wings were oriented with the hinge to the left. Image J was used to digitize either landmarks 3 and 13 to measure wing length, or the 15 landmarks to measure more accurately wing centroid size when indicated (S2 Fig). All wings were measured twice. Analysis of wing size FA, the variance of the difference between the left and the right wing in a population, was performed as described previously [6]. The FA10 index was used to estimate FA, i.e. FA corrected for measurement error, directional asymmetry and inter-individual variation [9]. For all genotypes, the interaction individual*side was significant, indicating that FA was larger than measurement error. F-tests were performed to compare the different genotypes.
Immunostainings were performed as described in [31]. Primary anti-H2AK119ub antibodies (Cell Signaling D27C4) were used at a 1:40 dilution.
Wing imaginal discs from da-Gal4/UAS-CycGΔP and da-Gal4/+ third instar female larvae were dissected, and total RNAs were extracted as previously described except that 150 discs homogenized by pipetting were used for each extraction [58]. Three biological replicates (wing imaginal discs dissected from three independent crosses) were generated for each genotype. Library preparation and Illumina sequencing were performed at the ENS Genomic Platform (Paris, France). PolyA RNAs were purified from 1 μg of total RNA using oligo(dT). Libraries were prepared using the TruSeq Stranded mRNA kit (Illumina). Libraries were multiplexed by 6 on 2 flowcell lanes. 50 bp single read sequencing was performed on a HiSeq 1500 device (Illumina). Number of reads are shown on S17 Table. Reads were aligned with the D. melanogaster genome (dm6, r6.07) using TopHat 2 (v2.0.10) [59]. Unambiguously mapping reads were then assigned to genes and exons described by the Ensembl BDGP5 v77 assembly, by using the “summarizeOverlaps” function from the “GenomicAlignments” package (v 1.2.2) in “Union” mode [60]. Library size normalization and differential expression analysis were both performed with DESeq 2 (v 1.6.3). Genes with adjusted p-value below 0.05 were retained as differentially expressed [61]. Gene Ontology analysis was performed using DAVID [62].
For RT-qPCR validations, RNAs were extracted from wing imaginal discs and treated with Turbo DNAse (Ambion). cDNA were synthesized with SuperScript II Reverse transcriptase (Invitrogen) using random primers. RT-qPCR experiments were carried out in a CFX96 system (Bio-Rad) using SsoFast EvaGreen Supermix (Bio-Rad). Two biological replicates (cDNA from wing imaginal discs of larvae coming from independent crosses) and three technical replicates (same pool of cDNA) per biological replicate were performed for each genotype. Expression levels were quantified with the Pfaffl method [63]. The geometric mean of two reference genes, Lamin (Lam) and rasputin (rin), the expression of which did not vary when CycGΔP was expressed, was used for normalization [64]. Sequences of primer couples are listed in S16 Table.
An interactome was built using Cytoscape (v 2.8.3) and the DroID plugin (v 1.5) to introduce protein-protein and transcription factor-gene interactions [40]. The jActiveModules plugin (v 2.23) was used to find sub-networks of co-deregulated genes in the interactome by using “overlap threshold” 0.8, “score adjusted for size”, and “regional scoring” [43].
Wing imaginal discs from +/UAS-CycGΔP; da-Gal4/+ and da-Gal4/+ third instar female larvae were used for ChIP-seq experiments. 600 wing imaginal discs were dissected (taking one disc per larva) in Schneider medium and aliquoted per 50 on ice. The 12 microtubes were treated as described in [58] with minor modifications. Discs were fixed at 22°C. 12 sonication cycles were performed (Diagenode Bioruptor sonifier; cycles of 30'' ON, 30'' OFF, high power). After centrifugation, the 12 supernatants were pooled, homogenized, and 2% were kept (Input). The remaining fragmented chromatin was redistributed into 12 tubes and each tube was adjusted to 1 mL with 140 mM NaCl, 10 mM Tris-HCl pH 8.0, 1 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, 0.1% BSA, Roche complete EDTA-free protease inhibitor cocktail. For immunoprecipitation, 3 μg of anti-Myc antibody (Abcam 9132) were added per tube. Two biological replicates were performed for +/UAS-CycGΔP; da-Gal4/+ and one for da-Gal4/+.
Library preparation and Illumina sequencing were performed at the ENS Genomic Platform (Paris, France). Libraries were prepared using NEXTflex ChIP-Seq Kit (Bioo Scientific), using 38 ng of IP or Input DNA. Libraries were multiplexed by 10 on one flowcell run. 75 bp single read sequencing was performed on a NextSeq 500 device (Illumina). Reads were filtered by the "fastq_quality_filter" command from the "fastx-Toolkit" package (http://hannonlab.cshl.edu/fastx_toolkit/), using a threshold of 90% bases with mapping quality ≥ 20. Reads are shown on S18 Table. Those that successfully passed the filtering step were aligned to the D. melanogaster genome (dm6, r6.07) using Bowtie 2 (http://bowtie-bio.sourceforge.net/bowtie2/) (v 2.1.0) with default parameters [65]. Peaks were called by MACS2 (v 2.1.0) by comparing each ChIP to its input library, with fragment size fixed at 110 bp and otherwise default parameters [66]. Peak reproducibility between the two biological replicates was then analysed with the IDR method (https://www.encodeproject.org/software/idr/) [67]. Briefly, an IDR score was assigned to each peak by the "batch-consistency-analysis" function, using the recommended parameters for MACS peaks (peak ranking based on p-value). Peaks below the 0.05 threshold were considered reproducible. The overlapping reproducible peaks from both replicates were fused using the BEDtools suite "merge" function [68], resulting in the final list of peaks kept for subsequent analysis. Cyclin G-bound genes were defined as genes from the genome annotation file (dm6, r6.07) which overlapped at least one of these Cyclin G peaks, as obtained by the BEDtools suite "intersect" function.
For ChIP-qPCR validations, ChIPs were performed similarly with the anti-Myc antibody. Rabbit IgG (Diagenode) were used as negative control (mock). qPCR experiments were carried out in a CFX96 system (Bio-Rad) using SsoFast EvaGreen Supermix (Bio-Rad). Three biological replicates–three technical replicates per biological replicate–were performed for each antibody and for the Input. Sequences of primer couples are listed in S16 Table. Data were normalized against Input chromatin.
Heatmaps and aggregation plots of Cyclin G signal over gene bodies and TSS were generated using the ngsplot package. (https://github.com/shenlab-sinai/ngsplot) [69]. Some genes with spurious signal (such as genes from the histone complex) were excluded from the analysis based on signal uniformity over the full length of the gene (cumulative derivative of Cyclin G signal over gene length = 0).
Genomic loci enriched for Polycomb (Pc), Posterior Sex Comb (Psc), Polyhomeotic (Ph), RNA Polymerase II (RNAPolII) and H3K27me3 in wild type imaginal discs of third instar larvae were retrieved from GEO (GSE42106) [70–71] (H3K27me3_WholeWingDisc GSM1032567, PcRJ_AnteriorWingDisc GSM1032571, PcRJ_PosteriorWingDisc GSM1032574, Ph_WholeWingDisc GSM1032576, PolII_WholeWingDisc GSM1032577, Psc_WholeWingDisc GSM1032578. Binding sites for Pc in the whole wing disc were defined as the overlap between Pc binding sites in the anterior and posterior wing disc compartment, as obtained by the BEDtools "intersect" function. For Asx and Calypso, the bed files were a kind gift from Dr. J. Müller [34]. The mappability file for dm6 genome with 25 nt reads (the smallest size in the compared data) was generated using the Peakseq code (http://archive.gersteinlab.org/proj/PeakSeq/ Mappability_Map/Code). The overall size of the mappable genome was used as the effective genome size for the GAT software (https://github.com/AndreasHeger/gat) to assess the significance of the overlap between peaks of Cyclin G and other factors [72]. As GAT performs a two-tailed test, it displays low p-values both for significant overlap and exclusion (as between Cyclin G and H3K27me3).
Gene overlap significance assessment was made as follows: under the null hypothesis, genes that are enriched for Asx, Calypso, Pc, Psc, Ph, RNAPolII or H3K27me3 in wild type imaginal discs of third instar larvae should not exhibit any bias towards Cyclin G targets. Thus, the overlap between n enriched genes and K Cyclin G targets genes should be explained by random sampling without replacement of n genes within the total amount N of D. melanogaster genes. The amount of overlap under the null hypothesis X follows a hypergeometric law: X~HY(K,N,n). The significance of the observed overlap k was computed as the probability of observing X higher or equal to k under the null hypothesis: P(X ≥ k).
The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus [70] and are accessible through GEO Series accession number GSE99462 for RNA-seq, and GSE99461 for ChIP-seq.
|
10.1371/journal.pcbi.1003251 | Cellular Adaptation Facilitates Sparse and Reliable Coding in Sensory Pathways | Most neurons in peripheral sensory pathways initially respond vigorously when a preferred stimulus is presented, but adapt as stimulation continues. It is unclear how this phenomenon affects stimulus coding in the later stages of sensory processing. Here, we show that a temporally sparse and reliable stimulus representation develops naturally in sequential stages of a sensory network with adapting neurons. As a modeling framework we employ a mean-field approach together with an adaptive population density treatment, accompanied by numerical simulations of spiking neural networks. We find that cellular adaptation plays a critical role in the dynamic reduction of the trial-by-trial variability of cortical spike responses by transiently suppressing self-generated fast fluctuations in the cortical balanced network. This provides an explanation for a widespread cortical phenomenon by a simple mechanism. We further show that in the insect olfactory system cellular adaptation is sufficient to explain the emergence of the temporally sparse and reliable stimulus representation in the mushroom body. Our results reveal a generic, biophysically plausible mechanism that can explain the emergence of a temporally sparse and reliable stimulus representation within a sequential processing architecture.
| Many lines of evidence suggest that few spikes carry the relevant stimulus information at later stages of sensory processing. Yet mechanisms for the emergence of a robust and temporally sparse sensory representation remain elusive. Here, we introduce an idea in which a temporal sparse and reliable stimulus representation develops naturally in spiking networks. It combines principles of signal propagation with the commonly observed mechanism of neuronal firing rate adaptation. Using a stringent numerical and mathematical approach, we show how a dense rate code at the periphery translates into a temporal sparse representation in the cortical network. At the same time, it dynamically suppresses trial-by-trial variability, matching experimental observations in sensory cortices. Computational modelling of the insects olfactory pathway suggests that the same principle underlies the prominent example of temporal sparse coding in the mushroom body. Our results reveal a computational principle that relates neuronal firing rate adaptation to temporal sparse coding and variability suppression in nervous systems.
| The phenomenon of spike-frequency adaptation (SFA) [1], which is also known as spike-rate adaptation, is a fundamental process in nervous systems that attenuates neuronal stimulus responses to a lower level following an initial high firing. This process can be mediated by different cell-intrinsic mechanisms that involve a spike-triggered self-inhibition, and which can operate in a wide range of time scales [2]–[4]. These mechanisms are probably related to the early evolution of the excitable membrane [5]–[7] and are common to vertebrate and invertebrate neurons, both in the peripheral and central nervous system [8]. Nonetheless, the functional consequences of SFA in peripheral stages of sensory processing on the stimuli representation in later network stages remain unclear. For instance, light adaptation in photoreceptors strongly shapes their responses [9], [10] and affects stimulus information in second-order neurons [11]. In a seminal work by Hecht and colleagues [12], it was shown that during dark adaptation, 10 or less photon absorptions in the retina were sufficient to give a sensation of light within a millisecond of exposure and the response variability could be largely accounted for by quantum fluctuations. This is an interesting empirical result, and still it is theoretically puzzling that the intrinsic noise of the nervous system [13] has only little influence on the detection of such an extremely weak stimulus. A proposal by Barlow [14] suggested that successive processing in sensory neural pathways decrements the number of response spikes and therefore the informativeness of each spike increases while the level of noise decreases. However, it remains unclear how such temporally sparse spike responses can reliably encode information in the face of the immense cortical variability [15] and the sensitivity of cortical networks to small perturbations [16], [17].
The widespread phenomenon of a dynamically suppressed trial-by-trial response variability in sensory and motor cortices [18]–[20] along with a sparse representation [21], [22] hints at an increased reliability of the underlying neuronal code and may facilitate the perception of weak stimuli. However, the prevailing cortical network models of randomly connected spiking neurons, where the balance of excitation and inhibition is quickly reinstated within milliseconds after the arrival of an excitatory afferent input, do not capture this dynamic [16], [17], [23]–[25]. Recent numerical observations suggest that a clustered topology of the balanced network [26] or attractor networks with multi-stability [27] provide possible explanations for suppressing cortical variability during afferent stimulation.
In this study, we introduce an alternative and unified description in which a temporally sparse stimulus representation and the transient increase of response reliability emerge naturally. Our approach exploits the functional consequences of SFA in multi-stage network processing. Here, we show that the SFA mechanism introduces a dynamical non-linearity in the transfer function of neurons. Subsequently, the response onset becomes progressively sparser when transmitted across successive processing stages. We use a rigorous master equation description of neuronal ensembles [28]–[30] and numerical network simulations to arrive at the main result that the self-regulating effect of SFA causes a stimulus-triggered reduction of firing variability by modulating the average inhibition in the balanced cortical network. In this manner the temporally sparse representation is accompanied by an increased response reliability. We further utilize this theoretical framework to demonstrate the generality of this effect in a highly structured network model of insect olfactory sensory processing, where sequential neuronal adaptation readily explains the ubiquitously observed sparse and precisely timed stimulus response spikes at the level of the so-called Kenyon cells [31]–[34]. Our experimental results qualitatively supports this theoretical prediction.
To examine how successive adapting populations can achieve temporal sparseness, first we mathematically analyzed a sequence of neuronal ensembles (Figure 1A), where each ensemble exhibits a generic model of mean firing rate adaptation by means of a slow negative self-feedback [2], [28], [35] (Materials and Methods). This sequence of neuronal ensembles should be viewed as a caricature for distinct stages in the pathway of sensory processing. For instance, in the mammalian olfactory system the sensory pathway involves several stages from the olfactory sensory neurons to the olfactory bulb, the piriform cortex, and then to higher cortical areas (Figure 1A).
The mean firing rate in the steady-state of a single adaptive population can be obtained by solving the rate consistency equation, , where is the equilibrium mean firing rate of the population, , and are coupling strength, mean and variance of the total input into the population, respectively, is the response function (input-output transfer function, or curve) of the population mean activity, is the quantal conductance of the adaptation mechanism per unit of firing rate, and is the adaptation relaxation time constant [2], [28], [35]. The firing rate model assumes that individual neurons spike with Poisson statistics, and that the adaptation level only affects mean input into neurons, resulting in a change to the steady-state mean firing rate. It is known that any sufficiently slow modulation () linearizes the steady-state solution, , due to the self-inhibitory feedback being proportional to the firing rate (see Materials and Methods) [36].
Here, for simplicity, we studied the case where all populations in the network exhibit the same initial steady-state rate. This is achieved by adjustment of a constant background input to population , given (doted arrows, Figure 1A), resembling the stimulus irrelevant interactions in the network. All populations are coupled by the same strength . First, we calculated the average firing rate dynamics of the populations' responses following a step increase in the mean input to the first layer (black arrow, Figure 1A). By solving the dynamics of the mean firing rate and adaptation level concurrently, we obtained the mean-field approximation of the populations' firing rates (Materials and Methods). As it is typical for adapting neurons, the responses of each population consisted of a fast transient following stimulus onset before it converges to the new steady-state (tonic response part) with a stable focus (Materials and Methods). The Figure 1B shows the mean firing rate of three consecutive populations. The phasic response to the step increase in the input is preserved across stages. However, the tonic response becomes increasingly suppressed in the later stages (Figure 1B). This phenomenon is a general feature of successive adaptive neuronal populations with a non-linear transfer function which, is linearized in steady-state due to adaptation negative feedback [36]. This result emerges as the change in the population mean rate that can be determined by solving the rate consistency equation now for a step change in the input . The necessary condition for the suppression of the steady-state responses is a sufficiently strong adaptation (Materials and Methods). It is worth to note that the populations exhibit under-shoots after the offset of the stimulus (Figure 1B). This is due to the adaptation level that accumulated during the evoked state.
The result in this sub-section (Figure 1B–C) was established with a current based leaky integrated-and-fire response function. However, the analysis presented here extends to the majority of neuronal transfer functions since the stability and linearity of the adapted steady-states are granted for many biophysical transfer functions [35], [36]. This simple effect leads to a progressively sparser representation across successive stage of a generic feed-forward adaptive processing. We assess temporal sparseness by computing the time-dependent integral , where is the mean firing rate of population and is the increasing observation time window. Normalization of this measure by the spike count in the first population indicates that responses in the later stages of the adaptive network are temporally sparser (Figure 1C). This is expressed in the sharp increase of the rate integral during the transient response, whereas the first population integral shows an almost constant increase in the number of spikes.
Does the suppression of the adapted response level impair the information about the presence of the stimulus? To explore this, we studied a secondary increase in stimulus strengths with an equal magnitude, after 1 second, when the network has relaxed to the evoked equilibrium (Figure 1D). The secondary stimulus jump induced a secondary phasic response of comparable magnitude in the first population (Figure 1D). However, in the later populations this jump evoked an increased peak rate in the phasic response (Figure 1D). Notably, the coupling factor between the populations shapes this phenomena. Here, we adjusted to achieve an equal onset response magnitude across the populations for the first stimulus jump at , and a slight increase in the population onset response in the first population is amplified in the later stages. This is due to the fact that the later stages accumulated less adaptation in their evoked steady-state (the level of adaptation is proportional to mean firing rate). Importantly, this result confirms that the sustained presence of the stimulus is indeed stored at the level of cellular adaptation [37], even though it is not reflected in the firing rate of the last population (Materials and Methods). Therefore, regardless of the absolute amplitude of responses, the relative relation between secondary and initial onset keeps increasing across layers. This type of secondary overshoot is also experimentally known as sensory sensitization or response amplification, where an additional increase in the stimulus strength significantly enhances the responsiveness of later stages after the network converged to an adapted steady-state [10], [11], [38], [39].
The mean firing rate approach as above is insufficient to determine how reliable the observed response transients are across repeated simulations. In a spiking model of neo-cortex (the balanced network), self-generating recurrent fluctuations strongly dominate the dynamics of interactions and produce highly irregular and variable activity [24]. This prevailing cortical model suggests that balance of excitation and inhibition is quickly reinstated within milliseconds after the onset of an excitatory input and adjusts the network fluctuations level [23]. Therefore, it has been questioned whether a few temporarily meaningful action potentials could reliably encode the presence of a stimulus [16].
To investigate the reliability of adaptive mapping from a dense stimulus to a sparse cortical spike response across successive processing stages we employed the adaptive population density formalisms [28], [29] (Materials and Methods) along with numerical network simulations. We embedded a two-layered sensory network with an afferent ensemble projecting to a cortical network (Figure 2A). The afferent ensemble consisted of 4,000 adaptive neurons that included voltage dynamics, conductance-based synapses, and spike-induced adaptation [28]. It resembles the sub-cortical sensory processing and each neuron in the afferent ensemble projects randomly to 1% of the neurons in the cortical network. This is a large circuit of the balanced network (Figure 2A) with 10,000 excitatory and 2,500 inhibitory neurons with a typical random diluted connectivity of 1%. The spiking neuron model in the cortical network again includes voltage dynamics, conductance-based synapses, and spike-induced adaptation [28]. All neurons are alike and parameters are given in Table 3 in Muller et al. [28]. With appropriate adjustment of the synaptic weights, the cortical network operates in a globally balanced manner, producing irregular, asynchronous activity [24], [40], [41]. The distribution of firing rates for the network approximates a power-law density [42] with an average firing rate of Hz (Figure 2B) and the coefficients of variation () for the inter-spike intervals are centered at a value slightly greater than unity (Figure 2C) indicating the globally balanced and irregular state of the network [41]. Noteworthy is that the activity of neurons in both stages is fairly incoherent and spiking in each sub-network is independent. Therefore, one can apply an adiabatic elimination of the fast variables and formulate a population density description where a detailed neuron model reduces to a stochastic point process [28], [29] that provides an analytical approximation of the spiking dynamics and helps understanding the network simulation results in this section. This framework allows for a detailed study of a large and incoherent network without the need of numerical simulations (Materials and Methods).
The background input is modelled as a set of independent Poisson processes that drive both sub-networks (dashed arrows, Figure 2A). The stimulus dependent input is an increase in the intensity of the Poisson input into the afferent ensemble (solid arrow, Figure 2A). Before the stimulus became active at time , a typical neuron showed an irregular spiking activity in both network stages (Figure 2B–D). Whenever a sufficiently strong stimulus is applied all neurons in the afferent ensemble exhibited a transient response before the population mean firing rate converges back to a new level of steady-state (Figure 2D,E). The population firing rate of neurons in the cortical network also exhibited a transient evoked response (Figure 2D,E). However, in the balanced network individual neurons are heterogeneous in their responses (Figure 2D), since the number of inputs from afferent and recurrent connectivity are random. In contrast to the rate model in the previous section where individual neurons were assumed to spike in a Poissonian manner, the adaptive neuron model in the neural network simulation operates far away for this assumption since the adaptation endows a long lasting memory effect on the spike times [28], [29] that extends beyond the last spike. The time constant of this memory is determined by the time constant of adaptation (τs = 110 ms). This non-renewal statistics determines the shape of the transient component of the population response in Figure 2E. The spiking irregularity shows that the evoked state in the afferent ensemble is more regular than its background. The balanced network still exhibits a fairly irregular spiking and its average stays approximately constant slightly above 1 (Figure 2D). The population firing rate in the numerical simulations (solid line, Figure 2E) follow well the adaptive population density treatment (filled circles, Figure 2E).
To measure the effect of neuronal adaptation on the temporal sparseness, we again computed the number of spikes per neuron after the stimulus onset, . We compare our standard adaptive network with an adaptation time constant of τs = 110 ms (solid lines, Figure 2F) to a weakly-adaptive control network (τs = 30 ms; dashed lines, Figure 2F). Note, that the adaptation time constant in the weakly adaptive network is about equal to the membrane time constant and therefore plays a minor role for the network dynamics. It showed that both sub-networks generated sharp population level phasic response, which in the case of the cortical network evoked a single sharply timed spike within the first ms in a subset of neurons (Figure 2B,F). In the control case, the response is non-sparse and response spikes are distributed throughout the stimulus period (dashed lines, Figure 2F). Overall, strong adaptation reduces the total number of stimulus-induced action potentials per neuron and concentrates their occurrence within an initial brief phasic response part following the fast change in the stimulus. This temporal sparseness is reflected in the cumulative number of spikes per neuron (Figure 2F) which increases sharply. Thus, in accordance with the results of the rate-based model in the previous section, one can conclude that the sequence of adaptive processing accounts for the emergence of a temporally sparse stimulus representation in a cortical population.
We also estimated the fraction of neurons that significantly changed their number of spikes after stimulus onset. By construction, all cells in the afferent ensemble, both in adaptive and weakly-adaptive cases, produce a significant response. However, neurons in the cortical layer are far more selective. In the weakly-adaptive network 58% of all neurons responded significantly. In the adaptive network this number drops to 36%. This is calculated by comparing the count distribution across trials in 200 ms windows before and after the stimulus onset (Wilcoxon rank sum test, p-value = 0.01).
To reveal the effect of adaptation on the response variability, we employed the time-resolved Fano factor [20], , which measures the spike-count variance divided by the mean spike count across repeated simulations. Spikes were counted in a 50 ms time window and a sliding of 10 ms [18]. As before, we compared our standard adaptive network (Figure 2G,I; τs = 110 ms) with the control network (Figure 2H,J; τs = 30 ms). Since the Fano factor is known to be strongly dependent on the firing rate, we adjusted the stimulus level to the latter such that the averaged steady state firing rates in both networks were mean-matched [18]. The input Poisson spike trains () translated into slightly more regular spontaneous () activity in the afferent ensemble (Figure 2G), as neuronal membrane filtering and refractoriness reduced the output variability. After the stimulus onset (), due to the increase in the mean input rate, the average firing rate increased, however the variance of the number of events per trial did not increase proportionally. Therefore, we observed a reduction in the Fano factor (Figure 2G). This phenomena is independent of the adaptation mechanism in the neuron model and a quantitatively similar reduction can be observed in the weakly adaptive afferent ensemble (Figure 2H). A comparison between our standard adaptive and the control case reveals that the adaptive network is generally more regular in the background and in the evoked state (Figure 2G,H). This is due to the previously known effect, where adaptation induces negative serial dependencies in the inter-spike intervals [29], [43] and as a result reduces the Fano factor [29], [44].
In the next stage of processing, the distribution of across neurons during spontaneous activity is high due to the self-generated noise of the balanced circuits [23], [24]. This closely follows a wide spread experimental finding where in the spontaneous activity of sensory and motor cortices [18]–[20] (Figure 2G,H). This highly variable regime can be achieved in the balance network with strong recurrent couplings [23]. Whenever a sufficiently strong stimulus was applied, the internally generated fluctuations in the adaptive balanced network were transiently suppressed, and as a result the Fano factor dropped sharply (Figure 2I). However, this reduction of the Fano factor is a temporary phenomenon and converges back to slightly above the baseline variability (Figure 2I). At the same time, the evoked steady-state firing remained in the irregular and asynchronous state (Figure 2D). Indeed, this transient effect corresponds to a temporally mismatch in the balanced input conditions to the cortical neurons since the self-inhibitory and slower adaption effect prevents a rapid adjustment to the new input regime. This can be observed in the time course of variability suppression that closely reflects the time constant of adaptation (Figure 2I). However, with stronger adaptive feedforward input we can prevent the return of the Fano factor to the base line, this phenomena is due to the regularizing effect of adaptation in the afferent ensemble. In this scenario, the afferent ensemble structured the input to the cortical ensemble, contributing to the magnitude of the observed variability reduction. Indeed, whenever the excitatory feedforward strength is considerably strong, relative to the recurrent connections, the cortical network moves away from the balanced condition. Thus, such strong input resets the internal spiking dynamics within the cortical network and as a result it regulates the spiking variability [45]. This mechanism evidently can be used to prevent the recovery of the high variability. However, we deliberately use a weak stimulation to focus on the transit suppression of cortical variability that is mediated by the slow self-regulation due to adaptation. For instance, under the control condition where a pure Poisson input (with similar synaptic strength) is provided to the cortical balanced network, the reduction in is reduced but the time scale of recovery remains unaltered (crosses in Figure 2I). We contrast this adaptive behavior with the variability dynamics in the weakly adaptive balanced network (Figure 2J). In this case there is no reduction in , because for a short adaptation time constant the convergence to the balanced state is very rapid [24]. The small increase in the input noise strength leads to an increase of the self-generated randomness of the balanced network [17], [23].
In the above comparison, we adjust the stimulus strength to achieve the same steady-state firing rate (tonic response) in the afferent and cortical ensemble. In a next step we studied the effect of adaptation on the detectability of a weak and transient peripheral signal, which might be impaired by the self-generated noise in the cortical network. To this end we employed the population density approach (Material and Methods) to study the mean and variability of the cortical network responses to a wide range of signal strengths. We change the stimulation protocol to a brief signal with a duration of ms over the spontaneous background. The stimulation magnitude is adjusted to elicit the same onset firing rate in the afferent ensemble network in both adaptive and weakly adaptive cases. By modification of the feed-forward coupling between afferent and cortical network relative to the intracortical recurrent coupling we study the circuit responses (Figure 3A). Evidently, the strength of the feed-forward coupling to the cortical ensemble modifies its spontaneous background, and therefore also the total adaptation level. The adaptive network proves more sensitive to brief and weak stimuli. It significantly magnifies the mean stimulus response in the adaptive network relative to the background. Even for a considerably weak stimulus the relative amplitude of the response to background firing is pronounced (Figure 3A). This result resembles the amplification of a transient in the sequence of adaptive networks as it is observed in the previous section (Figure 1A).
How reliable are the responses across trials? To answer this question we calculated the Fano factor (Material and Methods) for the cortical ensemble response in the above scenario. This calculation indicates that the response variability in the adaptive network is significantly lower than in the weakly adaptive network over a large range of the feed-forward coupling strength (Figure 3B). Interestingly, our results of the population density treatment quantitatively follow the former prediction based on a network simulation that in a balanced network without adaptation the variability initially increases with signal strength (dashed line in Figure 3B and Table 1 in [23]) and only after a critical level of the feed-forward strengths the recurrent noise is suppressed due to the stronger influence of excitatory inputs.
As a case study to demonstrate how the sequential effect of the adaptation shape responses, we investigated its contribution to the emergence of the reliable and sparse temporal code in the insects olfactory system, which is analogous to the mammalian olfactory system. We simulated a reduced generic model of olfactory processing in insects using the phenomenologically adaptive neuron model [28]. The model network consisted of an input layer with 1,480 olfactory sensory neurons (OSNs), which project to the next layer representing the antennal lobe circuit with 24 projection neurons (PNs) and 96 inhibitory local inter-neurons (LNs) that form a local feed-forward inhibitory micro-circuit with the PNs. The third layer holds 1,000 Kenyon cells (KCs) receiving divergent-convergent input from PNs. The relative numbers for all neurons approximate the anatomical ratios found in the olfactory pathway of the honeybee [46] (Figure 4A). We introduced heterogeneity among neurons by randomizing their synaptic time constants and the connectivity probabilities are chosen according to anatomical studies. Synaptic weights were adjusted to achieve spontaneous firing statistics that match the observed physiological regimes. The SFA parameters were identical throughout the network with ms (see Materials and Methods for details). Experimentally, the cellular mechanisms for SFA exist for neurons at all three network layers [47]–[53]. Notably, strong SFA mediating currents have been identified in the KCs of Periplaneta americana [52].
Using this model, we sought to understand how adaptation contributes to temporally sparse odor representations in the KC layer in a small sized network and under highly fluctuating input conditions. We simulated the input to each OSN by an independent Poisson process, which is thought to be reminiscent of the transduction process at the olfactory receptor level [53]. Stimulus activation was modelled by a step increase in the Poisson intensity with uniformly jittered onset across the OSN population (Materials and Methods). Following a transient onset response the OSNs adapted their firing to a new steady-state (Figure 4B,C). The pronounced effect of adaptation becomes apparent when the adaptive population response is compared to the OSN responses in the control network without any adaptation (τs = 0; dashed line, Figure 4C). In the next layer, the PN population activity is reflected in a dominant phasic-tonic response profile (green line, Figure 4D), which closely matches the experimental observation [54]. This is due to the self-inhibitory effect of the SFA mechanism, and to the feedforward inhibition received from the LNs (magenta line, Figure 4D). Consequently, the KCs in the third layer produced only very few action potentials following the response onset with an almost silent background activity (red line, Figure 4E,G). The average number of emitted KC response spikes per neuron, , is small in the adaptive network (average ) whereas KCs continue spiking throughout stimulus presentation in the non-adaptive network (Figure 4G). This finding closely resembles experimental findings of temporal sparseness of KC responses in different insect species [31]–[33] and quantitatively matches the KC response statistics provided by Ito and colleagues [34]. The simulation results obtained here confirm the mathematically derived results in the first results section (Figure 1B) and show that neuronal adaption can cause a temporal sparse representation even in a fairly small and highly structured layered network where the mathematical assumptions of infinite network size and fundamentally incoherent activity are not fulfilled (Materials and Methods). We further investigated the effect of adaptation on the fraction of responding neurons by counting the number of KCs that emit spikes during stimulation. In the adaptive circuit and in presence of local inhibition only 9% of KCs produce responses (23% in the adaptive network when inhibition is turned off). In contrast, in the non-adaptive network with local inhibition 60% of KCs responded. The low fraction of responding neurons in the adaptive network quantitatively match the experimental findings in the moth [34] and the fruit fly [55].
To test the effect of inhibition in the LN-PN micro-circuitry within the antennal lobe layer on the emergence of temporal sparseness in the KC layer, we deactivated all LN-PN feedforward connections and kept all other parameters fix. We found a profound increase in the amplitude of the KC population response, both in the adaptive (red line, Figure 4F) and the non-adaptive network (dashed red line, Figure 4F). This increase in response amplitude is carried by an increase in the number of responding KCs due to the increased excitatory input from the PNs, implying a strong reduction in the KC population sparseness. Importantly, removing local inhibition did not alter the temporal profile of the KC population response in the adaptive network (cf. red lines in Figure 4E,F), and thus temporal sparseness was independent of inhibition in our network model.
How reliable is the sparse spike response across trials in a single KC? To answer this question, we again measured the robustness of the stimulus representation by estimating the Fano factor across simulation trials (Figure 4H). The network with adaptive neurons and inhibitory micro-circuitry exhibited a low Fano factor (median ) and a narrow distribution across all neurons. This follows the experimental finding that the few spikes emitted by KCs are highly reliably [34] (network 1, Figure 4H). Turning off the inhibitory micro-circuitry did not significantly change the response reliability (Wilcoxon rank sum test, p-value = 0.01; network 2, Figure 4H). However, both networks that lacked adaptation exhibited a significantly higher variability with a median Fano factor close to one (Wilcoxon rank sum test, p-value = 0.01; networks 3 and 4, Figure 4H), independent of the presence or absence of inhibition micro-circuits.
To explore whether neuronal adaptation could contribute to temporal sparseness in the biological network, we performed a set of Calcium imaging experiments, monitoring Calcium responses in the KC population of the honeybee mushroom body [33] (Materials and Methods). Our computational model (Figure 4) predicted that blocking of the inhibitory microcircuit would increase the population response amplitude but should not alter the temporal dynamics of the KC population response which is independent of the stimulus duration. In a set of experiments, we tested this hypothesis by comparison of the KCs' evoked activity in the presence and absence of GABAergic inhibition (Materials and Methods). First, we analyzed the normalized Calcium response signal within the mushroom body lip region in response to a 3 s, 2 s, 1 s and 0.5 s odor stimulus (Figure 5A). We observed the same brief phasic response following stimulus onset in all four cases with a characteristic slope of Calcium response decay that has been reported previously to account for a temporally sparse spiking response [33]. These responses, unlike those of PNs in the previous processing stage of the insect olfactory system [56], [57], are independent of the stimulation duration (Figure 5A). Bath application of the GABAA antagonist picrotoxin (PTX) did not change the time course of the Calcium response dynamics (Figure 5B–D). The effectiveness of the drug was verified by the increased population response amplitude in initial phase (Figure 5C). Next, we tested the GABAB antagonist hydrochloride (CGP) using the same protocol and again found an increase in the response magnitude but no alteration of the response dynamics (Figure 5E,F). This suggests the absence of inhibition does not change the temporal scale of KCs responses in line with the model prediction.
We propose that a simple neuron-intrinsic mechanism of spike-triggered adaptation can account for a reliable and temporally sparse sensory stimulus representation across stages of sensory processing. The emergence of a sparse representation has been demonstrated in various sensory areas, for example in visual [58], auditory [59], somatosensory [60], and olfactory [61] cortices, and thus manifests a principle of sensory computation across sensory modalities and independent of the natural stimulus kinetic. Our results show that adaptation allows to reliably represent a stimulus with a temporally restricted response to stimulus onset and thus more stimuli can be represented in time which is the basis for a temporally sparse representations of a dynamically changing stimulus environment.
At the single neuron level, SFA is known to induce the functional property of a fractional differentiation with respect to the temporal profile of the input and thus offers the possibility of tuning the neuron's response properties to the relevant stimulus time-scales at the cellular level [2]–[4], [62]–[65]. Our results indicate further that sensory processing in a feedforeward network with adaptive neurons focuses on the temporal changes of the sensory input in a precise and temporally sparse manner (Figure 1B; Figure 2E and Figure 4E) and at the same time the constancy of the stimulus is memorized in the cellular level of adaptation [37] (Figure 1D). The constancy of the environment is an important factor of state-dependent computations [66] that evidently should be tracked by the network. Such context-dependent modulations set the background and have been observed in different sensory systems where responses are strongly influenced by efferent contextual input [67]–[69]. In this paper, we show that information about the context of a given stimuli maybe stored in the adaptation level across processing stages while at the same time the network remains sensitive to changes. Thus, sequential adaptive populations adjust the circuit transfer function in a self-organizing manner to avoid response attenuation to secondary stimuli. These results add a further possibility of network level interactions to the previous suggestions that SFA optimizes the context depended responses and resolves ambiguity in the neuronal code [8], [70] at the single neuron level. This allows a sensory system to detect extremely small changes in stimulus over a large background by means of an adaptive response without contextual information loss [39]. One prominent example is primate vision where, in the absence of the self-generated dynamics of retina input due to microsaccades, observers become functionally blind to stationary objects during fixations [71].
A sparse temporal representation of stimulus permits very few spikes to transmit high quantities of information about a behaviorally significant stimulus [72]. However, it has been repeatedly questioned whether a few informative spikes can survive in the cortical network, which is highly sensitive to small perturbations [16], [73]. Our results show that a biologically realistic cellular mechanism implemented at successive network stages can transform a dense and highly variable Poisson input at the periphery into a temporally sparse and highly reliable ensemble representation in the cortical network. Therefore, it facilitates a transition from a rate code to a temporal code as required for the concerted spiking of cortical cell assemblies [74] (Figure 2D). These results reflect previous theoretical evidence that SFA has an extensive synchronizing-desynchronizing effect on population responses in a feedback coupled network [75], [76].
A balance between excitation and inhibition leads to strong temporal fluctuations and produce spike trains with high variability in cortex [16], [23]–[25], [77]. However, the adaption level adjusts with a dynamics that is slow compared to the dynamics of excitatory and inhibitory synaptic inputs. This circumstance allows for a transient mismatch of the balanced state in the cortical network and thus leads to a transient reduction of the self-generated (recurrent) noise (Figure 2I). This, in turn, explains why the temporally sparse representation can be highly reliable, following the experimental observations [21], [22]. Moreover, a recent and highly relevant in vivo data set hints toward our theoretical prediction, where adaptation may alter the balance between excitation and inhibition and increase the sensitivity of cortical neurons to sensory stimulation [78]. Here, our main result exploited the transient role of adaptation mechanisms on the cortical variability suppression, after which the variability recovers to the unstimulated values, even though the network remains stimulated (Figure 2I). One can achieve a longer time scale of variability suppression by an increase in the effect of the afferent strength (as a network mechanism), due to the reduction of the input irregularity in the evoked state. This proposal can be supported by the experimental evidence that thalamic inputs strongly drive neurons in cortex [79] and fits the previous theoretical suggestion by [45]. Noteworthy, in our model the irregularity of inter-spike intervals, measured by , in the balanced network does not change significantly in different conditions, which matches the experimentally reported evidence [80]. The recent theoretical studies [26], [27] show that the slow time scales variability suppression can be also achieved within a clustered topology in the balanced network [26] or likewise in an attractor-based networks of cortical dynamics [27]. In these approaches, the reduced variability can be attributed to an increased regularity of the spike trains. This hints that further research to understand the role of interactions between the network and cellular mechanisms in the cortical variability and other network statistics are certainly needed. Additionally, the link between the temporal sparseness achieved here by cascaded network of adaptive neurons with spatial sparseness of responses [81], [82] requires more elaborated research.
The insect olfactory system is experimentally well investigated and exemplifies a pronounced sparse temporal coding scheme at the level of the mushroom body KCs. The olfactory system is analog in invertebrates and vertebrates and the sparse stimulus representation is likewise observed in the pyramidal cells of the piriform cortex [61], and the rapid responses in the mitral cells in the olfactory bulb [83] compare to those of projection neurons in antennal lobe [54], [84]. Our adaptive network model, designed in coarse analogy to the insect olfactory system, produced increasingly phasic population responses as the stimulus-driven activity propagated through the network. Our model results closely match the repeated experimental observation of temporally sparse and reliable KC responses in extracellular recordings from the locust [31], fruit fly [85] and manduca [32], [34], and in Calcium imaging in the honeybee [33]. Although Calcium responses are slow, it has been suggested that they closely correspond to the population activity dynamics [86]. In our experiments we could show that systemic blocking of GABAergic transmission did not affect the temporal sparseness of the KC population response in the honeybee (Figure 5) signified by the transient Calcium response [33]. Therefore, the stable temporal activity in the mushroom body qualitatively matches with our theoretical predication of population rate dynamics (Figure 1). This result might seem to contradict former studies that stressed the role of inhibitory feed-forward [87] or feedback inhibition [88], [89] for the emergence of KC sparseness. However, the suggested inhibitory mechanisms and the sequential effect in the adaptive network proposed here are not mutually exclusive and may act in concert to establish and maintain a temporal and spatial sparse code in a rich and dynamic natural olfactory scene. In this paper, we deliberately focus on the temporal aspect of the responses, since it seems that spatial sparseness is mediated by connectivity schemes [90]–[92].
The adaptive network model manifests a low trial-to-trial variability of the sparse KC responses that typically consist of only 1–2 spikes. In consequence, a sparsely activated KC ensemble is able to robustly encode stimulus information. The low variability at the single cell level (Figure 4H) carries over to a low variability of the population response [29], [30]. This benefits downstream processing in the mushroom body output neurons that integrate converging input from many KCs [46], and which were shown to reliably encode odor-reward associations in the honeybee [93].
Next to the cellular mechanism of adaptation studied here, short-term synaptic plasticity may produce similar effects. The activity-dependent nature of short term depression (STD) produces correlated presynaptic input spike trains [94]. Hence, it facilitates weak signal detection [94] similar to adaptation [95]. Moreover, STD can also generate a sharp transient in the stimulus response [96], [97] that can propagate to higher layers of the network. Therefore it is plausible to utilize short-term synaptic plasticity to achieve similar results to the ones obtained here with SFA. However, STD may have some drawbacks in comparison to adaptation, namely a low signal-to-noise ratio, and a low-pass filtering of input that is more sensitive to high frequency synaptic noise [62], [68]. Evidently, STD takes effect at the single synapse while SFA acts on a neuron's output. The combination of both mechanisms that are encountered side-by-side in cortical circuits [99], [100] may provide a powerful means for efficient coding [98].
Our results here are of general importance for sensory coding theories. A mechanism of self-inhibition at the cellular level can facilitate a temporally sparse ensemble code but does not require well adjusted interplay between excitatory and inhibitory circuitry at the network level. This network effect is robust due to the distributed nature of the underlying mechanism, which acts independently in each single neuron. The regularizing effect of self-inhibition increases the signal-to-noise ratio not only of single neuron responses but also of the neuronal population activity [29], [30], [37] that is post-synaptically integrated in downstream neurons.
To address analytically the sequential effect of adaptation in a feedforward network, we consider a model in which populations are described by their firing rates. Although firing rate models typically provide a fairly accurate description of network behavior when the neurons are firing asynchronously [101], they do not capture all features of realistic networks. Therefore, we verify all of our predictions with a population density formalism [29] as well as a large-scale simulation of realistic spiking neurons. To determine the mean activity dynamics of a consecutive populations, we employed an standard mean firing rate model of population as(1)where is the transfer input-output function, is the adaptation time-scale, is the coupling factor between two populations and is the adaptive negative feedback for the population with strength and is the standard deviation of the input. In our rate model analysis, we use the transfer function of the leaky-integrate and fire neuron that can be written as(2)where and , , and are membrane capacitance, membrane time-constant, spiking threshold and reset potential, respectively. Here, we assume is the injected current to the population independent of the stimulus and constant over time. Given , the equilibrium can be determined by(3)The condition for the stability reads and [35] (Figure 6A). It is important to note that whenever the conditions for stability are satisfied, the fix point is reached via a focus attractor (Figure 6 C,D) since the Jacobin of this system (under the physiological condition of ) always has a complex eigenvalue with a negative real part. It is also known that the is a linear function with respect to its input, given a sufficiently slow or strong adaptation and a non-linear shape of [35], [36], [102]. It can also be shown that whenever the adaptation is ineffective () we have(4)where and . This derivative scales with (Figure 6B). Now, we can plug back the adaptation into the steady state solution, which has a magnitude of . In Figure 6 B, we numerically determine the condition for , that it reads , given the parameters , , , and as they are stated in the caption. An increase in the population rate leads to a reduced increase in the next population, and therefore the adapted level of responses satisfy . For realistic adaptation values this mapping closely follows the result of a previous study where it was shown that the effect of increasing the cells input conductance on its f-I curve is mainly subtractive [102]. Note that for very weak adaptation the steady-state is not affected considerably.
The magnitude of the transient response firing rate for an adaptive population lies between the adapted steady-state rate and the response rate without adaptation. Given the level of new input it can be calculated analytically [2]. Hence, the slow dynamics of adaptation and fast response f-I curve reflect two states of operation where the onset response very closely follows the properties of the non-adapted response curve and the adapted steady state produces a subtractive input-output relationship. Noteworthy, the assumption that all populations have a same background firing rate is not a necessary condition. One can achieve the same result by using heterogeneous couplings or stimulus independent private input that may induce more realistic variations in background rates as observed in different stages of sensory processing. The crucial point of the inherited dynamics due to adaption is the fundamental non-linearity that (1) the transient response amplitude is hardly affected by adaptation, (2) the adapted steady-state is fall apart from it and can become subtractive.
In Muller et al. [28] it is shown that by an adiabatic elimination of fast variables a detailed neuron model including voltage dynamics, conductance-based synapses, and spike-induced adaptation, in the incoherent state (Asynchronous and Irregular state) reduces to a stochastic point process. Thus, we define an orderly point process with a hazard function argument with state variable as(5)where is the number of events in . We assume the dynamic of the adaptation variable is(6)where is the time of th spike in the ensemble. Thus, the state variable distribution at time in the ensemble is governed by a master equation of the form(7)We solve Eq. 7 with the help of the transformations and numerically [28]. The master equation here belongs to a non-renewal process [28], [29] and its renewal correspondence can be seen in [28]. It turns out that indeed is the input-out transfer function of neurons in the network where its instantaneous parameters are give by the input statistics [28]. For instance, the transfer function of a conductance based leaky integrate and fire neuron can be written as(8)where, and are refectory period and an input-dependent effective time constants, respectively. The and appearing in are the average and variance of the free (i.e., spike-less) membrane voltage distribution [35] and is the membrane capacitance. Here, we used the mean-field formalism developed by [23] and [103] to approximately determine the averaged input within a standard balanced network, as the parameters of the hazard function suggested by [35], which uses the calculated average firing rate of inhibitory and excitatory neurons in the randomly connected network. The analytical results in this paper assumed the standard as a from of the hazard function and the value of is estimated form simulations of the detailed neuron model with an step like input increase. Here, it is important to note that conductance-based model approximately follows the current based neuron model with a colored noise, where is shorter than the membrane time constant of the neuron, which now depends on the total conductance [103], [104]. Therefore, given the estimate, we can approximate the numerical solution of the master equation (eq. 7) by applying the exponential Euler method for the death term, and reinserting the lost probability as it is fully described in [28]. The functional form of the solution in a compact form can be written as(9)where is the initial condition state of the system and is a constant defined by [28], [29]. Similarly, one can derive the distribution of just after the event, [28]. Then, the relationship between and the ordinary ISI distribution can be written as(10)where . Now the moment of the distribution and its coefficient of variation can be numerically determined. Note that the framework here is closely connected to the spike response model, also known as the generalized linear neuron model [30]. Alternatively, the same ISI distribution can be also derived form the discretization of the master equation as it is demonstrated in [37]. It can be shown that the firing rate and the consistency equation of the ensemble is(11)Now to calculate the counting statistics, we applied the techniques are introduced by Farkhooi et al [29], and defined a joint probability density as(12)where an event occurs at time and the state variable is . We can write event time and state of adaptation joint density recursively,(13)To simplify the integral equations, we use Bra-Ket notation following a suggestion by [105],
defined as(14)and(15)Thereafter, we derive the Laplace transform () of the joint density in the eq. 13 by(16)where . Next, we define the operator ,(17)Now, by employing as in [29], we derive(18)where is the Laplace transform of the probability density of observing events in a given time window. Now we derive(19)where is the identity operator. This equation represents the Laplace transform of the auto-correlation function. Using auto-correlation function , we can calculate the Fano factor that provides an index for the quantification of the count variability. It is defined as , where and are the variance and the mean of the number of events in a certain time window . It follows from the additive property of the expectation that and in the case of a constant firing rate, it is simply . To calculate the second moment of , we require in eq. 19. Thus, the Fano factor is and the inverse Laplace transform is(20)where . In [29], we demonstrate in detail that the asymptotic property of at equilibrium can be derived as(21)where is the linear correlation coefficient between two lagged intervals. Provided the limit exits, we find the familiar relationship of in the steady-state.
Our model neuron is a general conductance-based integrate-and-fire neuron with spike-frequency adaptation as it is proposed in Muller et al. [28]. The model phenomenologically captures a wide array of biophysical spike-frequency mechanisms such as M-type current, afterhyperpolarization (AHP-current) and even slow recovery from inactivation of the fast sodium current [28]. The model neuron is also known to perform high-pass filtering of the input frequencies following the universal model of adaptation [2]. Neuron parameters used follow the Table 3 in Muller et al. [28]. The conductance model used for the static synapses between the neurons is alpha-shaped with gamma distributed time constants from and for excitatory and inhibitory synapses, respectively. All simulations were performed using the NEST simulator [106] version 2.0beta and the Pynest interface.
The network connectivity is straight forward: each PN and LN receives excitatory connections from 20% randomly chosen OSNs [57], [107]. Additionally, every PN receives input from 50% randomly chosen inhibitory LNs [57], [107]. In our model the PNs do not excite one another and each PN output diverges to 50% randomly chosen KCs [33], [87], [90].
We tuned the simulated network by adjusting the synaptic weights to achieve the same spontaneous firing rate as reported experimentally: OSNs 15–25 Hz [53], LNs 4–10 Hz [107], PNs 3–10 [54] and KCs 0.3–1.0 Hz [34].
Experiments were performed following the methods published in Szyszka et al [33]. In summary, foraging honeybees (Apis mellifera) were caught at the entrance of the hive, immobilized by chilling on ice, and fixed in a plexiglas chamber before the head capsule was opened for dye injection. We retrogradely stained clawed Kenyon cells (KC) of the median calyx, using the calcium sensor FURA-2 dextran (Molecular Probes, Eugene, USA) with a dye loaded glass electrode, which was pricked into KC axons projecting to the ventral median part of the -lobe [33]. After dye injection the head capsule was closed, bees feed and kept in a dark humid chamber for several hours.
The processing of imaging data was performed with custom written routines in IDL (RSI, Boulder, CO, USA). In summary, changes in the calcium concentration were measured as absolute changes of fluorescence: a ratio was calculated from the light intensities measured at 340 nm and 380 nm illumination and the background fluorescence before odor onset was subtracted leading to F with F = F340/F380. Odor stimulation was preformed under a 20× objective of the microscope, the naturally occurring plant odor octanol (Sigma Aldrich, Germany), diluted 1:100 in paraffine oil (FLUKA, Buchs, Switzerland), was delivered to both antennae of the bee using a computer controlled, custom made olfactometer. To this, odor loaded air was injected into a permanent airstream resulting in a further 1:10 dilution. Stimulus duration was 3 seconds if not mentioned otherwise. The air was permanently exhausted.
For GABA blockage, a solution of 150 µl GABA receptor antagonist dissolved in ringer for final concentration (10−5M picrotoxin (PTX, Sigma Aldrich, Germany) or 5×10−4M CGP54626 (CGP, Tocris Bioscience, USA)) was bath applied to the brain after pre-treatment measurements. Measurements started 10 min after drug application. The calcium signals are analyzed in Matlab (The MathWorks Inc., Natick, USA). The normalization of the responses are performed per animal and the plotted traces are the averaged values across subjects.
|
10.1371/journal.ppat.0030009 | Neutralizing Antibody Fails to Impact the Course of Ebola Virus Infection in Monkeys | Prophylaxis with high doses of neutralizing antibody typically offers protection against challenge with viruses producing acute infections. In this study, we have investigated the ability of the neutralizing human monoclonal antibody, KZ52, to protect against Ebola virus in rhesus macaques. This antibody was previously shown to fully protect guinea pigs from infection. Four rhesus macaques were given 50 mg/kg of neutralizing human monoclonal antibody KZ52 intravenously 1 d before challenge with 1,000 plaque-forming units of Ebola virus, followed by a second dose of 50 mg/kg antibody 4 d after challenge. A control animal was exposed to virus in the absence of antibody treatment. Passive transfer of the neutralizing human monoclonal antibody not only failed to protect macaques against challenge with Ebola virus but also had a minimal effect on the explosive viral replication following infection. We show that the inability of antibody to impact infection was not due to neutralization escape. It appears that Ebola virus has a mechanism of infection propagation in vivo in macaques that is uniquely insensitive even to high concentrations of neutralizing antibody.
| Ebola virus is one of the most feared of human pathogens with a mortality that can approach 90% and an extremely rapid disease course that can lead to death within days of infection. Antibodies able to inhibit viral infection in culture, neutralizing antibodies, can typically prevent viral infection in animals and humans when present prior to infection, at sufficient concentration. Such neutralizing antibodies may be provided through passive administration or induced by vaccination. We have previously shown that a human neutralizing antibody can protect guinea pigs against Ebola virus. However, here we show that this antibody does not protect monkeys against Ebola virus and surprisingly appears to have very little impact upon the rapid course of infection, despite being present at very high levels in the blood of the monkeys. We conclude that administering antibody prior to or immediately following exposure to Ebola virus, for example, after an accident in a research setting or a bioterrorist attack, is unlikely to be effective in preventing disease. Recent successes in protecting monkeys against Ebola virus through vaccination may be independent of antibody, or, more likely, critically dependent on the cooperation of antibody and cellular immunity.
| Editor's note: The potential efficacy of pre- and post-exposure prophylaxis against Ebola virus infection, as well as the fundamentally important question of whether neutralizing bodies are important for Ebola virus resistance, is addressed by a related manuscript in this issue of PLoS Pathogens. Please see doi:10.1371/journal.ppat.0030002 by Feldmann et al.
Passive transfer of relatively high concentrations of neutralizing antibodies can protect against challenge with a range of viruses in animal models and in humans [1–3]. Protection in some cases is in the form of sterilizing immunity, i.e., no viral replication is observed following challenge [2,4,5]. In other cases (e.g., [5,6]), some replication is observed but protection from disease is achieved, presumably because neutralizing antibody sufficiently blunts infection for T cell and innate immunity to resolve infection [7]. It might be expected that passive neutralizing antibody would be most effective against challenge with acute viruses. Many acute viral infections are resolved even in the absence of neutralizing antibody, and the blunting effect of passive antibody would provide more time for the development of effective cellular immune responses. In contrast, chronic viruses may present a greater challenge to passive antibody, since, in the absence of sterilizing immunity, there is a window of opportunity for the virus to establish a chronic infection before cellular immunity can be mobilized.
Ebola virus (EBOV) causes a severe acute infection in humans [8]. Infection with the Ebola Zaire strain, Zaire ebolavirus (ZEBOV), produces mortality in the range of 60%–90% [9] with death generally occurring around 7–11 d following the appearance of symptoms [8]. There is a single report describing the use of convalescent sera to treat EBOV infection [10]. However, the patients in this report may have already been through the worst stages of the disease, and it is not clear that serum antibodies were responsible for their recovery [10]. Further, neutralizing antibody titers in survivors of EBOV infection tend to be rather low, although we have isolated a neutralizing human monoclonal antibody (mAb), KZ52, of good potency from a convalescent individual [11].
The ability of passive antibody to protect against EBOV has been investigated in a number of animal models. The guinea pig and mouse models use EBOVs that have been serially passaged to adapt to replication in the respective animals and are highly lethal. Protection has been demonstrated in the guinea pig model using neutralizing horse, sheep, and goat immunoglobulin G (IgG) against EBOV [12,13] and the human anti-EBOV GP mAb, IgG KZ52. This antibody neutralizes ZEBOV (1995, Kikwit) with a 50% inhibitory concentration (IC50) of 0.05–0.3 μg/ml and an IC90 of 0.5–2.6 μg/ml in Vero cells [11,14] and an IC50 of approximately 0.05–1 μg/ml and a IC90 of 0.5–2 μg/ml in primary human monocytes/macrophages [14]. We showed that when administered subcutaneously at a dosage of 25 mg/kg up to 1 h after challenge, the antibody protects against robust ZEBOV challenge (10,000 plaque-forming units [pfu]) in the guinea pig model [6].
Macaques provide a model of EBOV infection that is likely closer to human infection. The human virus can be used directly in macaques without need for adaptation and the course of disease mirrors that seen in humans [8]. In cynomolgus macaques (Macaca fascicularis), ZEBOV infection produces a mortality rate of 100% with death occurring 6–8 d following infection with 1,000 pfu [15], while in rhesus macaques (Macaca mulatta) ZEBOV produces about 100% mortality with death occurring 7–10 d after infection with 1,000 pfu [16]. In contrast to the guinea pig experiments, the passively transferred polyclonal equine neutralizing IgG described above provided only some minor benefit in the form of a slight delay in the onset of viremia from day 5 to day 7 [13] following ZEBOV challenge of cynomolgus monkeys. No significant reduction in mortality was observed. However, protection against EBOV in primates has been observed in a low dose challenge model. Thus, neutralizing equine IgG protected baboons from <30 LD50 (50% lethal dose) ZEBOV challenge when the IgG was given up to 1 h after infection and the serum contained high neutralizing antibody titers (1:128 to 1:512) [17,18], and, similarly, neutralizing ovine serum protected baboons against 0.6 LD50 ZEBOV challenge [19].
Here, we studied the ability of passively transferred neutralizing human mAb KZ52 to protect against ZEBOV challenge in rhesus macaques. This passive transfer failed to protect the macaques against challenge with ZEBOV, and, furthermore, had a minimal effect on the explosive viral replication following infection. We showed by ELISA that antibody was present at high levels in serum of the monkeys and that neutralization escape was not responsible for the resistance of virus to antibody prophylaxis.
To evaluate whether IgGl KZ52 could protect against Ebola virus infection in a nonhuman primate animal model, antibody was passively transferred to rhesus macaques followed by challenge with the 1995 ZEBOV (Kikwit) isolate 24 h later. Protection against virus challenge by neutralizing antibodies in naive animals often requires high doses of antibody [2]. Therefore, we used a high dose of 50 mg/kg KZ52, which was close to the maximum practically achievable. In addition, we gave a second bolus of 50 mg/kg of KZ52 on day 4 following infection. The results for the four antibody-treated animals show a steady increase in plasma viremia up to 105–107 pfu/ml on day 7 (Figure 1). These levels of plasma viremia closely parallel those seen in the control animal and typically seen in historical controls [15]. The second bolus of antibody given on day 4 did not appear to have any impact upon the rate of increase of plasma virus (Figure 1). Three of the treated animals were euthanized when moribund at day 9 or 10 post infection. The fourth treated animal showed a decrease in plasma viral load after the peak and survived to day 28 before becoming moribund when it too was euthanized. Although monkey CH46 had less severe symptoms than the other animals in the study, it was concluded that this animal, too, was suffering from disease due to ZEBOV, as evidenced, for example, by copious Ebola virus antigen in the lungs (see below).
Serum antibody (KZ52) loads were measured 1 d before virus challenge (day −1) and 4 d after challenge but before antibody boosting (day 4) by an ELISA designed to detect KZ52 as a human antibody that has bound to immobilized ZEBOV glycoprotein. The serum KZ52 antibody levels on day 4 were in the approximate range 200–400 μg/ml (Table 1). The two control monkeys, EHD (untreated and challenged), and a negative monkey (neither treated nor challenged), had very similar background levels of reactivity to the glycoprotein and anti-human antibody as each of the treated monkeys before treatment. A 50-mg/kg dose typically produces serum mAb concentrations in animals on the order of 500 μg/ml after injection [6]. Since the neutralization titer of KZ52 (IC90) for ZEBOV is on the order of 0.5–2.5 μg/ml, depending upon the target cell and the presence of complement, the concentrations of KZ52 in the animals at the time of challenge and for the first few days were, as expected, greater than, or on the order of 100 × IC90. These concentrations typically provide sterilizing immunity against challenge by a number of viruses [4,20].
One formal possibility is that the neutralizing antibody has little effect on the course of infection in the treated monkeys because of the rapid emergence of neutralization escape mutants. Accordingly, virus was isolated from a selection of plasma from day 4 (monkeys CH56, CH57, and CH83) and day 7 (monkeys CH56 and CH57). All of these viruses were sensitive to KZ52 so that essentially 100% neutralization was observed in vitro at 40 and 400 μg/ml KZ52 in a plaque assay (see Materials and Methods) using Vero E6 target cells.
In order to gain a better understanding of any differences in pathology between the control and antibody-treated animals, the levels of virus in different organs were surveyed postmortem. Viral levels in the liver, spleen, kidney, adrenal glands, lung, and mesenteric and inguinal lymph nodes were high (104–106 pfu/g) in the control and three of the four treated animals (Table 2). However, monkey CH46, who survived much longer than the other animals (to day 28) showed some major histopathological differences from the other infected monkeys and from the norm for ZEBOV infection [15]. Relatively low viral levels were observed in most of the organs of CH46, and none in the liver, spleen, and adrenal glands. In addition, large immunoreactive monocytes were found in the blood of monkeys CH56, CH57, and CH83, but not in CH46 (Figure 2). Typically, smaller immunoreactive monocytes are seen with ZEBOV infection [15]. The presence of large immunoreactive monocytes may simply reflect uptake of antibody-coated virions via Fc receptors and subsequent viral clearance. However, it is interesting to note that previous studies have implicated mononuclear phagocytes as vehicles for transport of filovirus particles to specific organs such as liver and spleen [21–24]. If virus particles could remain infectious following Fc receptor-mediated uptake in a subset of cells (compare DC-SIGN mediated uptake of HIV-1 by dendritic cells [25]), then the course of disease in monkeys CH56, CH57, and CH83 might represent the net result of inhibition by neutralization and enhancement by antibody-mediated cellular uptake. Interestingly, monkey CH46, who fared somewhat better than the others and lacked virus in the liver and spleen, did not show the presence of large immunoreactive monocytes. This is suggestive of lowered Fc receptor-mediated uptake of virus or reduced activity of the mononuclear phagocytic system.
Here, we describe the case of a potent neutralizing human monoclonal antibody, administered to give a high serum concentration, which is shown to be unable to protect macaques against challenge with a lethal dose of ZEBOV. The antibody appeared to have very little effect on the course of virus replication or disease in three of four treated animals. Neutralization escape does not appear to explain the lack of protection observed by the antibody. In one animal, a more limited infection was observed, but this macaque did also eventually succumb to the effects of viral disease.
The challenge dose of 1,000 pfu used corresponds to the amount of EBOV contained in a relatively small quantity of fluid (on the order of 1 μl) from an infected individual given the high titers of virus typically found in such individuals (on the order of 106 pfu/ml of blood, for example). Therefore, the challenge dose was not unreasonable in terms of a natural exposure to virus.
The negative results with passive antibody contrast strongly with recent successes in preventing EBOV infection in macaques through vaccination [26,27]. Does this mean that neutralizing antibody is unimportant in vaccine protection? The answer to this question must await further studies. However, a plausible hypothesis to explain all the data would still allow for an important contribution of antibody to vaccine protection. This hypothesis would argue that passive antibody is unable to completely block all EBOV entry to cells, and once a few cells are infected, virus replication is so explosive that it cannot be contained by a de novo generated cellular immune response. Vaccination, on the other hand, will provide CD8+ memory T cells that can be rapidly recruited to become effector cells and limit infection. Certainly, although mAb KZ52 was able to provide protection from disease following ZEBOV challenge in the guinea pig model, immunity was not sterilizing and some viral replication was noted [6]. Since 1 pfu of EBOV is a lethal dose for primates [8], a failure of passive antibody to achieve sterilizing immunity may be critical. Immunohistochemistry, as discussed above, gives some intriguing hints that uptake of antibody-coated virions by monocytes may possibly have a role to play in the course of infection following antibody treatment. We note, however, that previous in vitro studies using isolated human monocytes/macrophages did not find evidence of infectivity-enhancing antibodies [14]. More detailed in vitro and in vivo investigations will be required before any firm conclusions can be drawn. We also note that our experiments were carried out with a single neutralizing monoclonal antibody. It is possible that a more favorable outcome may have been apparent for a combination of neutralizing antibodies or even a combination of neutralizing and nonneutralizing antibodies [2]. However, these possibilities should be weighed against the very high concentrations of neutralizing monoclonal antibody used in the experiments and the efficacy of the antibody in the guinea pig model.
In summary, the inability of high concentrations of neutralizing antibody to even slow viral replication in infected macaques is remarkable and implies a mechanism of infection propagation that is virtually insensitive to antibody. Overall, the results suggest that monoclonal antibody prophylaxis or post-exposure prophylaxis alone are unlikely to be effective strategies in protecting against EBOV, for example, following a needle-stick accident in a research setting or a bioterrorist attack.
50 mg/kg of KZ52 IgG1 human antibody [11] was given intravenously to rhesus macaques (weight, 3.9–4.4 kg) 1 d before challenge (day 0) with 1,000 pfu intramuscularly of the 1995 ZEBOV (Kikwit) isolate and again 4 d later (day +4). One monkey was not given any antibody treatment. The animals were carefully monitored for signs of disease, and Ebola virus plasma viremia was determined at days 4, 7, 9, and 10 in serum by plaque assay as described below.
The investigators adhered to the Guide for the Care and Use of Laboratory Animals when conducting research, using animals [28]. The United States Army Medical Research Institute of Infectious Diseases (USAMRIID) animal facilities and animal care and use program are accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International. All infectious material and animals were handled in a maximum-containment biosafety level 4 facility at USAMRIID under standard operating conditions.
IgGl KZ52 was produced and purified as described by Parren et al. [29] and was >98% pure, as determined by sodium dodecyl sulfate-polyacrylamide gel electrophoresis and contained <1 IU of endotoxin/ml, as determined in a quantitative chromagenic Limulus amoebecyte lysate assay (BioWhittaker, Cambrex, http://www.cambrex.com).
Plasma viremia and viral loads in organs was determined by virus titration in a conventional plaque assay on Vero E6 cells, as described elsewhere [13,21].
Samples were diluted into Eagle's minimal essential medium (EMEM; Invitrogen, http://www.invitrogen.com) with 5% heat-inactivated fetal bovine serum (FBS). In the presence and absence of a constant dilution of 1:10 or 1:100 of 4 mg/ml KZ52 (thus, 0.4 mg/ml or 0.04 mg/ml), plasma were titrated from 10−1 to 10−6 dilutions. Viremia was determined by counting pfu on Vero E6 cell monolayers. Cells grown to confluence in 6-well plates were given 0.2 ml of plasma with and without additional KZ52. The titrated samples were incubated in the presence of KZ52 for 1 h at 37 °C in 5% CO2. After absorption, the cells were overlaid with 2 ml of EMEM containing 5% FBS, 25 mM HEPES buffer, 50-μg gentamicin per ml, and 1% agarose. After 10 d, plaques were visible and the cells were removed from the humidified 37 °C incubator to visualize plaques with an inverted phase microscope. 2 ml of neutral red (1:6,000 final concentration; Sigma-Aldrich, http://www.sigmaaldrich.com) was added to each well, and after an additional 24-h incubation, the plaques were counted [11,30].
Nunc-Immuno Maxisorp enzyme-linked immunosorbent assay (ELISA) plates (Nunc, http://nuncbrand.com) were coated with 100 μl/well of 10 μg/ml lectin from Galanthus Nivalis (Sigma-Aldrich) in PBS and incubated overnight at 4 °C. The plates were then blocked in phosphate-buffered saline (PBS) containing 10% FBS for 2 h. The wells were then washed twice with wash buffer, PBS containing 0.2% Tween 20 (Sigma-Aldrich). 293 cells were transfected with a mammalian expression plasmid coding for transmembrane domain-deleted Ebola glycoprotein. Supernatant (100 μl) from these cells, which contains 0.8–1.3 mg/ml total protein, was used to coat each well for 1 h at room temperature (RT) after the blocking solution was removed from the ELISA plates. Plates were then washed six times with wash buffer. Monkey sera were added in 10-fold dilutions from 1:10 to 1:105 in dilution buffer (PBS containing 1% BSA and 0.02% Tween) and incubated at RT for 1 h. The wells were then washed six times, and a secondary antibody alkaline phosphatase-conjugated goat anti-human immunoglobulin G (IgG) against the F(ab′)2 portion of the antibody (Pierce, http://www.piercenet.com) diluted 1:500 was added, and this was incubated for 1 h. Finally, the plates were washed again six times and developed by one tablet of phosphatase substrate (Sigma-Aldrich) in 5 ml of alkaline phosphatase stain buffer (pH 9.8) per plate. The assay was performed as per manufacturer's directions. The plates were read at an optical density of 405 nm on a microplate reader (Molecular Devices, http://www.moleculardevices.com) at 30 min after adding substrate. A panel of normal sera was run each time the assay was performed.
Sections were pretreated with Dako Ready to Use Proteinase K (Dako, http://www.dako.com) for 6 min at RT after deparaffinization and rehydration through a series of graded ethanols. Blocking was performed with Dako's Serum-Free Protein Block for 20 min pre-antibody exposure. The tissue sections were then incubated overnight at 4 °C in primary antibody using an equal mixture of mouse monoclonal antibodies to EBOV GP and VP40 (1:5,000). An alkaline phosphatase-labeled polymer (Dako Envision System, alkaline phosphatase) was incubated on the sections for 30 min, and then color was developed by exposing tissue to 6-bromo-2-hydroxyl-3-naphtholic acid (HistoMark Red; Kikegaard and Perry Laboratories, http://www.kpl.com) substrate for 50 min in the dark. Counterstaining was done with hematoxylin. Positive controls included archived EBOV-infected cynomolgus tissue, and negative controls included replicate sections exposed to anti-Marburg virus antibodies and uninfected cynomolgus macaque tissue [15]. |
10.1371/journal.ppat.1002318 | SAG101 Forms a Ternary Complex with EDS1 and PAD4 and Is Required for Resistance Signaling against Turnip Crinkle Virus | EDS1, PAD4, and SAG101 are common regulators of plant immunity against many pathogens. EDS1 interacts with both PAD4 and SAG101 but direct interaction between PAD4 and SAG101 has not been detected, leading to the suggestion that the EDS1-PAD4 and EDS1-SAG101 complexes are distinct. We show that EDS1, PAD4, and SAG101 are present in a single complex in planta. While this complex is preferentially nuclear localized, it can be redirected to the cytoplasm in the presence of an extranuclear form of EDS1. PAD4 and SAG101 can in turn, regulate the subcellular localization of EDS1. We also show that the Arabidopsis genome encodes two functionally redundant isoforms of EDS1, either of which can form ternary complexes with PAD4 and SAG101. Simultaneous mutations in both EDS1 isoforms are essential to abrogate resistance (R) protein-mediated defense against turnip crinkle virus (TCV) as well as avrRps4 expressing Pseudomonas syringae. Interestingly, unlike its function as a PAD4 substitute in bacterial resistance, SAG101 is required for R-mediated resistance to TCV, thus implicating a role for the ternary complex in this defense response. However, only EDS1 is required for HRT-mediated HR to TCV, while only PAD4 is required for SA-dependent induction of HRT. Together, these results suggest that EDS1, PAD4 and SAG101 also perform independent functions in HRT-mediated resistance.
| Plant immunity to pathogens requires several proteins, including EDS1, PAD4, SAG101, and these are thought to act downstream of resistance protein-mediated signaling. EDS1 interacts with both PAD4 and SAG101 but no interaction has been detected between SAG101 and PAD4. We show that SAG101 interacts with PAD4 via EDS1 and that the SAG101-EDS1-PAD4 ternary complex is present in the nucleus. EDS1, which is present in the cytoplasm and nucleus, is detected preferentially in the nucleus in the presence of SAG101. The presence of PAD4 restores the cytoplasmic localization of EDS1. Conversely, the SAG101-EDS1-PAD4 ternary complex, which is detected primarily in the nucleus, is redirected to cytoplasm in the presence of an extranuclear form of EDS1. These results show that protein localization changes in relation to the subcellular localization and/or relative levels of their interacting partners. We further show that Arabidopsis plants encode two functional isoforms of EDS1. Both isoforms interact with self and each other, as well as form ternary complexes. SAG101, which is thought to serve as a substitute for PAD4, functions independently in defense signaling against turnip crinkle virus. Our results suggest that EDS1, PAD4, SAG101 function independently as well as in a ternary complex to mediate plant defense signaling.
| One of the most studied plant defense mechanisms involves the deployment of resistance (R) proteins, which primarily provides protection against specific races of pathogens carrying corresponding avirulence (Avr) genes (“gene-for-gene” interactions [1]). R gene-mediated or race-specific immunity is induced when a strain-specific Avr protein from the pathogen associates directly/indirectly with a cognate plant R protein [2]–[4]. Induction of R-mediated responses is often accompanied by the formation of a hypersensitive response (HR) at the site of pathogen entry [5]. Although HR is considered one of the first visible manifestations of pathogen-induced host defense, whether it is the cause or consequence of resistance signaling remains unclear. Concurrent with R-mediated response, defense reactions are triggered in both local and distant parts of the plant. These include a local and systemic increase in the endogenous salicylic acid (SA) levels and the upregulation of a large set of defense genes, including those encoding pathogenesis-related (PR) proteins [6]–[7].
The SA signal transduction pathway plays a key role in plant defense signaling [8]. Arabidopsis mutants that are impaired in SA responsiveness, such as npr1 (nonexpressor of PR [9]–[10]), or are defective in pathogen-induced SA accumulation, such as eds1 (enhanced disease susceptibility 1 [11]), eds5 [12], sid2 (isochorishmate synthase [13]) and pad4 (phytoalexin deficient 4 [14]), exhibit enhanced susceptibility to pathogen infection and show impaired PR gene expression. The EDS1, EDS5, PAD4, NPR1 proteins and the SA synthesizing enzyme SID2 participate in both basal and R protein-mediated defense responses [9]–[14]. EDS1 interacts with PAD4 and SAG (senescence associated gene) 101 and the combined activities of EDS1 and PAD4 proteins are required for HR formation and the restriction of pathogen growth [15]–[17]. EDS1 is thought to form two distinct complexes with PAD4 and SAG101 since direct interaction between PAD4 and SAG101 has not been detected. EDS1 and PAD4 are present in the nucleus and cytoplasm, whereas SAG101 preferentially localizes to the nucleus. A recent study suggested that both the nuclear and cytosolic fractions of EDS1 are required to complement eds1-conferred enhanced susceptibility [18]. SAG101 is thought to be functionally redundant with PAD4, because a mutation in SAG101 alone does not confer bacterial susceptibility. However, sag101 pad4 double mutant plants do exhibit significantly enhanced susceptibility in comparison to pad4 single mutants [17], [19]. EDS1, PAD4, and SAG101 are structurally related to lipase/esterase-like proteins although lipase-like biochemical activities have not been demonstrated for EDS1 or PAD4 [11], [16], [17].
EDS1 was thought to participate in the resistance signaling mediated by toll-interleukin-nucleotide binding site-leucine rich repeat (TIR-NBS-LRR) category of R proteins [20]. However, recent results have shown that EDS1 and SA function redundantly in R-mediated signaling and this masks the requirement for EDS1 [21]. Thus, the requirement for EDS1 by R proteins previously thought to be independent of EDS1, became evident only in plants lacking the capacity to synthesize pathogen-responsive SA [21]. This includes RPS2, RPP8, and HRT, which encode coiled coil (CC)-NBS-LRR type R proteins and confer resistance to bacterial, oomycete and viral pathogens, respectively.
The R protein HRT confers resistance to turnip crinkle virus (TCV) and requires EDS1 and SA for resistance signaling; mutations in either EDS1 or SA synthesizing enzyme SID2 are sufficient to compromise resistance to TCV [22]. However, EDS1 and SA fulfill redundant functions in HR mediated by HRT; HR to TCV is only compromised in plants lacking EDS1 as well as SA [21]. Besides EDS1 and SA, HRT-mediated resistance also requires PAD4 and EDS5, a recessive locus rrt (regulates resistance to TCV), and the blue-light photoreceptors [22]-[24]. Although SA appears to function downstream of the HRT-derived recognition of TCV, it cannot confer resistance in the absence of HRT [22], [23], [25], [26]. Exogenous SA confers resistance in HRT background by upregulating expression of HRT [22]–[25]. Interestingly, the requirement for rrt in resistance can be overcome by increasing the levels of HRT via exogenous application of SA or by transgenic overexpression of HRT [22]–[25], [27]. HRT-mediated signaling is activated in the presence of TCV coat protein (CP) [27]–[29]. However, direct interactions between HRT and CP have not been detected.
Here, we examined the roles of EDS1, PAD4 and SAG101 in HRT-mediated signaling. We find that EDS1, but not PAD4 or SAG101, is required for CP-triggered HR in HRT expressing plants. This correlates with direct interactions between EDS1 and HRT. We also show that SAG101, which forms a ternary complex with EDS1 and PAD4, is an essential component of HRT-mediated signaling. Not only does SAG101 interact with PAD4 in the presence of EDS1, but it also induces the nuclear localization of EDS1. Conversely, the subcellular localization of the SAG101-EDS1-PAD4 ternary complex is driven by the location of EDS1. These results suggest that the inability of extranuclear EDS1 to complement eds1-1 phenotypes might be due to the altered localization of PAD4 and SAG101. Our studies also show that the Arabidopsis genome encodes two functionally redundant EDS1 isoforms, both of which can function in the R-mediated response to TCV or Pseudomonas syringae.
Genetics analysis of F2 plants derived from crosses between resistant ecotype Di-17 and susceptible plants eds1-1 (Ws ecotype) or eds1-2 (Ler ecotype) mutants showed that all HRT eds1 plants were susceptible to TCV (Table S1, [22], [24]). In comparison, ∼25% of F2 progeny (homo/heterozygous for HRT, but homozygous for rrt) were able to resist TCV infection in control crosses between Di-17 and Col-0/Ws/Ler [Table S1, 22]. Surprisingly, F2 progeny derived from a Di-17 x eds1-22 (At3 g48090; Col-0 ecotype, [21]) cross showed normal segregation of resistant plants; 25% of HRT plants showed resistance (Table S1). We investigated this further and realized that previous reports had indicated the presence of two EDS1 isoforms in the Arabidopsis ecotype Col-0 [17], [30], only one of which (encoded by At3 g48090, redesignated EDS1-90) has been functionally characterized [16], [17]. The other isoform (encoded by At3 g48080, designated EDS1-80) exhibits ∼85% amino acid (aa) identity with EDS1-90 (Figure S1A) and its transcript is induced by SA and TCV, similar to EDS1-90 (Figures S1B, S1C). Similar to Col-0, both Di-17 and Ler plants expressed both EDS1-80 and EDS1-90 but the EDS1-80 gene in Ler and Di-17 contained a 28 bp deletion in the second exon (Figures S1D, S1E). This deletion would result in the expression of a truncated EDS1-80 protein comprising of only the first 162 aa instead of the 629 aa long full-length protein. Thus, Ler-eds1-2 plants would essentially be defective in both EDS1 isoforms. Similarly, RT-PCR analysis showed that Ws and Ws-eds1-1 genotypes express EDS1-90, but not EDS1-80 (Figure S1F), suggesting that similar to Ler-eds1-2, Ws-eds1-1 plants are also compromised in the activities of both isoforms. These results also suggested that the presence of a functional EDS1-90 isoform in Di-17 was sufficient for HRT-mediated resistance to TCV. To reconfirm this we isolated a T-DNA knockout (KO) mutant in EDS1-80 in the Col-0 background (designated eds1-80; Figure S2A), crossed this KO line with Di-17 and analyzed segregation of resistance in the F2 plants (Table S1). Similar to Di-17 x Col-0 cross, plants of the HRT eds1-80 genotypes segregated normally for resistance; ∼25% plants were resistant to TCV (Table S1, Figures S2B, S2C). Genetic analysis based on EDS1-90 and EDS1-80 KO mutants suggested that either of the EDS1 isoforms can mediate HRT-mediated resistance to TCV.
We tested this further by evaluating the response of another R gene RPS4, which mediates resistance to Pseudomonas syringae expressing AvrRps4 (Figure S3) and is known to require EDS1. Unlike wild-type Ws plants, inoculation of avrRps4 bacteria induced prominent chlorosis and cell death in Ws-eds1-1 plants. The Col-0-eds1-80 and Col-0-eds1-90 plants on the other hand showed a similar response as wild-type Col-0 plants (Figures S3A, S3B). Similarly, pathogen inoculation induced SA and PR-1 levels in wild-type and eds1-80 or eds1-90 plants, but not in eds1-1 (Figures S3C, S3D). The eds1-80 or eds1-90 plants supported similar levels of bacterial growth as wild-type plants, which were ∼40-fold lower than that of the eds1-1 plants (Figure S3E). Together, these results show that single mutations in EDS1-80 or EDS1-90 in Col-0 background were insufficient to compromise RPS4-mediated resistance against avrRps4 bacteria.
To determine if EDS1-80 and EDS1-90 encoded functional proteins, corresponding to their orthologs in Ws and Ler ecotypes, we tested their ability to complement eds1-1 phenotypes. The EDS1-80 and EDS1-90 isoforms were expressed under the 35S promoter in the eds1-1 background (Figures S4A, S4B) and the T2 plants obtained from four independent lines expressing low or high EDS1 transcripts were analyzed for resistance to avrRps4 bacteria. Typical chlorosis and cell death phenotypes associated with avrRps4 infection on eds1-1 plants were not evident in plants expressing EDS1-80 or EDS1-90, regardless of their transcript levels (Figures 1A, S4C). Concurrently, these plants showed wt-like levels of ion-leakage (Figure 1B), PR-1 expression (Figure 1C), and SA levels (Figure 1D) in response to avrRps4 inoculation. The eds1-1 plants expressing EDS1-80 or EDS1-90 also supported wt-like growth of avrRps4 bacteria (Figure 1E). Together, these results suggest that both EDS1-80 and EDS1-90 encode functional proteins and expression of either gene complements the enhanced disease susceptibility phenotype in eds1-1 plants. Together, these results suggest that the two EDS1 isoforms likely function redundantly and that simultaneous mutations in both EDS1 isoforms are required to compromise HRT-mediated resistance to TCV.
To determine if EDS1 was required for the activation of HRT-mediated signaling we developed a transient system based on reconstitution of HR in Nicotiana benthamiana. This was essential since EDS1 and SA act redundantly to regulate HRT-mediated signaling in Arabidopsis [21], thereby rendering it difficult to test the function of EDS1 alone. This assay was facilitated by the fact that co-infiltration of HRT and its cognate avirulence effector CP induced a delayed and weak HR in N. benthamiana. Thus, any factor participating in the activation of HRT-mediated signaling should promote HR formation. Interestingly, co-infiltration of EDS1-80 or EDS1-90 together with HRT and CP promoted HR formation (Figure 2A), suggesting that both EDS1 isoforms likely facilitate the recognition of CP. This was further confirmed by assaying ion-leakage (Figure 2B). Unlike EDS1, co-infiltration of the eds1-1 mutant protein, SAG101 or PAD4 did not induce a strong HR in the presence of HRT and CP (Figures 2A, 2B). HRT and CP were expressed at comparable levels in the presence or absence of SAG101 and PAD4, suggesting that lack of HR in the HRT+CP+SAG101/PAD4 plants was not due to insufficient levels of HRT and/or CP (Figure 2C). Likewise, expression levels of SAG101 and PAD4 were similar to EDS1. Notably, as in Arabidopsis [17], eds1-1 protein was unstable and accumulated to very low levels in N. benthamiana (Figure 2C).
To determine if EDS1 promoted HRT+CP-dependent HR via interactions with HRT and/or CP, we carried out bimolecular fluorescence complementation (BiFC) assays (Figure 2D). As expected, HRT associated with its interacting partner CRT1 [21], [31], but no interaction was detected between either EDS1 isoforms and HRT or CP (Figure 2D). However, co-immunoprecipitation (IP) assays showed that EDS1 interacted with HRT, but not CP (Figures 2E, 2F). Neither HRT nor EDS1 interacted with GST (data not shown). Consistent with their ability to promote HR formation, both EDS1 isoforms associated with HRT in co-IP assays. Interaction between EDS1 and HRT was further verified by expressing these proteins under their native promoters in N. benthamiana plants (Figure S5A) as well as in Arabidopsis protoplasts (Figure S5B). Notably, EDS1 accumulated to similar levels when expressed under the 35S or its native promoter (Figure S5C). In comparison, HRT accumulated to higher levels when expressed under its native promoters, compared to 35S (Figure S5D). These results argue that interaction between HRT and EDS1 was not due to overexpression of these proteins. Together these results suggest that HRT associates with EDS1, albeit indirectly, and this likely facilitates the CP-triggered induction of HR in the presence of HRT.
In view of the functional redundancy between EDS1-80 and 90 and their association with HRT, it was important to determine if EDS1-80 was capable of forming a complex with PAD4 and SAG101. Indeed, similar to EDS1-90, EDS1-80 interacted with both PAD4 and SAG101 but not GST; the EDS1-80-PAD4 interaction was detected in both the periphery and nucleus of plant cells (Figure 3A). In comparison, the EDS1-80-SAG101 complex was primarily seen in the nucleus (Figure 3A). Co-IP assays further confirmed results obtained in the BiFC (Figures 3B, 3C). Since EDS1 and PAD4 are well known to regulate pathogen-induced accumulation of SA [11], [14], we next tested if SA altered the EDS1-80-PAD4 or EDS1-80-SAG101 interactions. No obvious differences in the intensity or site of interactions were noticed (data not shown), suggesting that increased SA might not alter these interactions. Unlike EDS1-90, the EDS1-80 isoform did not interact with itself or with EDS1-90 in BiFC assays (Figures 3A, S6A). In contrast, IP assays detected EDS1-80 interaction with itself and EDS1-90 (Figures 3D, 3E). This suggests that the homo and heterodimerization of EDS1-80 and EDS1-90 was likely indirect.
We next tested whether the presence of PAD4 or SAG101 affected the formation of the EDS1-80-90 heterodimer. EDS1-80 and EDS1-90 were co-expressed with PAD4 or SAG101, and immunoprecipitates were assayed for the EDS1-90, PAD4 or SAG101 proteins. Interestingly, EDS1-80 preferentially bound PAD4 in the presence of EDS1-90 (Figures 4A, S6B). EDS1-80 also showed slightly more affinity for SAG101 over EDS1-90 (Figure 4B). We next compared the relative affinities of EDS1-90 for PAD4 and SAG101. Similar levels of EDS1-PAD4 complex were detected in the absence or presence of SAG101 (Figure 4C). Similarly, levels of the EDS1-SAG101 complex did not alter significantly in the presence or absence of PAD4 (Figure 4D). We next assayed interaction of SAG101 and PAD4 with the lipase (LP; N-t 351 aa) and EDS1-PAD4-like (EP; C-t 351–623) domains of EDS1. Both SAG101 and PAD4 interacted with the LP, but not the EP, domain of EDS1 (Figure S6C). Together, these results suggest that SAG101 and PAD4 likely interact with different residues within the LP domain of EDS1 and therefore do not compete for binding with EDS1.
We noticed in our BiFC assays that the EDS1-SAG101 interaction occurred primarily in the nucleus (Figure 3A), even though EDS1-80 or 90 were present in both the cytosol and the nucleus (Figure 5A). We tested whether SAG101 influenced the subcellular localization of EDS1 and/or PAD4. Coexpression of EDS1-GFP with PAD4-RFP did not alter the localization of either protein; EDS1-80, EDS1-90 and PAD4 localized to the nucleus and periphery of the cell, similar to when expressed individually (Figures 5A, 5B, S7A). Similarly, coexpression of PAD4-GFP and SAG101-RFP did not alter the localization of either protein (Figure 5B). However, co-expression of EDS1-GFP with SAG101-RFP or SAG101-MYC altered the localization of EDS1, but not SAG101; in the presence of SAG101, EDS1 was preferentially detected in the nucleus (Figures 5B, S7A, S7B). This SAG101 triggered nuclear localization of EDS1-80 and EDS1-90 was not due to increased expression of EDS1 in the presence of SAG101 (Figure 5C). This result is in agreement with the previous report where co-localization of EDS1 and SAG101 was tested in wild-type Arabidopsis [17]. Interestingly, coexpression of PAD4-MYC or PAD4-Cerulean together with EDS1-GFP and SAG101-RFP retained some portion of EDS1 in the cytosol (Figures 5D, S7C). Notably, nuclear-cytoplasmic localization of EDS1 was only observed when PAD4 was coexpressed with EDS1 and SAG101. EDS1 remained preferentially in the nucleus when PAD4 was expressed 24 or 48 h after EDS1 and SAG101 (data not shown). This suggested that, rather than relocalizing EDS1, PAD4 merely retained it in the cytosol. This further suggested that EDS1 might be retained inside the nucleus in the presence of SAG101, although the nuclear localization of EDS1 was not dependent on SAG101 (data not shown). Unlike PAD4, SAG101 accumulated to higher levels when expressed under 35S, compared to its native promoter (Figures S7D, S7E). Thus, it was possible that the nuclear relocalization of EDS1 was dependent on the levels of SAG101. Indeed, when expressed under its native promoter, SAG101 was unable to relocalize all the cytosolic EDS1 into the nucleus (Figure 5E). These results, together with the observation that EDS1 and PAD4 levels change during pathogen infection [16], suggest that relative levels of EDS1, PAD4 and SAG101 might regulate distribution and/or localization of these proteins in response to pathogen stimulus.
To further confirm that the SAG101 triggered nuclear relocalization of EDS1 was a specific phenotype, we tagged EDS1 with a nuclear export signal (NES), or its mutant derivative (nes) [18]. As expected, both EDS1-80-NES and EDS1-90-NES were preferentially detected outside the nucleus while EDS1-80-nes and EDS1-90-nes localized like wild-type EDS1 (Figure 6A). Coexpression experiments showed that only EDS1-nes, but not EDS1-NES, relocalized to the nucleus in the presence of SAG101 (Figure 6B). Coexpression with PAD4 did not alter the localization of either form of EDS1. However, nuclear localization of PAD4 was affected by the presence of EDS1-NES, but not EDS1-nes (Figure 6B). Similar levels of EDS1-80-NES/nes protein in plants infiltrated with SAG101-RFP or PAD4-RFP suggested that localization of EDS1 was not associated with levels of protein expression (Figure 6C). Furthermore, EDS1-80-NES or EDS1-90-NES failed to induce the nuclear exclusions of RFP-tagged EDS1-90 (Figures 6B, S7D), suggesting that EDS1-NES-dependent extranuclear localization of SAG101 and PAD4 was a specific phenotype.
We next evaluated the interaction of PAD4 and SAG101 with EDS1-NES or EDS1-nes. As expected, EDS1-nes behaved similar to EDS1 (Figures 3A, 6D, S8). Notably, although EDS1-NES associated with both PAD4 and SAG101, these interactions occurred preferentially outside the nucleus (Figures 6D, S8). This was particularly evident in the case of SAG101, since the EDS1-SAG101 complex is normally located inside the nucleus. Together, these results suggest that the selective retention of EDS1 in a subcellular compartment can drive the localization of the EDS1-PAD4 and EDS1-SAG101 complexes. The fact that EDS1 can induce the cytosolic relocalization of SAG101 further suggests that, the previously reported inability of EDS1-NES to fully complement eds1-1 phenotypes might be due to the altered/mis-localization of PAD4 and/or SAG101, rather than the absence of EDS1 in the nucleus [18].
To determine if altered localization of EDS1-NES affected its ability to promote HRT-CP triggered cell death phenotype, we monitored visual phenotypes and ion-leakage in N. benthamiana plants infiltrated with HRT+CP+EDS1-NES. No significant difference was noticed in HRT+CP-mediated cell death phenotype induced in the presence of EDS1, EDS1-NES or EDS1-nes (Figures 7A, 7B). This further correlated with positive interaction seen between EDS1-NES and HRT proteins (Figure 7C). Together, these results suggested that the extranuclear retention of EDS1, and by extension that of PAD4 and SAG101, do not suppress HRT+CP-mediated HR.
The ability of EDS1, PAD4, SAG101 proteins to relocalize their interacting partners suggested that these proteins might be present in a ternary complex. However, consistent with earlier observations [17], we were unable to detect interactions between SAG101 and PAD4 in BiFC or co-IP assays (Figures 8A, 8B). We considered the possibility that factors present only during induced defense might be required for the PAD4-SAG101 association, if any. Since both EDS1 and PAD4 are known to regulate SA levels and because exogenous SA can induce reducing conditions required for relocating proteins [32], we tested binding between SAG101 and PAD4 in plants pretreated with SA. No SAG101-PAD4 interaction was detected in SA pretreated plants (Figure 8A). Another possibility was that SAG101-PAD4 interacted via a third protein, possibly EDS1, since both SAG101 and PAD4 interacted with EDS1. We tested the SAG101-PAD4 interaction in the presence of EDS1-80 or EDS1-90 (Figure 8A). Indeed, SAG101 and PAD4 associated with each other in the presence of either EDS1 isoform. This was further confirmed using co-IP assays (Figure 8C). Increased nuclear fluorescence in the BiFC assays suggested that a majority of the SAG101-EDS1-PAD4 complex was present in the nucleus. Pretreatment with SA did not alter formation or localization of the SAG101-EDS1-PAD4 complex. Interestingly, full length EDS1 was required for SAG101-EDS1-PAD4 complex formation, even though EDS1-LP domain alone was sufficient for interaction with SAG101 or PAD4 (Figure 8A). We next assayed the interaction between SAG101 and PAD4 in the presence of EDS1-NES and EDS1-nes, which were expressed as MYC tagged proteins (Figure S9). Surprisingly, in the presence of EDS1-NES, the SAG101-PAD4 interaction was preferentially detected outside the nucleus (Figure 8A), suggesting that the subcellular location of EDS1 might drive the localization of the SAG101-EDS1-PAD4 complex.
The fact that SAG101 forms a ternary complex with EDS1 and PAD4, and that it can drive the nuclear localization of EDS1 is inconsistent with a proposed redundant role for SAG101 in plant defense [17], [19]. We therefore investigated the requirement for SAG101 in HRT-mediated signaling. We crossed Di-17 plants with sag101 and analyzed HR and resistance in the F2 population. Approximately 75% of the plants showed HR to TCV, regardless of their genotype at the SAG101 locus (Figure 9A). HR phenotype correlated with increased expression of PR-1 gene in both HRT SAG101 and HRT sag101 plants (Figure 9B). However, all the HRT sag101 plants showed susceptibility to TCV and allowed increased accumulation of TCV in the distal tissues (Figures 9C, 9D, Table S1). The susceptible phenotype correlated with a significant reduction in SA and SAG accumulation in TCV inoculated HRT sag101 plants (Figure 9E). Together, these results suggested that SAG101 is required for HRT-mediated resistance.
To determine if the sag101 mutation compromised resistance to TCV by affecting the accumulation of HRT, we mobilized the HRT-FLAG transgene into HRT sag101 plants and analyzed HRT-FLAG levels. The HRT sag101 plants contained wt-like levels of HRT protein, before and after TCV inoculation (Figure 9F). In addition, as in Di-17 plants, all the HRT protein was present in the membranous fraction of extracts from HRT sag101 plants (Figure 9G). These results indicate that the inability of HRT sag101 plants to induce pathogen-responsive SA accumulation was not due to the altered levels or localization of the R protein. Notably, pretreatment with SA or its analog BTH restored resistance in HRT sag101 plants (Figure 9H). SA pretreatment also restored resistance in HRT eds1, but not in HRT pad4 plants (Figure 9H). The resistant and susceptible phenotypes in HRT sag101/HRT eds1 and HRT pad4 plants, respectively, correlated with HRT transcript levels; BTH treatment increased HRT transcript levels in HRT sag101 and HRT eds1, but not in HRT pad4 plants (Figure 9I). Thus, PAD4, but not EDS1 or SAG101, is required for the SA-mediated induction of HRT. Together, these results suggest that SAG101 does fulfill an independent function in HRT-mediated signaling.
Two functionally redundant isoforms of EDS1 in the Arabidopsis genome participate in resistance signaling such that the presence of either isoform is sufficient to mediate R-derived defense against microbial pathogens. Consistent with the co-operative roles of PAD4 and SAG101 in EDS1 function, either EDS1 isoforms can interact with PAD4 and SAG101, as well as form ternary complexes with these proteins. While SAG101 can drive the nuclear localization of EDS1, the presence of PAD4 can disrupt this to retain some EDS1 in the cytosol. This raises the possibility that the relative levels of SAG101 and PAD4 may drive the subcellular localization of EDS1. Conversely, EDS1 can also drive the localization of SAG101. For example, even though the majority of the EDS1-SAG101 or the SAG101-EDS1-PAD4 complexes are present in the nucleus, preferential retention of EDS1 in the cytosol via the addition of a nuclear export signal relocates SAG101 and the ternary complex to the cytosol. Furthermore, these data suggest that dynamic changes in the levels of EDS1/PAD4/SAG101 could drive the subcellular localization of the binary/ternary complexes to regulate defense signaling. These results also offer a possible mechanistic explanation for the inability of EDS1-NES to fully complement eds1 phenotypes [18].
A significantly large recovery of the EDS1-PAD4 and EDS-SAG101 complexes versus the SAG101-EDS1-PAD4 complex in co-IP studies suggests that EDS1 might be primarily present in a complex with PAD4 and SAG101. However, at this point we cannot discount the possibility that SAG101-EDS1-PAD4 complex is inherently unstable in cell free extracts. Indeed, earlier studies were unsuccessful in isolating SAG101-EDS1-PAD4 complex from cell free extracts [17], [33]. Notably, both PAD4 and SAG101 interact with the LP (1–350 aa in EDS1-90 and 1–357 aa in EDS1-80) domain of EDS1-80, which could potentially lead to steric hindrances resulting in protein instability. Intriguingly, our interaction studies with EDS1-LP and EP domains are not consistent with a previous report [16], which showed that the EP domain (351–623 aa) of EDS1-90 can self-interact, however the full-length EDS1-90 protein is required for interaction with PAD4. We find that the LP, but not the EP domains of both EDS1 isoforms, are required for interactions with self as well with PAD4 and SAG101. One possibility for these discrepancies is that our studies were carried out in planta, which likely better mimic the native environment than those in the Feys et al [17] study. Nonetheless, detection of the SAG101-EDS1-PAD4 complex required the full length EDS1 protein.
Unlike EDS1 and PAD4, which are well known to be essential for R protein-mediated defense to several pathogens [11], [16], [17], [20], [28], [29], SAG101 has been assigned a redundant role [17], [19]. This is because mutations in SAG101 do not compromise RPM1, RPS4, or RPP2-mediated resistance to the respective bacterial or oomycete pathogens, but does enhance susceptibility in pad4 plants [17]. In comparison to single mutants, simultaneous mutations in SAG101 and PAD4 also confer increased susceptibility to non-host pathogens [19], further supporting the functional redundancy between SAG101 and PAD4. We find that at least in the case of HRT-mediated signaling, SAG101 is as important as EDS1 and PAD4, since the absence of SAG101 alone can compromise HRT-mediated resistance to TCV. A PAD4-independent role for SAG101 is further supported by the fact that unlike PAD4, SAG101 is not required for the SA-mediated induction of HRT. Interestingly, similar to the eds1-1 and pad4-1 mutations [22], the sag101-1 mutation also reduced TCV-induced SA levels. Thus, similar to EDS1 and PAD4, SAG101 also regulates TCV-induced SA accumulation and might function in a feedback loop with SA.
The sag101 plants also show a nominal reduction in the levels of EDS1-90 and PAD4 proteins [17], which could contribute towards susceptibility to TCV in these plants. However, this is unlikely because sag101 plants are unaltered in RPS4-mediated resistance, which like HRT, is dependent on EDS1 and PAD4. In this regard, it is interesting to note that while loss of both EDS1 isoforms is thought to destabilize PAD4 and SAG101 [17] and thereby pathogen resistance, lack of a single isoform does not compromise resistance to TCV or avrRps4 bacteria. Clearly, the levels of EDS1, PAD4, SAG101 essential for initiating signaling response needs further clarification. It is possible that these proteins initiate normal signaling even at very low levels. This is not unusual as HRT cry2 and HRT phot2 plants show normal HR even though they contain significantly lower levels of HRT compared to wild type plants [24].
An important aspect that has not been addressed thus far is whether EDS1, PAD4, and SAG101 function as individual proteins or in a complex. Clearly, at least PAD4 fulfills a unique function in HRT-mediated signaling on its own; SA-mediated increase in HRT requires PAD4, but not EDS1 or SAG101. Similarly, only EDS1 facilitated HRT+CP-mediated HR to TCV. The EDS1-80-EDS-90 complex also does not appear to be essential for HRT- or RPS4-mediated signaling, since these continue to function in the eds1-80 or eds1-90 mutants. Interaction between EDS1 and HRT and the fact that EDS1 forms a complex with PAD4 and SAG101 suggests that these proteins might be part of a multi-protein complex and thus regulate signaling by modulating the activity of HRT. The absence of EDS1-HRT interaction in the BiFC assay suggests that EDS1 may associate indirectly with HRT. Interestingly, EDS1 does not dissociate from HRT in the presence of CP, suggesting that CP-triggered activation of HRT does not involve the release of EDS1. Whether CP-triggered activation of HRT utilizes the EDS1, PAD4, and SAG101 proteins individually or as complexes needs further clarification. However, the requirements for all three proteins do support the notion that the SAG101-EDS1-PAD4 ternary complex might be important. It will indeed be important to establish the biochemical activities of EDS1, PAD4 and SAG101 proteins in order to accurately access the importance of the ternary complex in R protein-mediated signaling.
Plants were grown in MTPS 144 Conviron (Winnipeg, MB, Canada) walk-in-chambers at 22°C, 65% relative humidity and 14 hour photoperiod. The photon flux density of the day period was 106.9 µmoles m−2 s−1 and was measured using a digital light meter (Phytotronic Inc, Earth city, MO). All crosses were performed by emasculating the flowers of the recipient genotype and pollinatng with the pollen from the donor. F2 plants showing the wt genotype at the mutant locus were used as controls in all experiments. The wt and mutant alleles were identified by PCR, CAPS, or dCAPS analysis [21]–[26]. The EDS1 KO mutant in At3 g48080 and At3 g48090 were isolated by screening SALK_019545 and SALK_071051 insertion lines, respectively. This EDS1 KO in At3 g48090 was previously designated eds1-22 and redesignated here as eds1-90. The homozygous insertion lines were verified by sequencing PCR products obtained with primers specific for the T-DNA left border in combination with an EDS1-specific primer (Table S2).
The full length cDNA corresponding to EDS1-80 and EDS1-90 genes were PCR amplified using linkered primers and cloned downstream of 35S promotor in pRTL2.GUS. For Arabidopsis transformation, the fragment containing the promotor, cDNA and terminator was removed from pRTL2-EDS1 vectors and cloned into binary vector pCambia or pBAR. These clones in binary vectors were moved into Agrobacterium tumefaciens strain MP90 by electroporation and were used to transform Arabidopsis via the floral dip method. Selection of transformants was carried out on medium containing hygromycin or soil sprayed with the herbicide BASTA.
Small-scale extraction of RNA from one or two leaves was performed with the TRIzol reagent (Invitrogen, CA), following the manufacturer's instructions. Northern blot analysis and synthesis of random-primed probes for PR-1 and PR-2 were carried out as described previously [26].
RNA quality and concentration were determined by gel electrophoresis and determination of A260. Reverse transcription (RT) and first strand cDNA synthesis were carried out using Superscript II (Invitrogen, CA). Two-to-three independent RNA preparations were used for RT-PCR and each of these were analyzed at least twice by RT-PCR. The RT-PCR was carried out for 35 cycles in order to determine absolute levels of transcripts. The number of amplification cycles was reduced to 21–25 in order to evaluate and quantify differences among transcript levels before they reached saturation. The amplified products were quantified using ImageQuant TL image analysis software (GE, USA). Gene-specific primers used for RT-PCR analysis are described in Table S2.
The leaves were vacuum-infiltrated with trypan-blue stain prepared in 10 mL acidic phenol, 10 mL glycerol, and 20 mL sterile water with 10 mg of trypan blue. The samples were placed in a heated water bath (90°C) for 2 min and incubated at room temperature for 2–12 h. The samples were destained using chloral hydrate (25 g/10 mL sterile water; Sigma), mounted on slides and observed for cell death with a compound microscope. The samples were photographed using an AxioCam camera (Zeiss, Germany) and images were analyzed using Openlab 3.5.2 (Improvision) software.
The bacterial strain DC3000 derivatives containing pVSP61 (empty vector), or avrRps4 were grown overnight in King's B medium containing rifampicin and kanamycin (Sigma, MO). The bacterial cells were harvested, washed and suspended in10 mM MgCl2. The cells were diluted to a final density of 105 CFU/mL (A600) and used for infiltration. The bacterial suspension was injected into the abaxial surface of the leaf using a needle-less syringae. Three leaf discs from the inoculated leaves were collected at 0 and 3 or 6 dpi. The leaf discs were homogenized in 10 mM MgCl2, diluted 103 or 104 fold and plated on King's B medium.
Transcripts synthesized in vitro from a cloned cDNA of TCV using T7 RNA polymerase were used for viral infections. For inoculations, the viral transcript was suspended at a concentration of 0.05 µg/ µL in inoculation buffer, and the inoculation was performed as described earlier [31]. After viral inoculations, the plants were transferred to a Conviron MTR30 reach-in chamber maintained at 22°C, 65% relative humidity and 14 hour photoperiod. HR was determined visually three-to-four days post-inoculation (dpi). Resistance and susceptibility was scored at 14 to 21 dpi and confirmed by northern gel blot analysis. Susceptible plants showed stunted growth, crinkling of leaves and drooping of the bolt.
A protocol adapted from Dellagi et al. [34] was used for conductivity measurements.
Briefly, 5 leaf discs per plant (7 mm) were removed with a cork borer, floated in distilled water for 50 min, and subsequently transferred to tubes containing 5 ml of distilled water. Conductivity of the solution was determined with an NIST traceable digital Conductivity Meter (Fisher Scientific) at the indicated time points. Standard deviation was calculated from four replicate measurements per genotype per experiment.
SA and SAG quantifications were carried out from ∼300 mg of leaf tissue as described before [23].
SA or BTH treatments were carried out by spraying or subirrigating 3-week-old plants with 500 µM SA or 100 µM BTH, respectively.
Proteins were extracted in buffer containing 50 mM Tris-HCl, pH7.5, 10% glycerol, 150 mM NaCl, 10 mM MgCl2, 5 mM EDTA, 5 mM DTT, and 1 X protease inhibitor cocktail (Sigma-Aldrich, St. Louis, MO). Protein concentration was measured by the Bio-RAD protein assay (Bio-Rad, CA).
For Ponceau-S staining, PVDF membranes were incubated in Ponceau-S solution (40% methanol (v/v), 15% acetic acid (v/v), 0.25% Ponceau-S). The membranes were destained using deionized water.
For soluble versus microsomal fractionations, proteins were extracted in buffer containing 50 mM Tris-MES, pH 8.0, 0.5 M sucrose, 1 mM MgCl2, 10 mM EDTA, 10 mM EGTA, 10 mM ascorbic acid, 5 mM DTT and 1X protease inhibitor cocktail (Sigma-Aldrich, St. Louis, MO). Total protein extract was centrifuged at 10,000 g followed by a second centrifugation at 125,000 g. The microsomal fraction was suspended in a buffer containing 5 mM potassium phosphate pH 7.8, 2 mM DTT and 1 X protease inhibitor cocktail.
Proteins (30–50 µg) were fractionated on a 7–10% SDS-PAGE gel and subjected to immunoblot analysis using α-CP, α-MYC, α-FLAG (Sigma-Aldrich, St. Louis, MO) or α-GFP antibody. Immunoblots were developed using ECL detection kit (Roche) or alkaline-phosphatase-based color detection.
Coimmunoprecipitations were carried out as described earlier [24].
Protoplasts were isolated from three-week-old Arabidopsis Col-0 plants as described earlier [35]. For protoplast transfection, ∼104 protoplasts were incubated at room temperature with 20 µg of plasmid DNA and an equal volume of solution containing 40% PEG 4000, 0.1 M CaCl2 and 0.2 M mannitol. After 5 min, 3 ml of wash solution containing 154 mM NaCl, 125 mM CaCl2, 5 mM KCl, 5 mM glucose, 2 mM MES (pH 5.7) was added slowly to the protoplast and the protoplasts were pelleted by centrifugation at 100 x g for 1 min. The protoplasts were washed twice and finally suspended in 1 ml of wash solution. The protoplasts were incubated in a round bottom glass vial for 12 h prior to protein extraction.
For confocal imaging, samples were scanned on an Olympus FV1000 microscope (Olympus America, Melvile, NY). GFP (YFP), CFP (Cerulean) and RFP were excited using 488, 440, and 543 nm laser lines, respectively. Constructs were made using pSITE [36] or pEarlyGate binary vectors using Gateway technology and introduced in A. tumefaciens strain LBA4404 or MP90 for agroinfiltration into N. benthamiana or Arabidopsis, respectively. Agrobacterium strains carrying various constructs were infiltrated into wild-type or transgenic N. benthamiana plants expressing CFP-tagged nuclear protein H2B or Arabidopsis plants. 48 h later, water-mounted sections of leaf tissue were examined by confocal microscopy using a water immersion PLAPO60XWLSM 2 (NA 1.0) objective on a FV1000 point-scanning/point-detection laser scanning confocal 3 microscope (Olympus) equipped with lasers spanning the spectral range of 405–633 nm. RFP, CFP and GFP overlay images (40X magnification) were acquired at a scan rate of 10 ms/pixel. Images were acquired sequentially when multiple fluors were used. Olympus FLUOVIEW 1.5 was used to control the microscope, image acquisition and the export of TIFF files.
|
10.1371/journal.pgen.1004117 | Loss of Trabid, a New Negative Regulator of the Drosophila Immune-Deficiency Pathway at the Level of TAK1, Reduces Life Span | A relatively unexplored nexus in Drosophila Immune deficiency (IMD) pathway is TGF-beta Activating Kinase 1 (TAK1), which triggers both immunity and apoptosis. In a cell culture screen, we identified that Lysine at position 142 was a K63-linked Ubiquitin acceptor site for TAK1, required for signalling. Moreover, Lysine at position 156 functioned as a K48-linked Ubiquitin acceptor site, also necessary for TAK1 activity. The deubiquitinase Trabid interacted with TAK1, reducing immune signalling output and K63-linked ubiquitination. The three tandem Npl4 Zinc Fingers and the catalytic Cysteine at position 518 were required for Trabid activity. Flies deficient for Trabid had a reduced life span due to chronic activation of IMD both systemically as well as in their gut where homeostasis was disrupted. The TAK1-associated Binding Protein 2 (TAB2) was linked with the TAK1-Trabid interaction through its Zinc finger domain that pacified the TAK1 signal. These results indicate an elaborate and multi-tiered mechanism for regulating TAK1 activity and modulating its immune signal.
| Chronic activation of immune responses results in health problems including gastrointestinal infections, metabolic imbalances and inflammatory bowel diseases that may lead to colorectal cancer. Central to this, is the balance of activation/restriction of nuclear factor-κB (NF-κB) during innate immune responses. To study signaling through NF-κB, we use the fruit fly Drosophila melanogaster as a genetically tractable model system that reflects human biology (due to the evolutionary conservation between innate immunity in flies and mammals), while reducing the complexity of the human disease of interest. We have found a new negative regulator of the Drosophila NF-κB pathway named Trabid. Its loss released the pathway and resulted in constitutive immune activation both in the gut as well as in the whole fly. This spontaneous immune activation reduced life span in the absence of infection, especially when it was combined with loss of another known negative regulator of the same pathway, a protein named Pirk. Stem cell activity in the gut in a pirk;trabid double mutant was found to be significantly increased, as the gut was trying to balance enterocyte loss. Trabid was acting at the level of TGF-beta Activating Kinase 1 (TAK1), which triggers both immunity and cell death.
| Conjugation of Ubiquitin (Ub) and formation of polyubiquitin chains on proteins can stimulate the assembly of reversible, short-lived signalling centres [reviewed in 1]. The most studied of the different polyubiquitin chain types are the K48-linked and K63-linked chains. K48-linked polyubiquitin chains target a protein for proteosomal degradation while polyubiquitin chains linked by K63 function in processes including signal transduction, DNA repair and transcription, through a degradation-independent mechanism [2]. Ubiquitin-mediated signalling is particularly important for both activating and restricting the activity of nuclear factor-κB (NF-κB) during innate immune responses where deregulation leads to chronic inflammation and cancer [reviewed in 3].
In Drosophila, the IMD pathway, which shows striking similarities to the ones stimulated by members of the mammalian TNF-receptor super-family, is strongly triggered by DAP-type peptidoglycan, a cell wall component of Gram-negative bacteria and Gram-positive bacilli [reviewed in 4]. It is assumed that fragments of peptidoglycan released by these bacteria bind the peptidoglycan recognition proteins LC or LE at the cell surface or the cytosol respectively leading to their multimerization [5]. The signal is then transduced through a receptor-adaptor complex comprising Imd itself (homologous to the mammalian Receptor Interacting Protein RIP, minus the kinase domain) [6], dFADD (FAS-associated death domain protein homologue) [7], [8] and the caspase-8 homologue Dredd (death-related Ced-3/Nedd2-like protein) [9]. DREDD is K63-linked ubiquitinated by the Drosophila Inhibitor of apoptosis-2 (dIAP-2), which acts as an E3-ligase promoting DREDD activation [10]. DREDD then cleaves Imd thus unmasking a domain of interaction with dIAP-2 for further dIAP-2-dependent Ub-conjugation this time on Imd itself [11]. Through its RING domain, dIAP-2 ubiquitinates and stabilises Imd, which then acts like a scaffold for the recruitment of downstream components [11]. These components may include the Drosophila TGF-beta-activating-kinase 1 (dTAK1), together with adaptor TAK1-associated Binding Protein 2 (dTAB2) [11]–[13]. TAK1/TAB2 play a critical role in activating the Drosophila IκB kinase (IKK) complex and also transiently activate the c-Jun-N-terminal kinase (JNK) pathway [14]. In this IKK/JNK dichotomy, the IKK complex represents the branch of the pathway that phosphorylates the NF-κB transcription factor Relish [15]. It is probable (but not proven) that Dredd cleaves the inhibitory C-terminal domain of phosphorylated Relish helped by an IKK scaffold [15], [16] thus liberating the N-terminal portion to move to the nucleus and regulate expression of transcriptional targets such as the antimicrobial peptide (AMP) gene diptericin (dipt).
IMD signalling shows acute phase profile in terms of AMP triggering where induction is rapid following infection [17]. Negative regulation plays an important role in restricting the response both inside as well as outside of the cell in epithelia and systemic infection. Outside of the cell Peptidoglycan Recognition Proteins (PGRPs) with an amidase activity act to down-regulate the pathway following microbial sensing [18]. Inside the cell, Pirk negatively regulates the receptor PGRP-LC [19]–[21] while dUSP36 inhibits Imd itself [22] and CYLD the IKK complex [23]. Relish plays a crucial role in limiting the signal through proteosomal degradation of dTAK1 [24]. Nevertheless, it is still unclear how dTAK1 is activated although both Dredd and K63-polyubiquitin chains may be involved [25], [15]. Here we report the discovery of Trabid as a novel component of the IMD pathway and a negative-regulator of dTAK1. Trabid altered K63-linked polyubiquitination in dTAK1 through its OTU and NZF domains attenuating the immune-related but not the JNK-related signalling output of dTAK1. We found Lysine 142 of dTAK1 to be critical for its function in the pathway as the probable K63 polyubiquitin acceptor site. Further, K156 functioned as a potential K48 Ub acceptor site. In addition, dTAB2 was found to interact with Trabid and modulate dTAK1 activity through its Zinc finger domain. Together, our findings indicate an elaborate and multi-tiered process modulating dTAK1 signalling activity, during Gram-negative bacterial infection.
In mammals, K63-linked polyubiquitination of TAK1 at Lys158 is critical for activating several signalling cascades including Tumor Necrosis Factor alpha (TNFα), Interleukin-1beta (IL-1beta)-induced IkappaB kinase (IKK)/nuclear factor-kappaB (NF-κB) and c-Jun N-terminal kinase (JNK)/activator protein 1 (AP-1) pathway [26]. We aligned the sequences of human TAK1 (hTAK1; 1–230 amino acids) and its fruit fly orthologue (dTAK1) to determine the corresponding Lys in Drosophila (Figure 1A). Based on the sequence of residues around Lys158 in hTAK1, Lys142 of dTAK1 was identified as the most probable candidate. Previous studies had also implicated Lys34 and 209 as Ub acceptor sites in hTAK1 [27]. Based on sequence alignments we identified Lys194 as the Drosophila equivalent of Lys209 in hTAK1. However, the equivalent of Lys34 in humans is not present in Drosophila (Figure 1A). It was also plausible that the K63 & K48 putative Ub acceptor sites would be in close proximity to both each other and to the kinase activation domain. Hence, Lys134, 156 and 189 were also selected as likely candidates.
Mutation constructs dTAK1K142R-V5, dTAK1K134R-V5, dTAK1K156R-V5, dTAK1K189R-V5 & dTAK1K194R-V5 were made changing the Lys to Arg at these sites. These point mutations maintained the positive charge but could not serve as an acceptor site for Ub modification. The constructs were then screened for their ability to activate IMD immune signalling in cell culture using quantitative real time PCR (qPCR) to measure induction of dipt 48 hrs post-transfection. While over expression of dTAK1 activated IMD signalling as previously observed [14], concomitant overexpression with dTAB2 resulted in increased activation (Figure 1B). In this screen, dTAK1K142R and dTAK1K156R showed significantly reduced dipt induction when compared to wild-type dTAK1 while all other mutants activated dipt at wild type levels (Figure 1B; Figure S1).
However, the fact that the signalling capacity of dTAK1K156R and dTAK1K142R was significantly reduced could have been due to the two mutant proteins not folding properly and being therefore, non-functional. As a result, they would be targeted for degradation by the proteasome. If this were the case, using a proteasome inhibitor one could show increased accumulation of non-degraded mutants dTAK1K156R and dTAK1K142R in comparison to wild type dTAK1. A time-course expression analysis was performed after treatment with 26S proteasomal inhibitor MG132 (75 µM for 8 hrs) at concentrations that are known to block proteasome activity as previously described [16]. Results showed that expression profiles of wild type dTAK1, and those of dTAK1K142R and dTAK1K156R were similar, indicating that the mutant proteins were not accumulating more than wild type dTAK1 and thus were presumably folding correctly (Figure S2). Therefore, dTAK1K142R and dTAK1K156R were selected for further analysis with the working hypothesis that the mutated Lysines were essential for dTAK1 immune activity.
We next sought to determine whether there was a difference in the ubiquitination profile of dTAK1 and dTAK1K142R mutant. Co-overexpression of hTAK1 and TAB1 in cell culture results in hTAK1 polyubiquitination [26]. This assay was modified for Drosophila S2 cells. Expression vectors encoding for C-terminally V5 tagged dTAK1 or dTAK1K142R were co-transfected with dTAB2-HA and cMyc-Ub into S2 cells. Cells were lysed 48 hrs post-transfection, immunoprecipitated with anti-V5 antibody, resolved on 10% SDS PAGE and immunoblotted with anti-cMyc antibody (Figure 2A). In contrast to human TAK1, where mutation of the Lys158 Ub acceptor site to Arginine resulted in failure of TGF-β-induced ubiquitination [26], [27], ubiquitination of dTAK1K142R was dramatically increased. Examination of the cell lysates showed greater degradation of dTAK1K142R when compared with wild type dTAK1 (Figure 2A, see dTAK1/dTAK1K142R panel).
We then analysed the linkage type of these polyubiquitination chains to distinguish whether they were K48 or K63-linked chains. We co-transfected dTAK1-V5 or dTAK1K142R-V5 and dTAB2-HA together with cMyc-Ub mutants having only one Lys residue at position 48 or 63 (UbK48 or UbK63; see experimental procedures). Wild type dTAK1 showed primarily K63-linked polyubiquitination (Figure 2B). As expected from the degradation seen in the cell lysates of Figure 2A, elimination of Lys142 severely compromised the ability of dTAK1 to form K63-linked polyubiquitination chains, although it retained the ability to form K48-linked chains (Figure 2B). Therefore, in agreement with results on human TAK1, there appeared to be two separate Ub acceptor sites for K48 and K63-linked polyubiquitination chains in Drosophila TAK1 with Lys142 being the probable K63 Ub acceptor site.
We then sought to determine the K48 Ub acceptor site and asked whether there was a difference in the Ub profile of dTAK1 and dTAK1K156R. Expression vectors encoding dTAK1-V5 or dTAK1K156R-V5 were transiently transfected together with dTAB2-HA and cMyc–Ub into S2 cells and Ub assays performed as above. Results showed that overall ubiquitination in the K156R mutant was greatly reduced in comparison to wild-type dTAK1 (Figure 3A).
We next identified, which type of polyubiquitination (i.e. whether K48 or K63) had been affected by the K156R mutation. Expression vectors encoding dTAK1-V5 or dTAK1K156R-V5 were transiently transfected together with dTAB2-HA and either cMyc – UbK48 or cMyc-UbK63 into S2 cells, in combinations shown and Ub assays performed (Figure 3B). Results showed that K48-linked polyubiquitination was significantly diminished in dTAK1K156R. Moreover, K63-linked polyubiquitination was not reduced in dTAK1K156R (Figure 3B). The caveat in the above is that results have been obtained through overexpressing the relevant proteins and looking at ubiquitination. However, these data suggest that K63-linked ubiquitination must precede TAK1 activation whereas 48K-linked ubiquitination must in its turn follow to dampen the signal. We have observed such a sequence of events in a time-course experiment where we precipitated endogenous TAK1 with an antibody against it [15] and blotted for cMyc-UbK63 or cMyc-UbK48 following addition of peptidoglycan (PG). Our results show that 2 h following challenge with PG from E. coli there is a bias towards K63-linked ubiquitination, which is gradually shifted towards K48-linked ubiquitination at the end of the 6 h time course (Figure S3A). Moreover, activation of AMP-related immune responses with PG was comparable to TAK1 overexpression through transient transfection (Figure S3B).
As implied from the results presented so far, K63-ubiquitination played a critical role in activating dTAK1. It follows therefore, that de-ubiquitination would be required for terminating the dTAK1 signal. In mammals, the zinc-finger protein A20 (also known as TNFAIP3) has been identified as negative regulator of NF-κB transcription factors in both TNF and IL-1 signalling through its deubiquitinating (DUB) and ubiquitin editing functions. In this model, K63-linked chains are cleaved and re-arranged to form K48-linked Ub chains, thereby tagging the protein for proteosomal degradation [28], [ reviewed in 29].
Trabid (Trbd) was originally discovered as a positive regulator of both the mammalian and Drosophila Wnt pathway with a remarkable preference for binding to, and cleaving, K63-linked ubiquitin chains [30]. Trbd is also a representative of the A20 OTU family in fruit flies [30]. To determine whether Trbd also functioned in IMD signalling, we tested for its interaction with dTAK1. Co-immunoprecipitation assays showed that dTrbd bound to dTAK1 (Figure 4A). Trbd also bound TAB2 as shown by relevant co-immuno-precipitation experiments (see below).
We then explored the in vivo effects on IMD signalling of a trbd deletion generated by homologous recombination [30]. We assayed AMP expression in 3–6 days old whole animals without an infection. We used flies homozygous for the trbd deletion; flies homozygous for a mutation in pirk [pirkEY00723, ref 21]; yellow-white (yw) flies as the genetic background used for both the initial trbd targeting construct [30] and by Gene Disruption Project [31], which generated pirkEY00723; flies mutant for both trbd and pirk; white118 flies (w118) as an additional control. We observed a significant increase in expression of attacinD, drosocin (dro), dipt and CecropinA1 in both the trbd and pirk; trbd flies relative to the yw and w1188 controls as well as to the pirk single mutant (Figure 4B). This meant that removal of trbd (or both pirk and trbd) would increase the levels of AMPs in a systemic fashion leading to a chronic response in the absence of infection. Injection of bacteria did not increase AMPs further and following systemic infection with E. coli 1106 the levels of dipt or dro gene expression between control and trbd or pirk;trbd flies were statistically inseparable (Figure 4C; Figure 4D). No further increase in whole fly AMP levels following systemic infection was observed in mutants of other negative regulators (e.g. CYLD, Nubbin) [23], [32]. Further increases maybe tissue specific (gut; see below) and would therefore fail to be detected in whole fly preps.
The Drosophila IMD pathway is also the mediator of local immune responses in the gut. In contrast to whole animals, we observed that the gut of heterozygous trbd flies showed a 2-fold increased expression of dipt over and above the wild type levels of induction following infection with Erwinia carotovora carotovora (Ecc15) and assaying using qPCR 24 h later (Figure 4E). Loss of both copies of trbd resulted in dipt induction 3 times as much as the wild type control. Finally, concomitant loss of pirk and trabid led to a 5-fold induction of dipt over and above wild type activation levels (Figure 4E). This was suppressed in a dredd mutant background (Figure 4E). Over-activation of dipt was not due to a delay in bacterial clearance as exemplified by measuring Colony forming Units (CFUs) following oral Ecc15 infection. Clearance in trbd and pirk; trbd mutants was statistically indistinguishable from wild type controls or pirk single mutants (Figure S4). Interestingly, bacterial clearance following systemic infection in trbd and pirk; trbd was statistically significantly faster than wild type and pirk flies using both a low (approx. 400 cells; Figure S5A) and a high (approx. 4000 cells; Figure S5B) dose of E. coli 1106. These differences were just below the limit of statistical significance following Ecc15 systemic infection with the same doses (Figure S5C and S5D). Nevertheless, the E. coli 1106 result correlated with the observation that trbd and trbd;pirk flies had a much higher level of AMPs to begin with (Figure 4B), suggesting a protective effect.
This potential protective effect however, had a cost. We found that deletion of trbd severely compromised the life span of flies (in the absence of infection). Our results are shown in Figure 5. 50% of flies heterozygous for both pirk and trbd survived the 30-day mark (LT50 = 32; Figure 5A and 5C). A similar effect was observed in flies deficient for pirk in a trbd heterozygous background (LT50 = 27; Figure 5A and 5C). Just deleting trbd (in a pirk heterozygous background) had serious consequences, as 50% of flies were dead by 18 days (LT50 = 18; Figure 5A and Figure 5C). More significantly however, the double mutant pirk; trbd had a dramatic reduction in life span compared to either single mutant (+/pirk; trbd or pirk; +/trbd) or double heterozygote (+/pirk; +/trbd) since 50% of flies were dead before the 14-day mark (LT50 = 14; Figure 5A and 5C). This phenomenon was suppressed in dredd; pirk; trbd flies (LT50 = 37) where IMD was inactive (Figure 5A; Figure 5C).
Interestingly, there was no statistically significant change (LT50 = 14 to LT50 = 18) when pirk; trbd flies were grown on germ-free conditions showing that it was largely the host rather than a disturbance in microbiota that was the cause for the reduction in life span (Figure 5B; Figure 5C). Crucially for this assumption, flies that had a blocked IMD pathway (dredd; pirk; trbd) lived as long as wild type (yw) flies in germ-free conditions (Figure 5B; Figure 5C).
The above result meant that chronic activation of IMD in the absence of infection (seen in Figure 4B) was the cause of life span reduction. Moreover, gut homeostasis was disrupted. Upd3 is a ligand secreted by stressed enterocytes, which activates the JAK-STAT pathway in intestinal stem cells to promote both their division and differentiation [33]–[35]. In comparison to wild type flies, we observed a significantly higher level of JAK-STAT activity in the guts of unchallenged trbd and even more in pirk;trbd flies as monitored by the expression of upd3 and the JAK-STAT target gene Socs36E in qPCR assays (Figure S6A; Figure S6B, respectively). The presence of dredd fully suppressed this elevated JAK-STAT activity observed in pirk;trbd flies, demonstrating that excessive Imd pathway activation must cause gut damage, which in turn induces epithelium renewal (Figure S6A; Figure S6B). From the above experiments we concluded therefore, that 1) dTrbd physically interacted with dTAK1, 2) negatively regulated the IMD signal in vivo and 3) its deletion had an impact on life span and gut homeostasis due to chronic activation of immune signalling. Moreover, reduction of life span in trbd deletion flies was enhanced when Pirk, a known negative regulator of the IMD pathway was also mutated.
A20 proteins (including Trbd) belong to the ovarian tumour (OTU) family of DUBs. The OTU is a conserved cysteine protease domain that possesses DUB activity [reviewed in 36]. Trbd possesses three N-terminal NZF (Npl4 zinc finger) domains and a C-terminal OTU domain, which shows DUB activity with a preference for K63-linked ubiquitin [30]. Further, two or more NZF domains are required for binding to K63-linked Ub chains. The catalytic residue in the OTU domain in humans is C443. Sequence alignment of hTrbd and dTrbd showed that C518 was most probably the corresponding catalytic residue in Drosophila (Figure 6A). Two constructs were made: dTrbdC518S (where C518 was changed to S) and dTrbdC518S+3xNZFDel, where along with C518S the first Cys residue of each of the 3 NZF domains namely, C13, C94 and C238, were mutated to Ala. To determine their functional importance, these mutations, dTAK1 and dTAB2 were co-transfected along with dTrbdC518S and dTrbdC518S+3xNZFDel in S2 cells and relative dipt expression assayed using qPCR. As expected, over-expression of dTrbd significantly decreased the TAK1/TAB2-mediated dipt induction although it did not completely abrogate it (Figure 6B). The C518S mutation substantially relieved the suppressive effect of wild type Trbd on dipt expression levels, with the C518S+3xNZFDel showing no suppression whatsoever (Figure 6B). Expression levels of puckered (a target of the JNK cascade also induced through TAK1/TAB2 activity) revealed that the pathway was not affected, indicating that dTrbd functioned solely in the IMD pathway (Figure S7). It has been suggested that it is Plenty of SH3 (POSH), which terminates the JNK-related dTAK1 signal [37].
To determine the relative contribution of the OTU and NZF domains in the ability of dTrbd to cleave K63-linked Ub chains [30] we tested the overexpression of dTrbd as well as dTrbdC518S and dTrbdC518S+3xNZFDel on dTAK1 K63-linked ubiquitination. As expected, dTrbd substantially reduced K63-linked ubiquitination of dTAK1 when co-transfected with dTAB2 (Figure 6C). Ubiquitination was moderately affected by dTrbdC518S and not at all by dTrbdC518S+3xNZFDel (Figure 6C). We also tested the effects of dTrbd, dTrbdC518S and dTrbdC518S+3xNZFDel on K48-linked ubiquitination. Results showed that K48-ubiquitination on TAK1, was not affected by dTrbd wild type, or dTrbd mutants (Figure S8).
We finally wanted to explore the tripartite relationship between TAK1, TAB2 and Trbd. In mammals, TAB2 and TAB3 function as adaptors, which link TRAF2 and TRAF6 to TAK1, facilitating complex formation and activation of TAK1 in IL-1 and TNF-induced NF-kB activation [38], [39]. Both TAB2 and TAB3 contain a highly conserved C-terminal novel zinc finger domain, which binds preferentially to K63-linked polyubiquitin chains. Mutations in this domain abolish the ability of TAB2 and TAB3 to bind polyubiquitin chains, as well as their ability to activate TAK1 [39], [40].
We generated a dTAB2 mutant (dTAB2ZnFDel see Figure 7A) where the first Cys residues in the C-terminal Zinc Finger (ZnF) motif were changed to Ala (C769A and C772A) as previously performed with human TAB2 [40]–[42]. Thereafter, the effect of this mutation on IMD signalling was assayed in cell culture using qPCR. Drosophila TAK1 was transfected either alone or together with dTAB2 or dTAB2ZnFDel into S2 cells and dipt expression assayed 48 h post infection using qPCR. Interestingly, signalling intensity was doubled in the presence of dTAB2ZnFDel when compared with wild type TAB2 (Figure 7B). Increased dipt induction may be the result of greater dTAK1 activation. Therefore the level of K63-linked ubiquitination of dTAK1 was examined in an S2 cell-based Ub assay in the presence of either TAB2 or dTAB2ZnFDel. Our results showed that, K63-linked ubiquitination of dTAK1 was increased in the presence of dTAB2ZnFDel (Figure 7C, left lane). These results suggested therefore that the mutation of the dTAB2 ZnF domain might stabilise dTAK1 by enhancing its 63K-linked ubiquitination. This led to an increase in signalling capacity seen by increased dipt induction. It appeared therefore that the TAB2 C-terminal ZnF domain restricted dTAK1 activity and thereby the IMD signal. Interestingly, dTAB2ZnFDel interacted more strongly (judging from the intensity of the signal) with dTrabid than wild type dTAB2 (Figure 7D). This might indicate that the tighter dTrbd bound to dTAB2 the higher the signalling capacity of dTAK1.
Our results point to a multi-tiered regulation of dTAK1 signalling in the IMD cascade. Ubiquitination played a key role in this regulation. It has been previously shown that in humans Lys158 was the Ub acceptor for K63 ubiquitination while Lys72 was the Ub acceptor for K48 ubiquitination [26]–[29]. In dTAK1 both Lys142 and Lys156 proved to be important for immune signalling as seen in qPCR assays in cell culture following transient transfection of these mutants. Our results indicate that in Drosophila TAK1, Lys142 functioned as the probable K63-acceptor site and Lys156 functioned as the probable K48-acceptor site. Our working hypothesis therefore, is that there is a main K63-Ub acceptor site in dTAK1 namely Lys142, which positively modulates the IMD immune signal. An interesting aspect of our results is that mutations in Lys142 and Lys156 do not affect the JNK pathway, in the transient activation of which TAK1 is involved. This suggests that AMP induction and JNK target activation can be independent in agreement with Silverman and co-workers [14]. In contrast to our results, Delaney et al [43] found that a TAK1 null mutation (TAK1179) in the kinase domain affected both pathways. Interestingly, the effect of TAK1179 could be moderately rescued by overexpressing the JNK kinase Hemipterous [43]. Moreover, loss of function mitotic clones in the larval fat body of JNK pathway components downstream of TAK1 abolished expression of a dipt-lacZ marker [43]. One explanation for this difference could be that TAK1K142R and TAK1K156R may be important for the ubiquitination and activity status of TAK1 in regards to IMD pathway only-in contrast to TAK1179, which inactivates the kinase function presumably important for both IMD and JNK. However, ref 14 as well as this study conducted experiments in S2 cells, which limits analysis compared to the whole organism. An additional caveat of our experiments in cell culture was that they were conducted by transient transfection and overexpression of the relevant proteins in contrast to [14], which used stable cell lines. Transferring our TAK1 Lys142 and Lys156 mutants in flies and in a TAK1 mutant background will be able to solve this issue.
In our ubiquitination assays of dTAK1, a major band is seen at ∼84 kDa (coinciding with the molecular weight of monoubiquitinated dTAK1) and is visible when blotted with the antibody specific for Ub (Figure 2A; Figure 3A). Therefore, this band is likely to be the monoubiquitinated form of dTAK1. A similar band is observed in humans [26], [27]. Hence, it appears that a significant amount of total ubiquitinated dTAK1 is monoubiquitinated. Overexpression of dTAK1 alone is sufficient to activate the IMD pathway, as seen in previous studies [14] and in our qPCR assays. Even when expressed alone, Drosophila TAK1 immunoprecipitation shows ladder-like bands, indicative of mono and polyubiquitination (Figure 6C, see first lane). Therefore, we can assume that dTAK1 ubiquitin modification occurs even when overexpressed alone (with the aid of endogenous components).
Like K63 polyubiquitination, monoubiquitination is also implicated in degradation-independent functions including protein kinase activation, DNA repair, membrane trafficking and chromatin remodelling [reviewed in 44]. It was shown that chronic phosphorylation of IKKβ at Ser-177/Ser-181 leads to monoubiquitin attachment at nearby Lys-163, which in turn modulates the phosphorylation status of IKKβ during chronic inflammation [45]. Similarly, constitutive over-expression of dTAK1 may lead to its chronic activation and monoubiquitination. This may explain why singly overexpressed dTAK1 appears to have the same migration as a possibly monoubiquitinated form.
Trbd negatively regulated IMD signalling and reduced (the essential for activity) K63-linked ubiquitination of dTAK1, presumably through its DUB function. In the absence of infection, flies deficient in Trbd had a significant increase in steady-state expression of dipt (an IMD target) in their whole body. In addition, they exhibited induction over and above the wild type threshold following gut infection. This was not a consequence of delayed pathogen clearance as bacteria were cleared in trbd flies as fast as in wild type flies following both systemic and oral infection. Finally, the life span of trbd flies was dramatically reduced in the absence of immune challenge. This phenomenon became even more pronounced when another negative regulator of the IMD-mediated response (pirk) was also absent. This indicated the importance of negative regulators in shaping the immune reaction and highlighted the tight constraints, which IMD signalling is under inside the cell from the level of the receptor PGRP-LC through to Imd itself and the IIK complex. This reduction in life span was suppressed in a dredd mutant background indicating that it was the chronic over-activation of IMD signalling that was compromising long-term survival. Interestingly, the endogenous flora did not influence this phenomenon since germ-free pirk; trbd flies showed indistinguishable survival compared to normally reared pirk; trbd flies. This result showed that reduction in life span was flora-independent in contrast to mutations in PGRP-LB, affecting extracellular recognition [18] or in big bang, which disrupts septate junctions between gut cells [46]. Results from these in vivo experiments are consistent therefore, with the hypothesis that Trbd acts inside the cell to suppress Imd signalling.
Our work in cell culture suggests that this suppression happens at the level of dTAK1 attenuating the IMD signal and helping to maintain a transient and tightly regulated immune response. Both C518 (the putative catalytic cysteine) and the 3 tandem NZF domains of dTrbd were required for its function. However, both the qPCR data (Figure 7B) and ubiquitination assays (Figure 7C) indicated that while dTrbd significantly reduced dTAK1 K63-linked ubiquitination and subsequent signalling, it did not completely abrogate it, leaving the possibility that it functions either indirectly or in a partially redundant manner with another DUB. An alternative scenario however, may implicate the in vivo existence of unknown cofactor(s), which regulate Trbd and whose concentration was limiting in our cell culture over-expression experiments. Such co-factors have been described for A20 [reviewed in 47]. More work is needed to distinguish between these possibilities.
Contrary to the mammalian model where mutations in the ZnF domain of hTAB2 result in failure to activate TAK1, our results show that mutations in the C-terminal ZnF domain of dTAB2 provoke a significant enhancement of IMD signalling. Nevertheless, genetic evidence in Drosophila has shown unequivocally that loss of TAB2 leads to loss of TAK1 signalling capacity and therefore loss of the IMD signal [19]. However, our results point to a more refined relationship between the two proteins since the ZnF domain of TAB2 seemed to moderate TAK1 activity by restricting its signalling capacity. Hence, it is improbable that the ZnF domain of dTAB2 binds to and brings in the E3 ligase, which activates TAK1, as is the case in humans. One plausible scenario would be that dTAB2ZnFDel binds more strongly to Trbd than wild type dTAB2 (as suggested by Co-IP in Figure 7D) thus keeping Trbd away from its target (TAK1). An alternative explanation would be that dTAB2ZnFDel interacts more with TAK1 than wild type dTAB2 thus stabilising TAK1 and increasing its signalling capacity. However, we did not observe such an increase in our Co-IP experiments (Figure S9).
Our working model is that of a tripartite relationship involving dTAK1, dTAB2 and Trabid. TAB2 is needed to activate TAK1 but through its ZnF domain it modulates TAK1 signalling and the TAB2-Trbd interaction. The latter is important for turning down the immune branch of TAK1 signalling and thereby the IMD pathway during gut epithelia responses and keeping IMD in check systemically in the absence of infection. Mutations in the dTAB2 ZnF domain enhance the TAB2-Trabid interaction and result in a more stable TAK1 presumably by keeping away Trbd from its target. An alternative scenario would be that through its ZnF domain TAB2 recruits an additional protein. This hypothesis would predict the presence of a protein that would act in concert with Trbd sharing some of its characteristics (e.g. DUB activity). More work is needed to distinguish between these two possibilities.
Drosophila S2 cells (Invitrogen) were maintained at 25°C in Schneider's Drosophila Medium (BioWhittaker/Lonza), supplemented with 10% heat-inactivated FBS and antibiotics 100 U/ml penicillin G and 100 µg/ml streptomycin sulfate – all Invitrogen). Cells were transfected with 2 ug plasmid DNA using Effectene Transfection Reagent (Qiagen) according to the manufacturer's protocol. Empty pAc5.1/HA-His vector was used to ensure equal amounts of DNA were delivered in each transfection.
Cells were lysed 48 hrs post-transfection in RIPA buffer (Sigma-Aldrich) supplemented with Complete Mini Protease Inhibitor Cocktail tablets (Roche Applied Science) and Benzonase Nuclease (Sigma-Aldrich). Cell lysates were incubated rocking with 50 µl of Anti-V5 or Anti-HA Agarose Affinity Gel (Sigma-Aldrich) for 2 hours at 4°C. Antibody beads were pre-blocked in RIPA buffer supplemented with 0.2% BSA (NEB) at 4°C for 2 hrs. Immunoprecipitates were washed with 600 µl CoIP wash buffer 900 (50 mM Tris-HCl [pH 8.0], 900 mM NaCl, 5 mM EDTA [pH 8.0], 0.5% Igepal CA-6030) four times for 10 minutes each at room temperature, followed by a final wash with 600 µl CoIP wash buffer 150 (50 mM Tris-HCl [pH 8.0], 150 mM NaCl, 5 mM EDTA [pH 8.0], 0.5% Igepal CA-6030). Immunoprecipitates were eluted in 1X SDS sample buffer, resolved on 10% SDS-PAGE and transferred to PVDF membranes. Blots were probed with mouse anti-V5 (1∶5000, Invitrogen), mouse anti-c-Myc peroxidise (1 µg.ml−1.Roche Applied Science; Clone 9E10) or rat anti-HA antibodies (200 ng.ml−1, Roche Applied Science; Clone 3F10). Between probing with a different antibody, blots were stripped with Restore PLUS Western Blot Stripping Buffer (Pierce).
|
10.1371/journal.pcbi.1007036 | Enhanced flux prediction by integrating relative expression and relative metabolite abundance into thermodynamically consistent metabolic models | The ever-increasing availability of transcriptomic and metabolomic data can be used to deeply analyze and make ever-expanding predictions about biological processes, as changes in the reaction fluxes through genome-wide pathways can now be tracked. Currently, constraint-based metabolic modeling approaches, such as flux balance analysis (FBA), can quantify metabolic fluxes and make steady-state flux predictions on a genome-wide scale using optimization principles. However, relating the differential gene expression or differential metabolite abundances in different physiological states to the differential flux profiles remains a challenge. Here we present a novel method, named REMI (Relative Expression and Metabolomic Integrations), that employs genome-scale metabolic models (GEMs) to translate differential gene expression and metabolite abundance data obtained through genetic or environmental perturbations into differential fluxes to analyze the altered physiology for any given pair of conditions. REMI allows for gene-expression, metabolite abundance, and thermodynamic data to be integrated into a single framework, then uses optimization principles to maximize the consistency between the differential gene-expression levels and metabolite abundance data and the estimated differential fluxes and thermodynamic constraints. We applied REMI to integrate into the Escherichia coli GEM publicly available sets of expression and metabolomic data obtained from two independent studies and under wide-ranging conditions. The differential flux distributions obtained from REMI corresponding to the various perturbations better agreed with the measured fluxomic data, and thus better reflected the different physiological states, than a traditional model. Compared to the similar alternative method that provides one solution from the solution space, REMI was able to enumerate several alternative flux profiles using a mixed-integer linear programming approach. Using this important advantage, we performed a high-frequency analysis of common genes and their associated reactions in the obtained alternative solutions and identified the most commonly regulated genes across any two given conditions. We illustrate that this new implementation provides more robust and biologically relevant results for a better understanding of the system physiology.
| The recent advances in omics technologies have provided us with an unprecedented abundance of data spanning genomes, global gene expression, and metabolomes. Though these advancements in high-throughput data collection offer an excellent opportunity for a more thorough understanding of metabolic capacities of a wide range of species, they have caused a considerable gap between “data generation” and “data integration.”. In this study, we present a new method named REMI (Relative Expression and Metabolomic Integrations) that enables the co-integration of gene expression, metabolomics and thermodynamics data as constraints into genome-scale models. This not only allows the better understanding of how different phenotypes originate from a given genotype but also aid to understanding the interactions between different types of omics data.
| The turnover rates of metabolites through a pathway are called fluxes, and genome-wide intracellular metabolic fluxes are the ultimate regulator of cellular physiology. Perturbations on the normal physiology, such as those that occur in a disease state, directly influence the metabolic fluxes. The well-established experimental approach for determining these metabolic fluxes is 13C metabolic flux analysis, though this experimental technique that directly measures metabolite levels is costly and time-consuming, such that computational tools for flux prediction have become a very popular alternative. Genome-scale metabolic models (GEMs), which essentially associate an organism’s genotype with its phenotype, integrate genomic information with known information about metabolite levels to comprehensively describe an organism's metabolism [1]. These models can predict metabolic fluxes, growth rates, or the fitness of gene knockouts using constraint-based approaches, which mainly require the knowledge of network stoichiometry that is available from the annotated genome sequences and metabolic pathway databases. One of the most routinely used constraint-based approaches is flux balance analysis (FBA), which relies on the stoichiometry and optimization principles to predict the steady-state metabolic flux distribution according to an objective function in a given metabolic network [2]. Due to network complexity, FBA commonly results in a span of alternative optimal solutions indicating different flux distributions with the same objective value rather than a unique steady-state flux distribution profile, and then selects one of these solutions at random to present back to the user, which is a major limitation of this method. To remedy this, it has been shown that integrating additional layers of constraints, such as thermodynamics, can effectively reduce the overall solution space of feasible flux distributions in an organism to limit the number of alternative solutions [3, 4].
With the growing availability of high-throughput data for different organisms under a wide range of genetic or environmental perturbations, GEMs became popular because of their ability to incorporate omics data as additional regulatory constraints for FBA problems. Because GEMs associate a genotype with a phenotype, it is essential to understand that a single genome can result in thousands of different physiologies through different regulatory mechanisms. Therefore, the integration of static snapshots of the metabolism, obtained from transcriptomic and metabolomic data, provides more biologically relevant constraints for the system and helps to increase the precision of the flux prediction, therefore better deducing the observed physiology. However, despite the high number of methods that have been introduced in recent years for the integration of omics data into constraint-based metabolic models, the enhanced prediction of flux profiles using omics data, particularly in cases using multi-omics data, is still far from being resolved. Recently, these methods, their scopes, and limitations were extensively reviewed [5], and the authors concluded that using gene-expression data does enhance flux predictions, though they inferred that the accurate predictions of the physiology is not achievable with the available reviewed methods.
The existing methods for integrating gene-expression data into GEMs can be classified into two categories with the first relying on the integration of absolute gene-expression data into GEMs. This includes techniques such as gene inactivity moderated by metabolism and expression (GIMME) [6] and the use of continuous and discrete formulations to find a flux distribution that is consistent with given context-specific gene-expression data, including integrative metabolic analysis tools (iMAT) [7, 8] [5, 9–11]. However, the assumption that absolute gene-expression data can be directly correlated with flux values is questionable and might not hold true for all genes. Moreover, these methods require user-defined thresholds to identify and categorize the expression levels of metabolic genes (high, moderate, or low expression), and the results are sensitive to the set thresholds. These drawbacks motivated the development of (ii) the second class of methods, which integrate the relative gene-expression data while aiming to maximize the correlation between differential changes in gene-expression and reaction fluxes. The underlying assumption for this class of methods is that the relative changes in gene expression between two conditions correlate with the resulting differential flux profiles [12, 13].
The increasing availability and quality of metabolomic data have promoted the development of methods that can be integrated into GEMs to refine model reconstruction, to reduce the solution space of feasible fluxes, and to better predict the physiological state of a system. These methods, their scope, and their limitations have been reviewed by Töpfer et al. [14]. One of these methods, thermodynamic-based flux balance analysis (TFA), integrates the absolute metabolite concentration data into GEMs, as the metabolite concentrations are intrinsically associated with the Gibbs free energy of metabolic reactions [3, 4]. Another available method is gene inactivation moderated by metabolism, metabolomics, and expression (GIM3E), an extension of the GIMME algorithm with added metabolomic data in addition to gene-expression data [15]. However, this method only considers the presence/absence of metabolites to refine the model, therefore preventing a full utilization of the quantitative metabolomic data. A time-resolved expression and metabolite-based prediction of flux values, named TERM-FLUX, integrates time-series expression and metabolomic data, and predicts flux distribution for a given time point t. [16]. However, the application of TERM-FLUX is limited to studies with time-series data, which are not widely available. More recently, a method for the integration of relative metabolite levels for flux prediction, iReMet-flux, has been introduced to predict differential fluxes at the genome-scale [17], and it requires an assessment of the differential changes of all existing metabolites in a GEM. This limits its application, as metabolomic data are mostly measured not at a genome-wide level but rather for only a few metabolites in a system.
For multi-omic data, methods have recently been introduced for integrating different layers of data, such as genomic, transcriptomic, proteomic, and fluxomic, into metabolic models [18] or multi-scale models [19]. However, a method that couples the thermodynamic constraints into GEMs with relative transcriptomic and metabolomic data is not yet available.
To address this deficiency, we herein propose a novel method, termed Relative Expression and Metabolite Integration (REMI), to integrate relative expression and relative metabolite abundance data into thermodynamically curated GEMs. REMI is the first method that integrates thermodynamics together with relative gene-expression and metabolomic data as constraints for FBA. We demonstrate that REMI’s ability to integrate different layers of constrictive data significantly reduces the solution space of feasible fluxes. REMI also extensively enumerates alternative optimal and sub-optimal solutions, bringing a robustness and flexibility to the flux distribution analysis. We applied REMI to a GEM of E. coli to estimate the central carbon metabolism flux measurements that were determined by 13C metabolic flux analysis (13C-MFA) and were provided by two independent experimental studies [20, 21]. Although there is limited number of such fluxomic datasets for validation and the measurements are not available at a genome-wide level, our results suggest that the integration of gene-expression, metabolite abundance, and thermodynamic data within REMI’s optimization framework allows for improved flux predictions. Comparing REMI’s predictions with a similar method (GX-FBA [12]), we also show that REMI has on average a 32% higher Pearson correlation coefficient (r = 0.79) indicating a more precise exploration of organismal metabolism under wide-ranging conditions.
We designed REMI as the first method to integrate relative gene-expression and metabolite abundance data into thermodynamically curated GEMs, reducing the solution space of optimal fluxes to provide results that are better at predicting cell physiologies closer to the experimental observations than can be reached using existing methods. The REMI workflow along with an illustration of the method performed on a toy model is presented in Fig 1. REMI requires a GEM (FBA model) and sets of gene-expression and/or metabolomic data. The first step consists of data pre-processing wherein the FBA model is converted to a thermodynamic-based flux analysis (TFA) model [3] that incorporates the Gibbs free energy of metabolites and reactions into the model. The gene-expression/metabolite-level ratios are further systematically converted into reactions ratios to integrate them into the REMI methods. Based on the type of integrated data, there are three different REMI methods. REMI-TGex integrates thermodynamic and gene-expression data, REMI-TM integrates thermodynamic and metabolomic data, and REMI-TGexM integrates thermodynamic, gene-expression, and metabolomic data into an FBA model. Note that the REMI methods can be used without thermodynamic data, such as in REMI-Gex, which integrates gene-expression data into a FBA model.
The REMI framework was applied to integrate the E. coli transcriptomic and metabolomic data obtained from two studies under 8 [20] and 3 [21] different conditions into the thermodynamically curated E. coli GEM iJO1366 and to estimate the differential steady-state fluxes. We call the data and information from [20] “Dataset A” and data and information from [21] “Dataset B”. We formulated different optimization models which hierarchically integrated different combinations of available data to investigate the effectiveness of multi-omics data integration in reducing the metabolic flexibility of the provided solutions. REMI-TGex is an integrated model obtained by incorporating relative gene-expression data into a thermodynamically constrained model, which is represented by iJO1366 in this work. Furthermore, we integrated relative metabolite concentration data into the REMI-TGex model to produce REMI-TGexM and compared experimentally measured fluxes with the steady-state flux prediction results of REMI-TGex and REMI-TGexM. We also compared our prediction results with those of the previously existing GX-FBA [12], though as this method does not employ thermodynamic constraints, we used REMI to incorporate gene-expression data lacking thermodynamics constraints into E. coli GEM iJO1366 (REMI-Gex). The comparison of REMI-Gex and REMI-TGex highlights the significance of the thermodynamic constraints in reducing the solution space of flux analysis. We also performed some studies with only metabolite changes with thermodynamic constraints (REMI-TM) and without thermodynamic constraints (REMI-M).
The underlying assumption of the REMI method is that the perturbation of gene-expression and metabolite levels influences the flux levels in the metabolic network. To this end, REMI maximizes the consistency between relative experimentally observed changes in gene expression and metabolite abundance with the flux levels (the objective function of the REMI constraint-based method). The maximum consistency is then calculated as an integer number, called the maximum consistency score (MCS). This represents the maximum number of constraints that can be incorporated into a FBA model from a given set of constraints (gene-expression or metabolite abundance levels) while ensuring that the model still achieves the required metabolic functionalities and remains feasible. MCS is a unique number, however, in that the complex nature and interconnectivity of metabolic networks can result in several alternative solutions for a given MCS, meaning that numerous combination of different constraints from the input data could result in the same MCS. The theoretical maximum consistency score (TMCS) indicates the number of genes (or metabolites or both) with available experimental data that can potentially be integrated into the model, and MCS indicates the number of these available constraints that could be consistently integrated into the model.
We first applied REMI to the integration of eight datasets from Ishii et al. [20], which included genome-wide transcriptomics together with some metabolomic data obtained for one reference condition and seven different conditions or mutations, into an E. coli model. After integrating the gene-expression data of each condition into the model and comparing it with the reference model, we computed TMCSs, MCSs, and the number of alternative solutions for the REMI-Gex method (without thermodynamic constraints) and the REMI-TGex method (with thermodynamic constraints). In contrast to other methods, REMI finds all possible alternative solutions of a given maximum consistency score, which involves all possible combinations of the given set of constraints that always result in a feasible model. These alternative solutions provide flexibility in the biological interpretation of the results as they are equally consistent with the provided experimental data (applied as constraints to the model). Note that in the GEM analysis, the alternate flux solutions are conventionally considered as equivalent phenotypic states [22]. In this study, however, alternative solutions represent the equivalent states of the maximum consistency between gene-expression (or metabolite abundance or both) data and the flux levels. Therefore, each feasible alternative solution provides an opportunity to analyze and interpret the given phenotypic state based on the condition-specific omics data, from a different standpoint.
We further integrated the available metabolomics measurements into the E. coli model using REMI-TGexM and obtained the MCS for the integrated metabolites as well as the global maximum consistency score (GMCS), which encompasses both genes and metabolites (Table 1). Although different pairs of conditions showed a very close TMCS variability across the seven case studies based on gene-expression data integration (mean = 103.6, standard deviation [sd] = 0.5) and on metabolic data (mean = 4.7, sd = 0.5), the MCS significantly varied across the four REMI methods: REMI-Gex (mean = 58.7, sd = 4.1), REMI-TGex (mean = 49.7, sd = 3.5), REMI-GexM (mean = 63.3, sd = 4), and REMI-TGexM (mean = 54.3, sd = 3.3) (Table 1). As Table 1 shows, the gene-expression and metabolite abundance constraints for the deregulated metabolites were consistent across the conditions. Therefore, the REMI-TM and REMI-GexM consistency scores sum up to the REMI-TGexM consistency score. This means that there is no conflict between the gene-expression and metabolite abundance data and that they can be co-integrated without confronting each other. We found in our results that MCS was noticeably lower than TMCS, suggesting that the assumption that the relative changes in gene expression correlate with those in fluxes does not always hold. This is probably because the relative changes in gene expression depend on the mechanism of post-transcriptional and translational processes which are currently not captured in metabolic models. However, the number of enumerated alternative solutions highly differs across the conditions in all four methods: REMI-Gex (mean = 80.6, sd = 80.3), REMI-TGex (mean = 104.6, sd = 168.5), REMI-GexM (mean = 156.1, sd = 241.8), and REMI-TGexM (mean = 25.1, sd = 26.4) (Table 1), which suggests that the numbers of alternative solutions are condition-specific, as expected. As shown in the Table 1, wherever the sd is very high, for example sd = 241.8 in REMI-GexM for the rpe vs Ref case, we observe a high number of alternative solutions (n = 735 in this case). Different conditions (mutations) alter the cell metabolism differently, leading to different levels of metabolic adaptations and metabolic flux rerouting. Hence, we speculate that the differences in flux rerouting across conditions results in differences in the numbers of alternative solutions across the seven relative conditions. Note that for REMI-TM and consequently for REMI-M, the constraints for all the deregulated metabolites were consistently integrated into the model, so we found only one solution for the REMI-TM models without any alternative solution.
For the metabolomic integration, the GMCS was higher in the REMI-TGexM models compared to REMI-TGex because in REMI-TGexM, the GMCS was computed based on both relative metabolite (Table 1; Metabolites) and relative gene-expression levels (Table 1; Genes), whereas the MCS for the REMI-TGex model was computed based on only relative expression levels. We further investigated the consistency between gene-expression and metabolomic data and whether the data contradicted each other in certain scenarios. All the available experimental metabolomic data (Table 1; TMCS and MCS) were integrated using the REMI-TGexM method for the pgm vs Ref, gapC vs Ref, zwf vs Ref, wt5 vs Ref, and wt7 vs Ref comparisons. We observed that the number of alternative solutions for these five cases was identical between REMI-TGexM and REMI-TGex. This implies that the relative expression constraints and the relative metabolite constraints were not contradictory for these five cases. However, in rpe vs Ref and pgi vs Ref, all the metabolic data were integrated in the model, but the number of alternative solutions differed (and in the case of rpe vs Ref was noticeably reduced) between REMI-TGexM and REMI-TGex. To see if this indicated a contradiction, further investigation into the alternative solutions revealed that in the rpe vs Ref and wt5 vs Ref comparisons, REMI-TGexM and REMI-TGex have the same set of constraints, which means that the constraints from metabolomics and expression data were not contradictory. However, we found that the metabolomics integration resulted in a reduction in the number of alternative solutions (Table 1). We hypothesized that further integration of metabolomics (on the top of the gene-expression constraints) imposed a flux rerouting in the metabolic network.
As REMI allows enumerating all the possible alternative solutions for a given consistency score, we further interrogated the alternative solutions by High-frequency constraint (HFC) analysis.
The results of this analysis indicate the core constraints that consistently operate in all the alternative solutions (the constitutive part of all solutions). Meaning that such core constraints certainly perturb fluxes within each pair of conditions. Therefore, these constraints could potentially be the indicators of the regulators of the condition-specific metabolism, which assist biologist in determining which metabolic subsystems to deregulate or to mutate. We believe that the capability to analyze and identify these regulators is a key advantage of REMI.
As shown in the Table 1, the computed HFCs differ across conditions for all four cases: REMI-Gex (mean = 52, sd = 5.4), REMI-TGex (mean = 44.4, sd = 4.5), REMI-GexM (mean = 56.9, sd = 5.4), and REMI-TGexM (mean = 49.6, sd = 4.2). Constraints that were common amongst all the alternative solutions, indicating key regulators, were the potential candidates for further investigations. After analyzing HFCs across conditions and between the four cases, we found that a reaction catalyzed by glycolate oxidase (GLYCTO4) from the alternate carbon metabolism and another reaction from the murine recycling pathway (MDDEP4pp) were always deregulated in the pgm, gapC, zwf, rpe, pgi, and wt7 conditions. These reactions are likely key regulators of mutation in E. coli because they were found to be deregulated in all mutant conditions.
To study the effect of thermodynamics on the model, we compared the reduction in solution space for the predicted flux profiles from the REMI-TGex and REMI-Gex methods when coupled with the gene-expression data (Table 1). The MCS was consistently reduced in the REMI-TGex model compared to REMI-Gex for all pairs of conditions, as REMI-TGex eliminates flux solutions that are not thermodynamically feasible.
To better illustrate the positive influence of thermodynamic constraints in reducing the solution space, we show the example of pgm vs Ref as a case study, where we obtained MCS = 56 in REMI-Gex and MCS = 49 in REMI-TGex (Table 1). First, we enforced the models to satisfy any given consistency score (56 and 49 in this example) by adding a new constraint, which would further allow us to perform conditional FVA. Then, we performed the FVA that satisfies the consistency score (MCS = 56) in REMI-Gex and the consistency score (MCS = 49) in REMI-TGex. Comparing the FVA results of REMI-Gex and REMI-TGex revealed that there exist 45 reactions in REMI-Gex that operate in a thermodynamically infeasible direction and which also contribute to the MCS = 56. The flux ranges of these reactions are shown in S1 Table and indicate that the TGex method is indeed eliminating the infeasible solutions to enrich for more relevant results. For more clarification, two reactions out of the 45 are shown as examples in S1 Fig. As expected, the flux ranges for these reactions are less flexible for the REMI-Gex (MCS = 56) compared to the REMI-TGex (MCS = 49), which confirms some extent of the thermodynamic infeasibility in the REMI-Gex predictions as infeasible flux ranges directly indicate the model infeasibility. On the other words, if we integrate thermodynamic constraints to the model and allow the consistency score (MCS = 56) then the model certainly generates infeasible solutions. To investigate whether the higher consistency score caused thermodynamic infeasibility in the REMI-Gex, we performed a FVA of REMI-Gex while forcing lower consistency scores (MCS = 49 and 10). We found that the flux ranges of reactions became more flexible at lower consistency scores in the REMI-Gex model compared to the REMI-TGex model (S1 Fig), indicating that if both REMI-TGex and REMI-Gex have the same consistency scores, the REMI-Gex cannot allow thermodynamic infeasibility. In contrast, if the consistency score is higher in the REMI-Gex compared to the REMI-TGex, then it leads to thermodynamic infeasibility. The same results were obtained for all other reactions (S1 Table).
To further benchmark REMI with the available experimental data, we used a second data set (2 overexpression compared to the ref condition) from an independent study where the role of metabolic cofactors, such as NADH and ATP in different aspect of metabolism is studied by overexpressing NADH oxidase (NOX) and the soluble F1-ATPase in E. coli [21]. REMI integrated the gene-expression data from Holm et al. [21] into the E. coli model, and a summary of the results is shown in Table 2. Like the previous analysis, we observed a reduction in the MCS value within REMI-TGex as compared to REMI-Gex, as REMI-Gex satisfies fluxes that were not thermodynamically feasible. The number of alternative solutions highly differs between NOX overexpression and ATPase overexpression for both REMI-TGex and REMI-Gex, which is likely due to the condition-specific regulations (NOX vs ATPase overexpression) that do not necessarily involve the same set of deregulated genes.
To investigate the influence of thermodynamic constraints on flux ranges, we identified the overlapping constraints (HFCs) across all the alternative solutions and then enforced them to be active to build the most consistent model. An active HFC satisfies differential gene expression (or metabolite levels) between two conditions form a given experimental data. Thus, for each condition, we built the most consistent model despite having many alternatives. We next performed FVA on the REMI-Gex and REMI-TGex models. As REMI is based on pair-wise relative constraints (for two conditions) and builds two models that are then compared, as opposed to modifying one solution based on a given condition, we obtained two FVA solutions, i.e. one for each condition. We identified less bidirectional reactions (BDRs) in the REMI-TGex case compared to the REMI-Gex case (Table 3), which means that thermodynamic constraints reduce the solution space and consequently the number of BDRs. This is consistent with the fact that thermodynamic constraints eliminate infeasible reaction directionalities. The number of BDR reductions differs across conditions, and we identified the highest BDR reduction for the rpe vs. Ref case and the lowest BDR reduction for the NOX vs. Ref case, which therefore indicates more reduction in the feasible flux solution space in the rpe vs. Ref case compared to the NOX vs. Ref case. For the all comparisons, we found a further reduction in BDRs upon the integration of relative metabolomic data into the REMI-TGex model. In most of the cases, we found a similar decrease in BDRs, which means that the metabolomic data further constrained the solution space. Except for the wt7 vs Ref case, we observed a decrease in BDRs for all cases that were constrained by metabolites and expression data together (GexM) as compared to only expression (Gex) data. Unexpectedly and unlike all the other cases, by incorporating metabolomics data for the wt7 vs Ref case, we found an increase of one reaction in the BDRs. This suggests that for the wt7 vs Ref case the integration of gene expression and metabolites reroutes fluxes through the metabolic networks differently compared to other cases. As expected, we consistently find a reduction in BDRs for the REMI-TM model (thermodynamics and relative metabolomics) compared to without thermodynamics (the REMI-M model). This is in agreement with the fact that integrating thermodynamic constraints into a model eliminates infeasible reaction directionalities and consequently the flux feasible optimal space.
To further illustrate the positive influence of thermodynamic constraints in reducing the optimal solution space, we performed a relative flexibility (Materials and Methods) analysis using the REMI-TGex and REMI-Gex methods. To perform a relative flexibility analysis, a reference model is compared to a target model to investigate the relative flux reduction. For a reference, we used the iJ01366 model without integrating any data, meaning that the reference model implies only mass balance constraints. We took the pgi vs. Ref case as an example to demonstrate the average relative flexibility (ARF) reduction at a global (e.g. all reactions) level as well as at the subsystem level.
For the pgi vs. Ref case, we found a 10%, 20%, 50%, 77%, and 80% reduction in the global ARF in REMI-M, REMI-Gex, REMI-TGex, REMI-GexM, and REMI-TGexM models compared to the reference model, respectively (Fig 2A). We found 40% and 80% more reduction in the global ARF for the REMI-TGex and REMI-TGexM models compared to REMI-Gex (Fig 2A), which was expected as the REMI-TGex and REMI-TGexM models are more constrained by thermodynamic and metabolomic data compared to REMI-Gex. We further analyzed the ARF at the subsystem/pathway level to investigate the reduction in ARF for each specific subsystem using the REMI-TGex and REMI-Gex methods. Consistently, each subsystem for the REMI-TGexM and REMI-TGex models was more reduced than REMI-Gex (Fig 2B). For REMI-TGex and REMI-TGexM, we observed a remarkable ARF reduction in the glycerophospholipid metabolism, lipopolysaccharide biosynthesis, murein recycling and biosynthesis, and the biomass and maintenance function subsystems. We further performed the same analysis for the pgi vs. Ref, rpe vs. Ref, pgm vs. Ref, wt5 vs. Ref, wt7 vs. Ref, NOX vs. Ref, and ATPase vs. Ref data (S2 Fig). We found a similar reduction in ARF for REMI-TGex and REMI-TGexM compared to REMI-Gex for the cases of gapC vs. Ref and zwf vs. Ref and found a small reduction in pgi vs. Ref and rpe vs. Ref (S2 Fig). We identified a remarkable reduction in ARF (more than 90%) across all the comparisons using the REMI-TGexM method for the glycerophospholipid metabolism, murein recycling, and lipopolysaccharide biosynthesis/recycling subsystems (S2 Table). This suggests that these subsystems are more perturbed based on our available gene-expression and metabolite level data, which indicates that they might be key regulator pathways for the studied mutations.
To demonstrate the efficacy of the REMI methods in reducing the solution space and therefore predicting flux profiles close to the experimental measurements, we compared the flux predictions of the REMI-Gex, REMI-TGex, and REMI-TGexM methods with those of the alternative, previously used GX-FBA method and compared both methods to the available experimental measured fluxes from 13C experiments. In both GX-FBA and REMI methods, it is assumed that the differential changes in gene expression correlate with flux changes. GX-FBA uses a linear optimization to address the problem, while the REMI method employs mixed-integer linear optimization that enables enumerating alternative states of the maximum consistency. Additionally, unlike GX-FBA method, REMI allows the integration of thermodynamics and relative metabolite abundances. To implement the GX-FBA method, we integrated the relative gene-expression datasets into the iJO1366 model using GX-FBA and computed the flux distributions. For the comparisons, we computed two metrics: 1) the Pearson correlation between the predicted and measured intracellular fluxes, and 2) the average percentage error (see Materials and Methods) between the measured and predicted fluxes. A good prediction requires a noticeable correlation and a small average percentage error.
The results of the first set of experimental data [20] (pgm vs. Ref, rpe vs. Ref, zwf vs. Ref, wt5 vs. Ref, and wt7 vs. Ref) showed a considerably improved flux prediction for the REMI-Gex, REMI-TGex, REMI-TGexM, and REMI-GexM models as compared to the GX-FBA method, indicated by Pearson correlation and average percentage error (Fig 3A). The GX-FBA and REMI-Gex methods predicted a similar flux correlation for the experimental fluxes for the pgi vs. Ref and gapC vs. Ref cases (Fig 3A). For the second set of experimental data [21] (Nox vs. Ref and ATPase vs. Ref), REMI-TGex predicted better correlation than REMI-Gex and GX-FBA, and the average percentage error of GX-FBA was higher than that of REMI-TGex and REMI-Gex (Fig 3B). On average, across all nine comparisons (excluding references) we found that the REMI-Gex method has 32% higher Pearson correlation coefficient compared to the GX-FBA method, which indicates a remarkable improvement in the flux prediction. Since the REMI methods use an additional objective that is the minimization of the sum of fluxes (see Materials and Methods), we modified GX-FBA to imply the minimization of the sum of fluxes as an objective in order to perform an unbiased comparison. This modified GX-FBA prediction agreed less with the experimental results than the REMI predictions (S3 Fig), meaning that REMI outperforms GX-FBA in terms of predictions. REMI also has two advantages over GX-FBA and other relative expression methods in that, first, we do not need to estimate a reference flux distribution a priori, because two flux distributions for two different conditions are obtained in the same optimization framework in REMI (see Materials and Methods), second, REMI enumerates alternative solutions at the MCS, providing a higher confidence when investigating and analyzing results. Generating two separate flux distributions for the two compared conditions allows REMI to be more suitable to study the differential flux analysis between two conditions, and the extensive enumeration of alternative solutions provides robustness and flexibility in the biological interpretations of the provided data.
Although all REMI methods were in relative agreement with the experimental fluxomic measurements, we did not observe a significant difference in the predicted results of REMI-Gex, REMI-TGex, and REMI-TGexM. However, as the fluxomic measurements were very limited around the central carbon metabolism, we cannot draw any overarching conclusions about the accuracy of REMI from these results, as this could only indicate that the major fluxomic differences occur in pathways outside of this one specific metabolic pathway. We believe that to investigate the differences in flux predictions across REMI methods, fluxomic and metabolomic measurements will be required on a grander scale, such as the genome level.
Eleven total sets of experimental data that had been previously integrated into the genome-scale model (GEM) of E.coli by Kim et al. [23] and were originally obtained from two independent studies done by Ishii et al. (8 datasets) [20] and Holm et al. (3 datasets) [21] were used for the evaluation of the REMI methodology.
The three datasets from Holm et al. [20] comprise genome-wide transcriptomic data together with fluxomic data (21 measured fluxes) collected from three experimental conditions: wildtype E. coli, cells overexpressing NADH oxidase (NOX), and cells overexpressing the soluble F1-ATPase (ATPase). The eight datasets from Ishii et al. [20] include genome-wide transcriptomic, fluxomic (31 measured fluxes), and metabolomic (42 metabolites) data obtained under eight different experimental conditions: wildtype E. coli cells cultured at different growth rates of 0.2, 0.6, and 0.7 per hour along with single-gene knockout mutants of the glycolysis and pentose phosphate pathway (pgm, pgi, gapC, zwf, and rpe).
All analyses were performed using IJO1366, the latest GEM of E. coli [24]. The model comprises 2,583 reactions, 1,805 metabolites, and 1,367 genes. The REMI code is implemented in Matlab R2016a, and it is available on GitHub at https://github.com/EP-LCSB/remi. Mixed-integer linear programming (MILP) problems were solved using the CPLEX solver on an Intel 12-core desktop computer running Mac.
It has been previously shown that thermodynamic constraints not only effectively reduce the solution space of FBA by eliminating the thermodynamically infeasible fluxes from the solution space, but also allow the integration of metabolite concentrations. This provides important links between mass and energy balance and the phenotypic characteristics of the organism. The thermodynamic constraints, as depicted in Eq (1), were integrated into the IJO1366 model [3]. The standard Gibbs free energy ΔrGi° without corrections for the pH and ionic strength was estimated using the group contribution method [25].
For each reaction of a GEM, the Gibbs free energy of the reaction (ΔrGi′) was computed, which considers the charge and the activity (xj) of each metabolite j given the pH, the metabolite concentration range, and the ionic strength at the cellular compartment where the reaction occurs.
We used the gene-protein-reaction (GPR) association rules acquired from the E. coli GEM to translate the relative gene-expression levels (being relatively up- or downregulated) to the differential and relative flux values of corresponding reactions. GPRs are not mapped as one gene to one reaction, meaning there are many cases in which one gene is mapped to several reactions and multiple genes are mapped to a single reaction, which are depicted with “and” and “or” affiliations, respectively.
To capture this, we followed the same procedure for the mapping between GPRs and reactions fluxes introduced by Fang et al [13]. In that study, the authors showed that using a geometric mean of expression ratios where several genes are jointly required for a complex reaction to occur, is the most efficient way to capture the condition that all reaction ratios are required. For the isoenzyme case when any of the several potential genes are sufficient to carry out the reaction, the arithmetic mean of reaction ratios of the genes is suggested, as it captures the minimum condition where any of the reaction ratios is required. However, other types of assumptions were used by other methods for mapping GPRs to fluxes [7, 12, 15, 23, 26]. For example, minimum expression value is used for complex reactions and maximum or sum of expression values are used for isoenzyme case [23, 26].
In REMI, if the reaction R is associated with two genes (g1 “and” g2), the expression level ratios for genes g1 and g2 in the two corresponding conditions are calculated to obtain the geometric mean of the g1 and g2 ratios. Whereas, if the reaction R is associated with two genes (g1 “or” g2), the arithmetic mean of the obtained expression data ratios is calculated. Thus, from GPR associations, REMI computes the so-called tentative “reaction flux ratios” to further constrain the model. For the metabolomic data, the ratio of metabolite concentration for each metabolite (if available) is calculated for any two given conditions.
To evaluate whether a reaction or metabolite was up- or downregulated, we sorted the ratios (calculated as explained in the previous section), and selected the top 5% as upregulated and the bottom 5% as downregulated. The fold changes greater than two are considered as significant in many studies. For comparison purposes, we used the two-fold change as the cut-off threshold to identify the significant gene expression and metabolite changes. We found that across all mutants, the set of significant changes identified with our threshold of the top and bottom 5% encloses the corresponding set identified with the two-fold change, meaning that the proposed cutoff criterion is conservative. However, this threshold is a user-defined parameter and one could use different threshold cutoff.
For a given metabolic network that includes R reactions and M metabolites, bidirectional reactions are decomposed into forward and backward reactions to allow all fluxes to have positive values. Assuming that S is a stoichiometry matrix, Smr is the stoichiometric coefficient associated with the metabolite m (m = 1, …, M) in reaction r (r = 1, …, R). Positive and negative stoichiometric coefficients of metabolites signify the substrate or products of a reaction. A binary variable zr was assigned to each reaction r to ensure a positive flux vr (Eq (2)) through the reaction r, and when zr = 0, there was no flux. An additional constraint was formulated using Eq (3) to ensure that only one reaction directionality could be active and carry flux. α and β indicate the forward and reverse directions of a reaction.
In REMI, two models are described for each given condition. For both models, we constrained the cellular growth rate to be at least 0.1 mmol g-1 DW-1 h-1, to ensure that the model is able to synthesize all the biomass building blocks required for the cellular growth. Throughout the manuscript, the terms “wildtype” and “mutant” are used to better differentiate between the two conditions (or models) when describing the REMI framework. REMI can, however, be used for any two given conditions and is not restricted to the wildtype and mutant labels. Eq (4) specifies the mass balance constraints for the wildtype and mutant conditions at the steady state.
The relative information about the gene-expression levels or metabolite levels between the two given experimental conditions was formulated as additional constraints and integrated into the two representative models of the conditions. To do this, binary variables for the up- and downregulated reactions were assigned as u and d, respectively, where n is the total number of up- and downregulated reactions. For the upregulated reactions, a higher flux was enforced in the mutants as compared to the wildtype, while for downregulated reactions, a higher flux was enforced for the wildtype as compared to the mutant.
For u upregulated and d downregulated reactions, a total of n binary variables were generated (B1, …Bi, …Bn), where Bi = 1 indicates the up- or downregulation of a reaction. Next, n constraints (Eqs (6 and 7)) were added to enforce a basal flux in both the wildtype and mutant conditions. For u upregulated reactions, constraints (Eq (8)) were added to ensure a mutant flux could be higher (p*vrwild) than a wildtype flux, where p is a reaction ratio between the wildtype and mutant (computed from gene-expression ratio). Constraints were added (Eq (9)) for d downregulated reactions that ensured a mutant flux was lower compared to a wildtype flux. In Eq (10), n constraints were added to form the boundary for the slack variables that are used in Eqs (8) and (9), where ε = 10−5,M′ = 1000.
In GEMs, gene-level perturbations can mediate both reactions and their subsequent metabolites. Available studies show a correlation between gene changes and metabolite changes and infer that perturbations at the metabolite level are formed from perturbations in genes or reaction levels [27, 28]. Thus, if experimental evidence shows remarkable changes in a given metabolite abundance level across two conditions, the assumption is that there is an imbalance in the incoming or outgoing fluxes around that metabolite.
If the experimental data indicates that a metabolite is upregulated, it is assumed in REMI that either the sum of production ϕp in condition 2 is greater than the ϕp in condition 1 or the sum of consumption ϕc in condition 2 is less than the ϕc in condition 1 (Fig 4B). Due to mass balance, ϕp and ϕc will be equal.
In Eq (11), the sum of production and of consumption of a metabolite i is shown, where the metabolite is produced by reactions 1 and 2 and is consumed by reactions 3 and 4 (Fig 4A).
Based on available experimental measurements of metabolite abundance, REMI finds the total number (n’) of up- and downregulated metabolites, where u’ and d’ are up- and downregulated metabolites, respectively. For an up-regulated metabolite i (i.e. in the mutant vs. wildtype), either more production or less consumption is enforced in the mutant compared to the wildtype using Eqs (12) and (13). In Eq (12), a binary variable (Bi) is introduced, which switches to production if Bi = 1 and to consumption if Bi = 0. Similarly, for downregulated metabolites i, less production or more consumption is enforced in the mutant compared to the wildtype (Eqs (13) and (14), see supplementary description for more detail).
Based on the assumption that alterations in gene-expression or metabolite levels within two different physiological conditions results in differential flux profiles, REMI defines such alterations as constraints and integrates them accordingly into the two metabolic models corresponding to the two conditions. However, as additional constraints reduce the solution space of FBA, particularly in the case of multi-omic data integration, the resulting models might not be feasible. Therefore, the objective function (Eq (15)) was formulated in such a way as to obtain feasible models with a maximum agreement between the relative expression and metabolite levels and their corresponding constraints. Eq (15) maximizes the agreement with experimental data using mathematical optimization principles subject to Eqs (5)–(10), where n is the total number of up- and downregulated reactions. The maximum consistency score (MCS) is the sum of the binary variables (Eq (15)) in the outcome of the optimization that is formulated in REMI.
An aforementioned mathematical optimization model (objective function (15) subject to Eqs 1–9) allows us to maximize the total number of consistent reactions between the differential gene-expression or metabolite levels with the differential flux profiles between two models and to obtain a maximum consistency score (MCS). Depending on the flexibility of the model, many alternative flux distribution profiles for a given MCS, and subsequently MCS-n, are possible. MCS and MCS-n represent optimal and suboptimal consistency, respectively. To enumerate alternative solutions, integer cut constraints (Eq (16) [29] were used as follows:
∑i=1nB′iBi≤(∑i=1nB′i)−1
(16)
The left-hand side of Eq (16) determines the number of up- and downregulated reactions in the current solution that carries fluxes in the first MCS solution. The right-hand side represents the number of reactions that carry fluxes in MCS-1. The inequality ensures that the new solution differs at least by one new reaction that carries flux compared to the previous solution. Repeating this procedure allows the enumeration of alternative solutions for each MCS.
To concurrently integrate both the relative gene-expression data and the relative metabolite levels, an integrated mathematical optimization model was built with a global objective function (Eq (17) subject to a combined set of constraints, i.e. Eqs (5)–(14). This optimization model was then solved to maximize the objective, which is the combined consistency score of the two sets of constraints.
GlobalConsistencyscore(GCS)=Maximize∑i=1n+n′Bi,
(17)
where n and n’ represent a total number of up- and downregulated metabolites and up- and downregulated genes, respectively.
To compare the REMI-predicted fluxes with the experimentally measured ones, predicted flux distribution profiles were required. To obtain such predicted flux profiles, all the alternative solutions at MCS were first enumerated. REMI method optimizes consistency and identifies alternative sets of consistency. Then, for each consistency set we build a model by fixing binary variables which enforces constraints are applied in the model. Then, an additional optimization was performed by minimizing the sum of the fluxes for each alternative solution to obtain a representative flux profile for benchmarking REMI against the experimental flux measurements.
To effectively compare the predicted in silico fluxes from REMI with the corresponding 13C-determined in vivo intracellular fluxes, the following two metrics were used: the uncentered Pearson correlation coefficient (Eq 18), and the average percentage error in predicted fluxes (Eqs (18)–(20)). The uncentered Pearson correlation is a good metric for the flux comparison, as fluxes are usually not centered, and it has been used for comparing two flux vectors [23].
In Eq (18), vi and vm are the in silico and measured vectors of the fluxes, respectively. The correlation coefficients +1 and -1 indicate a strong positive and negative linear relationship between vi and vm, and the 0 correlation coefficient indicates no linear relationship between vi and vm.
The average percentage error has been used in the GX-FBA method [12] to compare two fluxes. In Eq (19), the dr is used to measure the relative deviation between the two fluxes in two conditions, where x and y correspond to the flux of a given reaction in condition 1 and condition 2, respectively. Since |dr| lies between 0 and 1, one can consider dr as a percentage flux change from condition 1 to condition 2. The average (per reaction) percentage error, e, in the predicted in silico fluxes was calculated using Eq (20), where diinsilico and diexp indicate relative deviation in predicted in silico flux using methods such as REMI and GX-FBA, and experimentally measured flux and N represent the number of reactions with available experimental flux data.
For a given system, the FBA results in a solution space of optimal flux profiles, and the magnitude of this solution space indicates the metabolic flexibility of the system. The integration of the thermodynamic knowledge of reactions as well as condition-specific experimental data, e.g. gene-expression or metabolomic data, constrains the metabolic system to a less flexible one. Thus, the solution space and the subsequent range of the metabolic responses are reduced. Comparing and quantifying the relative flexibility of a metabolic system before and after constraint is a decent indication of the effectiveness of the data integration [30]. Performing a flux variability analysis (FVA) outlines the flux variability range of each reaction in the system for the two conditions as follow:
FRi1=[vmin,i1,vmax,i1]
(21)
FRi2=[vmin,i2,vmax,i2]
(22)
The relative flexibility (RF) for reaction i is calculated using the following equation:
RFi=[(vmin,i2−vmax,i2)/(vmin,i1−vmax,i1)]
(23)
where FRi1 and FRi2 represent the flux variability range of reaction i at each of the two conditions, one condition is usually designated as a reference condition or reference state, such as when comparing the relative flexibility of a metabolic system with (condition 1) and without (condition2) thermodynamic constraints. The value of RF that is computed for each reaction i reflects the relative changes in the flux variability range of one condition compared to the other condition. The global relative flexibility change between two given condition is then computed by averaging the Fi values for each reaction i that carry flux in the reference state.
We developed the computational tool, REMI, which combines gene-expression, metabolomics, and thermodynamics constraints with the mass balance constraints imposed in metabolic models to predict phenotypic changes in an organism upon environmental or genetic perturbations. As the integration of these three additional physiological constraint results in a highly reduced flexibility of the predicted optimal flux profiles, REMI enhances the quality of the computationally predicted fluxes. REMI’s novel formulation permits the extensive enumeration of alternative solutions because there exist several alternative sets of pathway that result in the same phenotype due to the complexity and interconnectivity of metabolic networks, meaning that the results provided by REMI more accurately reflect natural biological states than previously existing methods. Within several examples, we showed the effectiveness of incorporating thermodynamic data with gene-expression and metabolomics in reducing the flexibility of predicted optimal flux profiles. This means that we can obtain manageable set of physiological consistent hypothesis and physiological interpretations which have a higher confidence as they are consistent with a larger set of data. Applying REMI to experimental data has shown that there is not always a full consistency between gene-expression and metabolomic data, which shows that there is still much to learn about how gene expression and metabolism are linked.
The application of REMI goes beyond the study of physiology of a mutant versus a wild-type cell presented in this work. With a slight modification in the formulation, REMI can be employed for investigating the physiology of several mutants simultaneously against the wild type physiology within a single optimization. Although in this study we showcased REMI for constraining internal fluxes, REMI can also be applied to study the perturbation of external fluxes and metabolites whenever omics data are available. This has a potential application in studying the overflow metabolism, e.g., the acetate overflow in fast-growing E.coli or the Warburg effect in the cancer cells. Furthermore, REMI can also be used to investigate metabolism of diseased states compared to the healthy one, where numerous sets of omics data are available.
Various REMI methods introduced in this work permit a wide range of applications depending on the type of available data (thermodynamics, single or multi omics data). However, whenever gene-expression, metabolite abundance, and thermodynamic data are available, our results suggest that the most extensive data integration method, REMI-TGexM, provides the best results and the most reduced optimal solution space. As systematic multi-omics integration remains a challenge, REMI opens the possibility of not only multi-omics integration, but also the identification of the crosstalk between the various omics present in a system.
|
10.1371/journal.ppat.1006462 | Human hantavirus infection elicits pronounced redistribution of mononuclear phagocytes in peripheral blood and airways | Hantaviruses infect humans via inhalation of virus-contaminated rodent excreta. Infection can cause severe disease with up to 40% mortality depending on the viral strain. The virus primarily targets the vascular endothelium without direct cytopathic effects. Instead, exaggerated immune responses may inadvertently contribute to disease development. Mononuclear phagocytes (MNPs), including monocytes and dendritic cells (DCs), orchestrate the adaptive immune responses. Since hantaviruses are transmitted via inhalation, studying immunological events in the airways is of importance to understand the processes leading to immunopathogenesis. Here, we studied 17 patients infected with Puumala virus that causes a mild form of hemorrhagic fever with renal syndrome (HFRS). Bronchial biopsies as well as longitudinal blood draws were obtained from the patients. During the acute stage of disease, a significant influx of MNPs expressing HLA-DR, CD11c or CD123 was detected in the patients’ bronchial tissue. In parallel, absolute numbers of MNPs were dramatically reduced in peripheral blood, coinciding with viremia. Expression of CCR7 on the remaining MNPs in blood suggested migration to peripheral and/or lymphoid tissues. Numbers of MNPs in blood subsequently normalized during the convalescent phase of the disease when viral RNA was no longer detectable in plasma. Finally, we exposed blood MNPs in vitro to Puumala virus, and demonstrated an induction of CCR7 expression on MNPs. In conclusion, the present study shows a marked redistribution of blood MNPs to the airways during acute hantavirus disease, a process that may underlie the local immune activation and contribute to immunopathogenesis in hantavirus-infected patients.
| Inhalation of hantavirus-infected rodent droppings can cause a wide range of disease ranging from mild symptoms to deaths in humans. Central to hantavirus disease is vascular leakage that can manifest in different organs, including the lungs. Although the virus can infect endothelial cells lining the blood vessels, it does not cause cell death. Instead, activation of the immune system in response to viral infection has been implicated in causing vascular leakage. In this study, we investigated how monocytes and dendritic cells (DCs) are involved in hantavirus disease, given their capacity to activate other immune cells. We obtained unique clinical material from 17 Puumala virus-infected patients including mucosal biopsies from the airways as well as multiple blood draws over the course of disease. In the airways of these patients, we observed an infiltration of monocytes and DCs. In parallel, there was a dramatic depletion in peripheral blood—more than ten-fold—of monocytes and DCs that was sustained throughout the first two weeks of disease. Taken together, this study provides novel insights into immune mediated processes underlying human hantavirus pathogenesis.
| Hantaviruses pathogenic to humans are rodent borne, but do not cause disease in their natural hosts. However, transmission to humans via inhalation of aerosolized virus-contaminated rodent excreta may lead to severe disease and death, thus representing a severe threat to public health [1, 2]. Hantaviruses in Europe and Asia primarily cause hemorrhagic fever with renal syndrome (HFRS) whereas hantaviruses in the Americas cause hantavirus pulmonary syndrome (HPS), with case fatality rates of 0.1–10% and 40% respectively [3]. Puumala virus (PUUV), the endemic strain in Sweden, has an incubation time of 2–3 weeks and can cause a mild form of HFRS, also referred to as nephropathia epidemica [2, 4, 5]. In humans, hantaviruses infect the vascular endothelium without causing cytopathic effects [6]. Yet, increased vascular permeability is a hallmark of hantavirus diseases. It has been suggested that an immune-mediated dysregulation of endothelial permeability might contribute to disease pathogenesis [1, 3, 7–9]. Hantavirus immunopathogenesis is most likely a complex multifactorial process involving both innate [10–12] and adaptive immune cells [13–15]. Cytotoxic T lymphocytes (CTLs) and natural killer (NK) cells as well as pro-inflammatory cytokines such as tumor necrosis factor (TNF) produced by these lymphocytes have been implicated in causing capillary leakage [16]. Supporting this notion, stronger CTL responses have been associated with a more severe disease outcome and even death [14, 17–20].
Monocytes and dendritic cells (DCs), together termed mononuclear phagocytes (MNPs), are able to present viral antigens to T cells, thus initiating and regulating virus-specific immune responses [21, 22]. In human blood, monocytes can be further subdivided into classical, intermediate and non-classical monocytes based on varying expressions of CD14 and CD16 [23]. During both bacterial and viral infections, intermediate and non-classical monocytes in blood of patients have been reported to increase in numbers [24–27]. Kwissa et al. further illustrated in acute dengue virus infection that the expansion of intermediate monocytes correlated with formation of plasmablasts, important for development of humoral responses [27]. Indeed, a robust production of hantavirus-specific plasmablasts in circulation, as well as IgG and IgM antibodies in serum may be necessary for patient recovery and even survival [28–30].
DCs, which are superior to monocytes in priming naïve T cell responses, consist of plasmacytoid DCs (PDCs) and myeloid DCs (MDCs) that can be further separated into CD1c+ MDCs and CD141+ MDCs [31, 32]. During viral infections, DCs are rapidly mobilized to peripheral tissues where they replenish the tissue-resident DCs that first encounter the pathogens and either die due to infection or migrate to draining lymph nodes [33–35]. In humans, both monocyte-derived cells as well as DCs have been observed in respiratory compartments at steady state with the capacity to detect and respond to invading pathogens [31, 36–39]. Since hantaviruses are transmitted predominantly via inhalation, studying immunological events in the airways where viral replication is first initiated is of importance to understand the processes leading to immunopathogenesis. Furthermore, pulmonary dysfunction has been reported in HFRS patients [40–42]. Expansion of cytotoxic CD8+ T cells in the airways of hantavirus-infected patients has been described as contributing to disease severity [13, 14]. This suggests that DCs may be involved in promoting the recruitment and local activation of T cells. However, little is known on the contributions of MNPs in human hantavirus-infected patients in vivo, especially at the site of entry.
In order to investigate the involvement of monocytes and DCs in local immune events in the airways, we obtained endobronchial biopsies from 17 PUUV-infected HFRS patients during the acute phase of disease and compared them to samples from uninfected controls (UC). We illustrated a significant infiltration of CD8+ T cells and MNPs into the bronchial tissue during acute HFRS compared to UC. As hantaviruses establish a systemic infection, we characterized MNPs in longitudinal peripheral blood samples from the same patients. Concurrent with the increase of MNPs in the airways, we observed a dramatic depletion of circulating monocytes and DCs during the acute phase of disease. By investigating the distribution of monocytes and DCs in different anatomical compartments during an acute viral infection in humans, we gained insights into the potential roles of specific cell subsets based on their migratory patterns and tissue-specific locations during the course of disease.
Although HFRS mainly manifests in the kidneys, respiratory involvement has been increasingly documented in these patients, including pulmonary edema that may lead to respiratory failure [14, 28, 40]. Of the 17 PUUV-infected HFRS patients included in this study, 10 experienced respiratory symptoms such as dry cough and dyspnea, and 5 needed oxygen treatment (S1 Table). Given that the airways are the initial site of infection after inhalation of hantavirus-contaminated rodent excreta, little is known regarding the early immune response taking place locally. In the current study, bronchoscopy was performed on PUUV-infected HFRS patients in order to sample their airways during the acute phase of disease. As soon as platelet counts had stabilized and the patients were able to withstand the procedure, endobronchial biopsies and bronchoalveolar lavage (BAL) were collected from each patient (median 9 days after onset of symptoms). In addition, longitudinal peripheral blood samples were collected from these patients during both the acute phase (2–14 days after onset of symptoms) and convalescent phase (>15 days after onset of symptoms) of HFRS. Results from blood and lung samples collected from HFRS patients were compared throughout the study with samples from UC, who similarly underwent bronchoscopies and blood draws (Fig 1A). In 15 patients, viral load was detected in BAL cells, suggesting local viral replication. However, viral load alone could neither explain the respiratory symptoms experienced by 10 patients, nor the need for oxygen treatment by 5 patients (Fig 1B). As CTLs have been implicated in contributing to hantavirus pathogenesis [13, 14, 16, 43], we investigated the absolute numbers of CD8+ T cells in the airways of patients. As previously described within the same study cohort [14], more CD8+ T cells were detected in BAL of patients that required oxygen treatment when compared to those who did not need oxygen treatment (p<0.05) (Fig 1C). Granzyme B, a cytotoxic protease released by CTLs and NK cells that can induce apoptosis, was also detected in BAL fluid of patients, with a trend towards higher amounts of granzyme B in patients requiring oxygen treatment (n = 5), although the difference was not significant (p = 0.08) (Fig 1D), possibly due to the relatively large variation between individuals. In order to assess whether CD8+ T cells were also present in bronchial tissue of patients, we performed immunohistochemistry on sections of endobronchial biopsies (Fig 1E). Indeed, significantly more CD8+ T cells were detected in biopsies taken from patients with acute HFRS than in biopsies from UC (Fig 1F). This prompted us to examine aspects of local T cell activation by investigating monocytes and DCs in the airways during hantavirus disease.
By immunohistochemistry, we investigated the distribution and frequency of cells expressing surface markers for monocytes and DCs on sections of endobronchial biopsies. We found significantly more HLA-DR+ cells (p<0.01) in biopsies taken from patients with acute HFRS than in biopsies from UC (Fig 2A and 2B). The increased HLA-DR staining, especially in the lamina propria, suggested an infiltration of MNPs. Additionally, the pulmonary epithelium also displayed increased HLA-DR staining in tissues from acute HFRS patients, possibly from local inflammation leading to upregulation of HLA-DR on epithelial cells [44] in addition to infiltration of immune cells to the site of infection. We also observed a significant increase in the number of cells expressing the myeloid cell marker CD11c during acute HFRS compared to UC (Fig 2C). Detailed analysis of bronchial tissue showed significantly increased numbers of CD11c+ cells in the lamina propria (p<0.05) and epithelium (p<0.05) of hantavirus-infected patients compared to UC (Fig 2D). In addition, the number of cells expressing the PDC marker CD123 was also significantly higher in the lamina propria (p<0.05) of these patients (Fig 2E and 2F). To address whether patients with high numbers of CD8+ T cells in the bronchial tissue also had high numbers of MNPs, we performed a Spearman correlation test and observed a positive association between CD8+ cells and CD11c+ cells (p = 0.08) (Fig 2G), and a significant correlation between CD8+ cells and CD123+ cells (p<0.05) (Fig 2H). Taken together, we observed an infiltration of MNPs into the airways during acute HFRS coinciding with the presence of CD8+ T cells at the site of infection. However, hantavirus infection is typically systemic and not limited to the airways [3]. Indeed, viral RNA copies can be detected in the plasma early during disease onset, but also on the day of bronchoscopy when the bronchial biopsies were obtained (Fig 2I and S1 Fig). Thus, we next explored the possibility that blood DCs and monocytes exposed to virus or virus-induced cytokines may have received signals to migrate into the airways, contributing to the significant influx of MNPs in the bronchi.
Since monocytes participate in inflammation [22], especially following viral infection, we hypothesized that the number of monocytes would expand in peripheral blood during the acute phase of HFRS, as has been observed in other acute viral infections [27, 45, 46]. To investigate how hantavirus infection may affect monocytes in circulation, the frequencies of classical (CD14+ CD16-), intermediate (CD14+ CD16+) and non-classical (CD14- CD16+) monocytes were analyzed using flow cytometry (Fig 3A and S2 Fig). The absolute number of classical monocytes per microliter of blood decreased significantly (p<0.001) during acute HFRS compared to samples from UC (Fig 3B and S2 Table), and subsequently normalized during the convalescent phase of the disease (S3 Table). Similarly, the numbers of intermediate monocytes and non-classical monocytes were also significantly reduced (p<0.001) during acute HFRS compared to UC (Fig 3B and S2 Table). Interestingly, for all the monocyte subsets observed, the low numbers of cells in circulation coincided with high viral load (viral RNA copies per mL of plasma) as assessed by quantitative reverse transcriptase polymerase chain reaction (qRT-PCR) (Fig 3C and S1 Fig). During convalescence, when virus was no longer detectable in plasma, the number of monocytes returned to comparable values as those in UC, except for intermediate monocytes. In summary, we observed a loss of monocytes in the peripheral blood of patients during the acute phase of HFRS.
The loss of monocytes in circulation during acute HFRS led us to also assess whether DCs, key determinants of viral disease outcome due to their capacity to initiate and activate T cell responses [47], would also be affected by hantavirus infection. We first analyzed the two MDC subsets found in human blood: CD1c+ MDCs and CD141+ MDCs (Fig 4A). We observed a dramatic reduction of both MDC subsets during acute HFRS as compared to UC (Fig 4A). The reduction in absolute numbers was statistically significant for CD1c+ MDCs (p<0.01) and CD141+ MDCs (p<0.001) (Fig 4B). On average, the number of CD1c+ MDCs was as low as 1.5 cells per microliter of blood during the early acute phase as compared to 16 cells per microliter found in UC (S2 Table). The number of CD141+ MDCs, already rare under steady state conditions, decreased by 97% during acute HFRS (S3 Table). For both CD1c+ and CD141+ MDCs, the cell numbers normalized during the convalescent phase of the disease (Fig 4B). Of importance, we excluded the possibility that the reduced number of DCs in acute HFRS blood samples was a consequence of the cells being more fragile to freeze-thawing, by confirming similarly low frequencies of DCs and monocytes in fresh samples from patients with acute HFRS (S3 Fig). As with the monocytes, we compared the kinetics of plasma viral load with the absolute numbers of MDCs in blood and found that high viral load coincided with low numbers of CD1c+ and CD141+ cells (Fig 4C).
We also assessed the numbers of PDCs in blood during acute and convalescent HFRS. PDCs are the major producers of antiviral type I interferon (IFN) in the body and are important in the defense against viral pathogens, despite their low frequency. Yet, levels of IFN-α are not elevated in blood of HFRS patients [48]. Here, we found that the number of blood PDCs, as defined by their CD123 and CD303 expression, was significantly reduced during acute HFRS as compared to UC (Fig 5A). The drop in absolute PDC number (p<0.001) was maintained also during early convalescence at days 15–21 after the onset of HFRS, but eventually returned to normal values (Fig 5B and S2 Table). Again, the loss of blood PDCs during acute HFRS coincided with high viral load (Fig 5C). Together, a massive depletion of both MDCs and PDCs was observed in peripheral blood during acute HFRS.
As hantaviruses are not known to cause cytopathic effects, the loss of monocytes and DCs in circulation could reflect a redistribution of MNPs from circulation into the airways, as we had observed an infiltration of MNPs in the bronchial tissue (Fig 2). To assess whether trafficking of blood DCs and monocytes to lymph nodes [49, 50] or other tissues [51, 52] could account for the reduced numbers of monocytes and DCs in blood during acute HFRS, we measured the surface expression of the chemokine receptor CCR7 on the few MNPs still in circulation. At steady state, few or no cells expressed CCR7, as exemplified by the UC (Fig 6A). However, during acute HFRS, a subset of the cells remaining in peripheral blood expressed CCR7 on their surfaces that progressively disappeared over time, as exemplified by CD1c+ MDCs (Fig 6A). Intermediate monocytes and non-classical monocytes presented with a higher frequency of CCR7+ cells in HFRS patients compared to UC (p<0.05) during acute HFRS (Fig 6B). CD1c+ MDCs (p<0.01) also upregulated CCR7 expression pattern throughout the acute phase, while the CD141+ MDCs showed no or very modest upregulation of CCR7 (Fig 6B). Although a subset of PDCs upregulated CCR7 during early acute HFRS (p<0.001), the frequency of CCR7+ PDCs returned back to low levels in the late stage of acute disease (days 11–14), at frequencies similar to those in UC (Fig 6B).
Although the overall maturation profile of monocytes and DCs in circulation as determined by upregulation of co-stimulatory molecules CD70 and CD86 was not pronounced, CCR7+ cells had a higher expression of CD70 (classical monocytes and CD1c+ MDCs) and CD86 (CD1c+ MDCs) than the CCR7- cells, consistent with a more mature phenotype (Fig 6C and 6D). Taken together, the data suggest that although the majority of monocytes and DCs are absent in circulation during acute HFRS, the cells that remain in blood appear to have received signals to upregulate migratory receptors such as CCR7, facilitating migration to lymph nodes or peripheral tissues.
Finally, we established an experimental system to address the loss of MNPs in blood of patients with acute HFRS. Classical monocytes and CD1c+ MDCs were isolated from blood of healthy volunteers (Fig 7A) and exposed to PUUV or UV-inactivated PUUV in vitro. We measured the frequency of cells expressing PUUV antigens over time by immunofluorescence staining using human anti-PUUV serum (Fig 7B). Forty hours post infection, 1.4% of classical monocytes and 0.4% of CD1c+ MDCs were infected by PUUV (Fig 7C), whereas PUUV antigen was undetectable in cells that were either uninfected or exposed to UV-inactivated PUUV. In addition to detecting viral proteins, PUUV RNA was detected by qRT-PCR in classical monocytes and CD1c+ MDCs exposed to replicating PUUV, at approximately 200-fold higher than in cells exposed to UV-inactivated PUUV (Fig 7D). However, no replicating viruses were detected in supernatants of infected cells, suggesting that PUUV replication is restricted in these cells (S4 Fig). Neither replicating nor UV-inactivated PUUV decreased the viability of these cells compared to uninfected cells (Fig 7E), typical of the non-cytopathic effects of hantavirus [1]. Interestingly, exposure to PUUV improved the viability of classical monocytes significantly at 40 hours (p<0.01) and 60 hours (p<0.001) (Fig 7E). When exposed to Hantaan virus (HTNV) that causes severe HFRS, we similarly observed that classical monocytes and CD1c+ MDCs were susceptible to infection in vitro, without the induction of cell death (S5 Fig).
Both classical monocytes and CD1c+ MDCs also responded to PUUV exposure by modulating their expression of chemokine receptors over time (S6 Fig). CCR2, important for mobilization of monocytes from bone marrow to peripheral tissues [22, 53], was downregulated on both classical monocytes (p<0.05) and CD1c+ MDCs after 40 hours of PUUV exposure (p<0.05) compared to uninfected controls (Fig 7F and 7G). CCR4 and CCR6, chemokine receptors that have been associated with tissue homing, were upregulated on classical monocytes (p<0.05) and CD1c+ MDCs respectively after 12 hours of PUUV exposure compared to cells that were not exposed to virus (Fig 7F and 7G). Both classical monocytes and CD1c+ MDCs (p<0.05) also responded to PUUV exposure by upregulating the migratory chemokine receptor CCR7 typically expressed by mature cells (Fig 7H and 7I). In line with their increased CCR7 expression, both classical monocytes and CD1c+ MDCs also upregulated the co-stimulatory molecule CD86 (p<0.05) upon exposure to replicating PUUV, but not to UV-inactivated PUUV (Fig 7F and 7G). In summary, both classical monocytes and CD1c+ MDCs were susceptible to PUUV infection in vitro, but infection did not result in cell death. The observed regulation of chemokine receptor expression on classical monocytes and CD1c+ MDCs upon exposure to PUUV in vitro provides a platform for further investigations into the molecular mechanisms governing the redistribution of MNPs observed in patients with acute HFRS.
In this study, we explored the involvement of monocytes and DCs during hantavirus infection by characterizing MNPs at the initial site of infection: the airways, where virions enter their human host upon inhalation of aerosolized hantavirus-containing rodent excreta. Viral RNA, as previously shown [14], can be detected in BAL cells of patients during acute HFRS. An expansion of cytotoxic CD8+ T cells in BAL has been shown to correlate with disease severity [14]. In addition to CD8+ T cells lining the airways as reflected by BAL sampling, we now demonstrate the presence of CD8+ T cells in the bronchial tissue of hantavirus-infected patients. In the same HFRS patients, high levels of HLA-DR+ or CD11c+ cells were observed in the bronchial biopsies, suggesting an influx of monocytes and/or MDCs. An increase in the number of CD123+ cells also suggests an infiltration of PDCs into the airways. The presence of monocytes and DCs in the airways might explain the observed increase in CD8+ T cells present in the airways during acute HFRS. During influenza virus infection in mice, recruitment of DCs to the lungs is necessary for mounting adaptive immune responses needed for efficient viral clearance [33–35]. Specifically, local interaction in the lungs between antigen-bearing DCs is required for protective CD8+ T cell responses [54]. A careful investigation of how the MNPs interact with CD8+ T cells in the airways of hantavirus-infected patients would facilitate understanding of whether MNPs contribute to pathogenesis or immunity, by activating or controlling CTL activity.
As hantavirus infection is systemic [3], we also characterized the absolute numbers of six distinct MNP populations in the blood of patients over the course of disease, from acute infection to convalescence. We found a depletion of all populations, especially MDCs, in the peripheral blood of PUUV-infected patients during the acute phase of HFRS, coinciding with the presence of viral RNA in blood. During acute HFRS, CCR7 was upregulated on several monocyte and DC populations, indicating a mobilization of cells from the blood toward lymph nodes or peripheral tissues. In vitro, our data further demonstrated that classical monocytes and CD1c+ MDCs were susceptible to PUUV infection and remained alive. Although Markotic et al. suggested a differentiation of monocytes into DC-like cells after hantavirus infection in vitro [55], we were not able to identify monocytes expressing DC markers by flow cytometry upon PUUV exposure. Our data corroborated earlier findings by Raftery et al. and Temonen et al. suggesting that human DCs and monocytes may contribute to pathogenesis: both monocytes and DCs remain alive after infection, potentially leading to viral dissemination due to their migratory properties [56, 57].
Hantavirus infection is not marked by a general loss of immune cells or leukopenia in circulation of patients, since these viruses do not cause obvious cytopathic effects [56, 58]. Instead, the numbers of NK cells in blood are expanded in PUUV-infected patients [11, 13, 59]. We anticipated that a similar expansion of MNPs would be detected in blood from our patients as has been reported in those with other acute viral infections, as both monocytes and DCs are mobilized from the bone marrow to partake in the innate response to a viral infection [27, 45, 46, 60, 61]. In contrast, we found a depletion of all monocyte and DC subsets in blood during acute HFRS. Although Tang et al. reported an expansion of intermediate monocytes in the blood of HFRS patients [62], no such increase was observed in the present study. The cause of this disparity could be technical due to our gating strategy that excluded all lineage+ and HLA-DR- cells. Alternatively, biological differences between the virus strains could yield differing results, as their patient cohort was infected with HTNV whereas our patients were infected with PUUV, the endemic hantavirus strain in Sweden. In line with our findings, the loss of DCs has been documented in patients with acute influenza A virus (IAV) and human immunodeficiency virus (HIV) infections, related to depletion and impaired function of DCs during acute infection [63–66].
While all DC populations in peripheral blood returned to normal values during the convalescent phase of HFRS, we noted that the absolute numbers of intermediate monocytes and non-classical monocytes remained low, even more than 100 days after onset of disease (S3 Table). Monocytes arise from bone marrow precursors, differentiating from classical monocytes via intermediate monocytes to non-classical monocytes in their lifetime [67, 68]. We speculate that in PUUV-infected patients, there may be a delay in the developmental progression of circulating classical monocytes, even during convalescence. The prolonged expansion of NK cells in the circulation of HFRS patients [11] could provide a source of IFN gamma, a cytokine known to activate classical monocytes to a more inflammatory phenotype [69]. Additionally, the emerging concept of trained immunity [70] suggests that monocyte precursors in the bone marrow may be epigenetically modified upon exposure to hantavirus such that they remain poised for future infections.
Our data from experiments performed in vitro suggest that even if blood MNPs were susceptible to the virus, hantavirus infection did not lead to cell death. A potential explanation for the stark depletion of DCs observed in blood could be that these cells have migrated out of circulation. The chemokine receptor CCR7 controls the homing of DCs to lymph nodes, where priming of T cells and initiation of adaptive immune responses can occur [49, 50, 71, 72]. Increased CCR7 expression on blood CD1c+ MDCs during acute HFRS could indicate that these cells have migrated to the lymph nodes, in response to either direct viral infection, as we could show in vitro, or to pro-inflammatory cytokines in serum of patients [73]. Specifically for hantavirus infections, activation of CD4+ T cells have been shown to be instrumental in viral control and improved clinical outcome [74].
By infecting monocytes and DCs in vitro, we have developed a platform for further dissection of the underlying mechanisms by which exposure to hantavirus determines cellular trafficking. For instance, the chemokine receptor expression may indicate where blood MNPs traffic to during acute HFRS. Accumulation of monocytes in the brain during West Nile virus infection has been related to expression of CCR2, which is important for the egress of monocytes from the bone marrow into tissue [60]. Detection of hantavirus RNA in the bone marrow of a patient [75] suggests that hantavirus could impede the release of monocytes into the bloodstream by downregulating expression of CCR2, as our data suggest. In other respiratory diseases such as chronic obstructive pulmonary disease (COPD), the increased presence of DCs in the lungs correlated with the upregulation of CCR6 on DCs and an increase of the CCR6 ligand (CCL20) in the airways [76]. In mice, expression of CCR4 on T cells imprints them to home to the lungs upon influenza infection [77]. However, these scenarios have not been carefully investigated in the homing of monocytes and DCs into the lungs during viral infection in humans.
In conclusion, blood monocytes and DCs were dramatically depleted during the acute phase of HFRS caused by PUUV. The high numbers of CD8+ T cells in the airways [14, 16], correlating with respiratory symptoms experienced by patients, may have been promoted by an infiltration of MNPs into the airways, as demonstrated in bronchial biopsies of hantavirus-infected patients in this study. As the viral load subsides in the blood, the numbers of blood monocytes and DCs also return to normal values. By establishing in vitro hantavirus infections of MNPs, the descriptive nature of patient data can be complemented in future in vitro studies to elucidate the mechanisms of how hantavirus infection can orchestrate the mobilization of monocytes and DCs from the blood into peripheral tissues such as the respiratory tract, and lymphoid organs. A better understanding on the role of monocytes and DCs during hantavirus infection is valuable in the development of immunomodulatory strategies to treat hantavirus-infected patients.
The study protocol was approved by the regional Ethical Review Board at Umeå University, Umeå, Sweden. Written informed consent was obtained from study subjects, all of whom were adults.
Peripheral blood, bronchoalveolar lavage (BAL) and endobronchial biopsies were prospectively obtained from 17 hospitalized PUUV-infected patients between 2008 and 2011. The criteria for study enrollment were described earlier [14]. Peripheral blood samples for flow cytometry analysis were collected during the acute phase (2–14 days after disease onset; median 6 days). Follow-up samples were taken throughout the first weeks of infection as well as the convalescent phase (>15 days after disease onset). Patients were monitored using qRT-PCR to assess plasma viral load until two consecutive measurements were negative (median 11 days) [78]. Briefly, viral RNA was extracted from plasma and cDNA was generated. qRT-PCR was performed in triplicate using primers designed based on PUUV RNA sequences. No fatal cases were observed in this study cohort. Twelve uninfected age- and sex-matched blood donors were included in this study. They underwent bronchoscopy for the collection of endobronchial biopsies and bronchoalveolar lavage (BAL) as well as peripheral blood draw. Standard clinical procedures, including differential cell counts, were used to obtain clinical data for all subjects used in this study (S1 Table) [79].
Bronchoscopy was performed on all patients 6 to 14 days (median 9 days) after onset of symptoms. Patients underwent bronchoscopy as soon as their clinical condition allowed them to withstand the procedure. This included absence of hypotension or hypoxemia as well as improvements of coagulation parameters to avoid bleeding. Bronchoscopy was performed as soon as possible and when blood platelet count was higher than 100 x 109. At that time, all patients were still in need of hospital care due to the acute infection. In brief, patients and UCs were treated with oral midazolam (4–8 mg) and intravenous glycopyrronium (0.2–0.4 mg) 30 minutes (min) before the bronchoscopy. For topical anesthesia, lidocaine was applied, and additional lidocaine was administered in the larynx and bronchi during the procedure. A flexible video bronchoscope (Olympus BF IT200) was inserted through the mouth via a plastic mouthpiece. From each patient, four to six endobronchial biopsies were taken from the main carina and the main bronchial divisions on the left side using fenestrated forceps (Olympus FB-21C). BAL was obtained with saline solution (3 x 60 mL) from the contralateral side. BAL samples were filtered through a 100 μm nylon filter (Syntab) and centrifuged at 400 x g for 15 min at 4°C.
Endobronchial biopsies were processed and embedded into glycol methacrylate resin (Polyscience), as previously described [80]. Sections from biopsies (2 μm) were stained in duplicates with anti-CD8, HLA-DR, CD11c, and CD123 (all BD Biosciences) followed by the rabbit anti-mouse (Dako) biotinylated secondary antibody. The immunostaining was performed as previously described [13]. All sections were visualized with 3-amino-9-ethylcarbazole (AEC), and cell nuclei were counterstained with Mayer hematoxylin (Histo Lab). Finally, all sections were analyzed using a high-resolution digital scanner, NanoZoomer-XR (HAMAMATSU) to convert them into digital images. A blinded analysis was performed using the scanned sections and NanoZoomer Digitial Pathway View2 software (NDP View; HAMAMATSU). The number of positive cells was expressed as cells/mm and cells/mm2 of epithelium and lamina propria, respectively. Quantification of HLA-DR molecules was carried out with a Leica DMR-X microscope (Leica Microsystems GmbH) coupled to computerized image analysis (Leica Qwin 5501W; Leica Imaging Systems) as described previously [81].
For isolation of peripheral blood mononuclear cells (PBMCs), whole blood from PUUV-infected patients and UC was collected in CPT tubes (BD) and centrifuged according to manufacturer’s instructions. The separated suspension of PBMCs was harvested and then washed in PBS. PBMCs were frozen in 90% human albumin (Octapharma), 10% DMSO (WAKO-Chemie Medical), and 50 IE heparin (LEO Pharma), and stored in liquid nitrogen for later analysis. Absolute numbers of all monocyte and DC subsets were calculated by using the absolute lymphocyte and monocyte counts obtained on the automated hematology analyzer and the percentages of events in each respective gate obtained from flow cytometry data.
Monocytes and primary CD1c+ MDCs were isolated from buffy coats obtained from Karolinska University Hospital (Stockholm, Sweden) as previously described [82]. Monocytes were isolated using the human monocyte enrichment kit (RosetteSep; StemCell Technologies) according to the manufacturer’s instructions. The blood was diluted, carefully layered on Ficoll-Paque PLUS (GE Healthcare Biosciences) and centrifuged for 20 min at 1800 x g at room temperature. For isolation of human CD1c+ MDCs, magnetic labeling using CD1c+ MDC isolation kit (Miltenyi Biotec) was used on enriched populations of monocytes. Monocytes and MDCs were cultured in RPMI1640 (Sigma-Aldrich) with 10% fetal bovine serum (FBS), 1% penicillin/streptomycin and 1% L-Glutamine (all Invitrogen). Cells were cultured at 1x106 cells per mL of complete medium. MDCs were additionally supplemented with 2 ng/mL GM-CSF (R&D Systems).
PUUV strain Kazan and HTNV strain 76–118 were propagated on Vero E6 cells (ATCC Vero C1008) as previously described [48]. The virus stocks were titrated on Vero E6 cells for calculations of multiplicity of infection (MOI). UV inactivation of hantaviruses was performed for 25 seconds using a VL215G Vilber Lourmat UV lamp (Torcy), as a negative control for productive infection. Cells were exposed to medium alone (uninfected), infected with hantaviruses or exposed to UV-inactivated hantaviruses at an MOI of 7.5 for 2 hours (h). Cells were then washed and incubated for 12–60 h at 37°C. Supernatants were collected after centrifugation and were stored at −80°C until further analysis. Cells were stained for flow cytometric analysis.
Cell suspensions were stained with Live/Dead Aqua fixable dead cell stain kit (Invitrogen) to exclude dead cells. Non-specific binding was prevented by adding FcR blocking reagent (Miltenyi Biotec) followed by surface staining with conjugated Abs (S5 Table). Briefly, cells were stained for 15 min at 4°C in FACS buffer (PBS with 2% fetal bovine serum) and fixed in 1% paraformaldehyde (PFA). For chemokine receptor staining, cells were stained for 15 min at 37°C prior to addition of cell surface antibodies for another 15 min at RT. For HTNV-infected cells, intracellular staining with anti-nucleocapsid protein (N) antibody (B5D9, Progen) was assessed using a standard protocol. In brief, cells were stained for surface markers, fixed and permeabilized using Transcription Factor Staining Buffer Set (eBioscience). Cells were analyzed by flow cytometry using a LSRII instrument or LSRFortessa (both BD Biosciences) and data were analyzed using FlowJo X software (Tree Star).
At 60 h post infection, RNA from DCs and monocytes infected with PUUV in vitro was isolated using 450 μl TriPure Isolation Reagent (Roche Diagnostics). The relative levels of PUUV RNA in PUUV-infected DCs and monocytes was assessed using a qRT-PCR assay, as previously described [83]. β-actin mRNA levels were measured in parallel, using a commercially available TaqMan gene expression assay (4333762; Applied Biosystems). The expression of PUUV S segment RNA was calculated against the housekeeping gene β-actin: 2-[Ct(PUUV gene)-Ct(B-Actin)].
At 40 h post infection, classical monocytes and CD1c+ myeloid DCs were adhered for 20 min on Alcian blue-coated coverslips at 100 000 cells per condition. Cells on the coverslips were gently washed in PBS and fixed with pre-warmed 4% paraformaldehyde for 20 min at room temperature. Cells were then blocked with PBS containing 1% normal goat serum and permeabilized with 0.1% Triton-X 100 (Sigma) and stained with human anti-PUUV serum for 1 hour. Secondary antibodies against human IgG conjugated to Alexa Fluor 488 were used. Additionally, CD1c+ MDCs were co-stained with anti-HLA-DR conjugated to Alexa Fluor 647. Coverslips were mounted on glass slides with Prolong Diamond Antifade mountant with DAPI (Molecular Probes). Confocal images were acquired on a Zeiss LSM700 using a 10x objective. PUUV+ cells were enumerated out of 1000–2000 cells per condition using FIJI ImageJ software (NIH).
Levels of granzyme B in BAL fluid were measured using the commercially available Human Granzyme B ELISA kit (Abcam).
For all patient data generated ex vivo, mean cell counts of monocyte and DC populations were modeled using Poisson regression with patient-specific random intercept and robust standard errors. The proportions of CCR7+ CMs, IM, NCM, CD1c+ MDCs, CD141+ MDCs, and PDCs were modeled using logistic regression with patient-specific random intercept. Random intercepts were used to account for the potential dependence among repeated blood measurements over time. Number of days since symptoms' onset in HFRS patients was the predictor of interest and was categorized as acute phase (2–14 days) or convalescent phase (>15 days). UC served as the reference group. Correlations were analyzed using Spearman’s rank correlation coefficient. For in vitro experiments, statistical significance was assessed using paired t-test. Comparisons for IHC data are by Mann–Whitney U-test. Data were analyzed using GraphPad Prism version 6.0 (GraphPad Software) and Stata version 14.1 (StataCorp, College Station, TX). All the reported p-values are two-sided and p-values<0.05 was considered statistically significant.
|
10.1371/journal.ppat.1006464 | Helicobacter pylori gene silencing in vivo demonstrates urease is essential for chronic infection | Helicobacter pylori infection causes chronic active gastritis that after many years of infection can develop into peptic ulceration or gastric adenocarcinoma. The bacterium is highly adapted to surviving in the gastric environment and a key adaptation is the virulence factor urease. Although widely postulated, the requirement of urease expression for persistent infection has not been elucidated experimentally as conventional urease knockout mutants are incapable of colonization. To overcome this constraint, conditional H. pylori urease mutants were constructed by adapting the tetracycline inducible expression system that enabled changing the urease phenotype of the bacteria during established infection. Through tight regulation we demonstrate that urease expression is not only required for establishing initial colonization but also for maintaining chronic infection. Furthermore, successful isolation of tet-escape mutants from a late infection time point revealed the strong selective pressure on this gastric pathogen to continuously express urease in order to maintain chronic infection. In addition to mutations in the conditional gene expression system, escape mutants were found to harbor changes in other genes including the alternative RNA polymerase sigma factor, fliA, highlighting the genetic plasticity of H. pylori to adapt to a changing niche. The tet-system described here opens up opportunities to studying genes involved in the chronic stage of H. pylori infection to gain insight into bacterial mechanisms promoting immune escape and life-long infection. Furthermore, this genetic tool also allows for a new avenue of inquiry into understanding the importance of various virulence determinants in a changing biological environment when the bacterium is put under duress.
| Helicobacter pylori is a bacterial pathogen that chronically infects half the global population and is a major contributor to the development of peptic ulcers and stomach cancer. H. pylori has evolved to survive in the stomach and one important adaptation is the enzyme urease. The bacteria cannot establish an infection in the host without this enzyme, and although widely postulated, the requirement of urease for chronic infection of the host has not been tested experimentally as conventional urease mutants are incapable of colonization. To overcome this constraint, a genetic system was introduced that allowed for the making of H. pylori strains in which urease expression could be turned off after the bacteria have colonised the stomach. We show for the first time that this enzyme is not only important for initial colonization but that it is also very important for maintaining chronic infection. We also show that if urease is turned off, the bacterium can mutate several different genes in order to restore urease expression. The genetic approach described here opens up opportunities to studying genes involved in the chronic stage of H. pylori infection to gain insight into how the bacterium is able to avoid clearance by the immune system and how it is able to adapt to changing biological environments.
| The human gut pathogen Helicobacter pylori has coevolved with humans over thousands of years to dominate the gastric niche [1–3]. The majority of infected individuals (80–90%) carry and transmit H. pylori without any symptoms of disease [4, 5]. However, H. pylori infection causes chronic active gastritis that may develop into peptic ulceration (10–20%) or gastric adenocarcinoma (0.5–2%) [6, 7] causing a significant burden on public health [8–10]. H. pylori infection is persistent and clinical disease usually develops after many years of chronic inflammation and epithelial damage. Furthermore, due to increasing rates of antibiotic treatment failure [11, 12] there is a pressing need for further research into the bacterium’s mechanisms for persistence and immune evasion strategies. These are of particular importance to understanding H. pylori pathogenesis and to identifying novel targets for the development of new treatment options.
H. pylori is highly adapted to colonizing and surviving in the harsh conditions of the gastric environment. One key adaptation is the virulence factor urease. This multimeric enzyme, consisting of 12 UreA and UreB heterodimers, catalyses the hydrolysis of urea to produce CO2 and NH3, which acts to buffer the acidity of the local environment around the cell [13, 14]. Urease is abundantly expressed by H. pylori at levels exceeding that of any other known microbe [15] and is estimated to constitute 10–15% of the bacterium’s total protein content [16]. In addition, urease is essential for establishing colonization as H. pylori urease mutants are unable to infect the host [17–21]. Several lines of evidence suggest that urease plays a significantly greater role in infection than simple acid neutralization. Elevating the gastric pH to 7.0 was shown to be insufficient in permitting colonization by a urease negative strain [18]. In several in vitro studies urease and its catalytic products contributed directly to virulence. Ammonia produced by urease activity caused damage to the gastric epithelium by disrupting tight cell junction integrity [22, 23] and CO2 protected against the bactericidal activity of the nitric oxide metabolite, peroxynitrite, produced by phagocytes to kill engulfed bacteria [24]. Furthermore, several studies suggest that urease may directly interact with host epithelial and immune cells. Urease has been shown to bind to major histocompatibility complex (MHC) class II molecules on gastric epithelial cells thereby inducing cell apoptosis [25] and the UreB subunit can stimulate monocytes to release proinflammatory cytokines by binding to cell surface CD74, a MHC class II associated invariant chain [26, 27]. In addition, an in vivo study demonstrated that changes to the surface of the urease complex resulted in the eventual clearance of H. pylori infection in mice [28]. Loss of colonization was attributed to the disruption in urease mediated interactions between H. pylori and host cells as urease activity was unaffected by the mutation, ruling out loss of acid resistance or nitrogen assimilation [29] as contributing factors [28]. Clinical isolates maintain high urease activity even after years of chronic infection when the bacterium has established itself in the relatively neutral environment of its gastric niche implicating that ongoing urease expression is required for persistence. However due to the lack of appropriate genetic systems this hypothesis could not be tested experimentally.
The necessity of urease activity in establishing colonization hinders the study of its function during persistence when using conventional knockout mutants. The availability of a conditional urease mutant would overcome this constraint by permitting changes to the urease phenotype during an established infection. To conclusively determine if urease is indeed required after colonization is established, we generated conditional H. pylori urease knockout mutants using a tetracycline repressor (tet) based system [30]. This system controls gene expression by way of a tetracycline repressor (TetR) that binds to specific operator sequences (tetO) in the target promoter and silences transcription of the downstream gene. Expression of the target gene can be turned on by the administration of a potent tetracycline inducer, such as anhydrotetracycline (ATc) or doxycycline (Dox) [31]. This system was recently adapted to H. pylori and gene expression was regulated in vivo during active infection [32]. In the current study, we adapted this system to generate conditional urease mutants. We demonstrate that in an established infection, loss of urease expression is detrimental to the bacterial survival in the host. Strong selective pressure on the bacteria for continuous urease expression is further demonstrated by the emergence of escape mutants that successfully repopulated the mouse stomach six weeks after genetic silencing of urease was initiated.
The urease structural genes, ureA and ureB, encoding for the 27 kDa UreA and 62 kDa UreB urease subunits, are transcribed as a single operon under the control of the ureA promoter, PureA [33]. To regulate the expression of urease in H. pylori, we placed the operon under tet control. Based on previous mutational studies of PureA [32, 34], the promoter was mutated to incorporate one or more tetO sequences to generate a series of different PureA derivatives, urePtetO(I-V) (Fig 1A and 1B). These tet-promoter constructs were made using PCR based techniques and used to replace the native chromosomal PureA by allelic replacement. This strategy involved first generating a recipient H. pylori strain in which PureA and ureA has been replaced with a rpsL-cat cassette. The urease negative recipient strain was then naturally transformed with the urePtetO PCR constructs to generate strains with tetO modified PureA derivatives and a restored ureA gene. H. pylori strains harbouring these constructs were characterized to identify a tet-promoter construct with regulatory properties that would permit the appropriate level of complementation to ensure colonization yet could be sufficiently silenced to prevent infection. The functionality of these tet-promoters was first assessed in the wild-type background for their ability to drive urease expression by measuring urease enzymatic activity, UreB expression and mouse colonization.
The urePtetO constructs were introduced into the wild-type X47 strain, replacing the chromosomal PureA and generating X47 urePtetO strains (OND2018—OND2022). The urease expression level and the urease enzymatic activity of these strains under standard growing conditions was measured and compared to the parent strain (Fig 1C and 1D). Despite expressing less urease, as determined by immunodetection of UreB in the total cell lysate (Fig 1C), the urease activity measured in strains transformed with urePtetOI, urePtetOII and urePtetOV was found to be comparable to wild-type (Fig 1D). In comparison, the urease activity in strains transformed with urePtetOIII and urePtetOIV were reduced by 60% and 25% respectively, which was also accompanied by substantially reduced amount of UreB in the total cell lysate (Fig 1C).
Given that replacement of the wild type PureA with the urePtetO promoters resulted in reduced urease expression which concomitantly may reduce their ability to establish infection, C57BL/6J mice were challenged with strains X47 urePtetO(I-V) (OND2018—OND2022) to assess if urease expression in these strains was still sufficient to facilitate colonization. All five X47 urePtetO strains were successfully re-isolated from mouse stomachs two weeks after oral challenge (Fig 1E). The infection rate and bacterial load in mice challenged with strains X47 urePtetOI, X47 urePtetOII and X47 urePtetOV was comparable to the control group challenged with the wild-type strain. However the infection rates were decreased for groups challenged with strains of lower urease activity, with strain X47 urePtetOIII displaying a major defect in colonization. To verify that colonization had not been established due to mutation or reversion of urePtetO, the ureA promoter region of re-isolated strains was sequenced which confirmed that the sequences of the urePtetO constructs remained unaltered after passage through mice.
After establishing that the tetO modifications to the ureA promoter did not abrogate colonization per se, we evaluated if these promoters could regulate urease expression in a tetracycline dependent manner. All five ureA promoter derivatives were transformed into a X47 recipient strain that expressed TetR under the control of the strong flaA promoter [32]. The resulting strains, X47 mdaB::ptetR4, urePtetO(I-V) (OND1954—OND1958), did not express urease when grown on standard CBA plates however urease expression could be induced when grown on CBA plates containing 50 ng/ml anhydrotetracycline (ATc) (S1 Fig). These results demonstrated that TetR effectively silenced the tet-promoters in these strains.
Regulation of urePtetO promoters by TetR in conditional urease knockout strains X47 mdaB::ptetR4, urePtetO(I-V) was assessed using the urease enzymatic activity assay. Bacteria were cultured in the absence or presence of 50 ng/ml ATc for two successive passages and then collected for analysis. When strains were cultured in the absence of ATc, urease activity was below the detection limits of the assay (2 U/ml of Type III urease from Jack bean) (Fig 2A). For strains grown in the presence of ATc, urease activity in strains X47 ptetR4; urePtetO-I, -II and -V was induced to wild type levels, while the urease activity for strains X47 ptetR4; urePtetO-III and -IV remained below 10% of wild-type activity. These results demonstrated that by using the appropriate tet-promoter urease activity can indeed be regulated by the presence of a small molecule inducer, confirming the generation of conditional H. pylori urease knockout mutants.
The tet-responsive promoters urePtetOI and urePtetOV have different genetic architectures (Fig 1B) and upon induction also promoted the greatest urease expression levels amongst the tested strains. Based on these results the regulation of these two promoters was further characterized. The kinetics of urePtetO induction and repression was analysed in strain X47 ptetR4; urePtetOI (OND1954) and X47 ptetR4; urePtetOV (OND1958) by immunodetection of the UreB protein (Fig 2B and 2C). After addition of 200 ng/ml ATc to the culture medium, UreB protein expression increased over time and reached maximum levels after 12 h and 8 h for urePtetOI and urePtetOV, respectively (Fig 2B). Withdrawal of ATc from induced cultures led to a significant decrease in UreB protein levels within 3 h, demonstrating that both urePtetOI and urePtetOV were quickly silenced (Fig 2C) and that the UreB protein was turned over efficiently, falling to the threshold of detection within 12 h.
With the knowledge that tetracycline dependent regulation of urease expression was attainable in vitro we next turned our attention to establishing a mouse model of infection. Based on previous studies involving in vivo tet-systems [35–37] the inducer molecule doxycycline (Dox) was first used as a model inducer to identify a maximal dosage of material that could be tolerated by the bacterium in vivo. We found that wild-type X47 could still infect mice when the animals were supplemented with up to 10 mg/l of Dox in their drinking water. Colonization by wild-type X47 was severely attenuated at 100 mg/l of Dox and bacteria could not be reisolated at 1000 mg/l of Dox supplement (S2A Fig). Furthermore, strain X47 ptetR4; urePtetOI (OND1954), which emerged as the prime conditional urease mutant candidate from the in vitro studies, was tested to verify its ability to establish initial colonisation and then used to optimise the dosage of inducer molecule to regulate urease expression in vivo. OND1954 was only capable of establishing infection in C57BL/6J mice when the animals received Dox supplementation in their drinking water, demonstrating that urease is essential for OND1954 to establish colonization in the mouse infection model (S2B Fig). Addition of Dox supplement at 1 mg/l supported colonization of OND1954 and although attenuated, the conditional mutant was also isolated from animals supplemented with Dox at 5 mg/l and 10 mg/l.
Having identified a minimal supplement dose of Dox inducer, we then sought to complete the infection model by investigating two more important factors; the use of the less toxic tetracycline derivative ATc, and attempting to improve the robustness of the conditional urease mutant strain. The original wild-type X47 strain underwent four consecutive transformations to generate the conditional urease mutant OND1954 and consequently the strain may have accumulated secondary mutations that would decrease its fitness in vivo. To address this, the output clones of OND1954 isolated from three individual mice were collected and each clone was verified to be a conditional urease mutant. These clones were then pooled, OND3241(A-E), and used in subsequent infection studies to test if passage though mice led to improved infection rates. Mice were challenged with either the wild-type strain, the original conditional mutant OND1954 or the mouse passaged urease conditional mutant OND3241, and supplemented without or with 5 mg/l Dox or ATc. Colonization of OND3241 remained dependent on inducer supplement and using ATc instead of Dox resulted in an improved infection rate and bacterial load in the infection model (Fig 3).
Having established an infection model in which the H. pylori urease phenotype could be regulated in vivo, we proceeded to investigate what effect tet-mediated silencing of urease expression had on established H. pylori infections. Mice were challenged with the conditional strain OND3241 and provided with 5 mg/l ATc supplement to establish infection. After two weeks, the supplement was withdrawn and the animals were sacrificed at indicated time points. The conditional H. pylori urease mutant could still be isolated on days 1 and 3 after supplement withdrawal, however on days 5 and 7 the bacterial load had decreased to below our detection limit (Fig 4A). This data demonstrated for the first time that continuous urease expression is required by H. pylori to maintain colonization even after the bacteria have become established in the gastric niche.
To test if H. pylori were under selective pressure to overcome tet-regulation the suppression experiment was repeated and the animals were sacrificed at a much later time point. Mice were challenged with either wild-type X47 or OND3241 and provided with 5 mg/l ATc supplement to establish infection. After two weeks, the supplement was withdrawn from half the groups (both OND3241 and wild-type) while the remaining groups were maintained on ATc supplement and the animals were sacrificed at different time points (Fig 4).
No differences in bacterial load or infection rate was observed for animals infected with wild-type X47 over the course of the experiment demonstrating that long-term ATc supplement (5 mg/l for 8 wks) does not interfere with H. pylori infection (Fig 4B). Animals challenged with OND3241 and maintained continuously on ATc supplement had a consistent infection rate of 60% (Fig 4B). In vitro tests confirmed that bacteria re-isolated from these groups remained conditional urease mutants, even after a total infection time of 8 weeks. Withdrawal of ATc supplement from the animal groups challenged with OND3241 resulted in reduced infection load on days 3 and 5 and, although not completely cleared to below our detection limit in all animals, the infection rate had decreased to 20% on day 5. However, when mice challenged with the OND3241 were left in the absence of ATc for 42 days, the bacterial load and the infection rate had increased resulting in 80% of the animals bearing bacteria in the stomach above our detection limit. Importantly, unlike the bacteria re-isolated at the earlier time points (day 3 and day 5), H. pylori re-isolated from this last group of mice were all urease positive and they were no longer conditional urease mutants as tested qualitatively in vitro. This result revealed that the strain was under selective pressure to restore urease expression.
In an effort to identify possible genetic mutations to overcome tet-regulation of the urease operon, whole genome sequencing of the original conditional urease mutant strain OND1954, the individual clones of input strain OND3241(A-E), and 34 output strains recovered at day 42 (5 conditional urease-negative strains and 29 urease-positive tet-escape mutants) was undertaken. Sequence data were mapped against reference strain OND1954 and variants specific to tet-escape mutants were identified (S1 Table). Sequence analysis of the ureAB locus, including the upstream regulatory region, revealed no changes between OND1954, OND3241 and the tet-escape mutants. Interestingly, the tet-escape mutants harboured non-synonymous substitutions or frameshift mutations in at least one of the following genes, tetR and the flagellar biosynthesis genes fliA, fliE and flgE. These data reveal that tet-regulation was overcome in the tet-escape mutants not by altering the tetO binding sites but through affecting the repressor protein.
A set of conditional urease mutants were generated to demonstrate for the first time that urease expression is essential in the persistence stage of H. pylori infection, which broadens our understanding of the role of this enzyme during chronic infection.
Genetic manipulation of PureA to place urease under tet-control led to a decrease in the basal levels of urease expression for all urePtetO constructs tested. However, under the conditions tested, strains transformed with urePtetOI, urePtetOII and urePtetOV were found to have comparable urease activity to that of wild-type. This data can be reconciled as it has been reported that under neutral in vitro growth conditions without added nickel, such as the growth conditions used in this study, a significant amount of urease in wild-type strains is in the inactive apoenzyme form and only a minor fraction of urease is active [38–40]. Urease activity in H. pylori is highly controlled and is modulated through several different mechanisms in response to various environmental cues [39]. Since our goal was to establish a working in vivo model, we decided to directly test if the decrease in urease expression could be tolerated by the bacteria by assessing if the urePtetO strains were capable of colonizing the murine stomach.
Interestingly, when analyzing the in vivo colonization data from the X47 urePtetO strains (OND2018—OND2022) a positive correlation between infection rate and in vitro urease expression and activity but not to bacterial load in colonized animals was observed (Fig 1E). This finding suggests that for initial colonization of the murine stomach the amount of urease activity is an important factor likely due to the fact that the bacteria need to withstand the acidity of the gastric lumen until they reach their gastric niche, deep into the gastric mucus near the epithelial surface. However, once the bacteria are established within their environmental niche, although urease is still required for growth, the level of urease expression may be less important for maintaining colonization as mice colonized with strains transformed with urePtetOIII and urePtetOIV had a similar bacterial burden compared to mice infected with strains expressing more urease.
Infection of the mouse host by the conditional urease mutant was strictly dependant on supplementation with a tetracycline inducer, confirming that genetic regulation of urease expression was stringent enough to prevent colonization. Furthermore, in the induced state, tet-mediated expression of urease was sufficient to allow and maintain infection by the conditional urease mutant. Withdrawal of the supplement resulted in clearance of the bacterium within 5 days. This is in line with the slow shut-off observed in other mouse models using tet-based regulation systems which has been attributed to the persistence of doxycycline in tissues [41, 42].
Notably the longer time of clearance of the bacterium in vivo provided the opportunity for the emergence and selection of tet-escape mutants. H. pylori possess several mechanisms, such as an error prone PolA [43] and efficient DNA homologous recombination and transformation systems [44, 45], that permit the bacterium to undergo rapid microevolution to adapt to changing environments in its specific host [46, 47]. The emergence of tet-escape mutants in this study suggests that there is strong selective pressure on the bacterium for continuous urease expression to maintain chronic infection. Whole genome sequence analysis of the tet-escape mutants identified several different mutations that likely explain how tet-regulation of the ureAB operon was overcome.
One group of escape mutants had missense or nonsense mutation within the tetR gene. TetR is a finely tuned transcriptional regulator [31] and therefore most amino acid changes are likely to have deleterious effects to the function of TetR by inhibiting repressor dimerization or DNA binding [48–50]. Interestingly, another group of tet-escape mutants harbored either amino acid substitutions within the DNA binding domains of the alternative RNA polymerase sigma factor FliA (σ28) or a truncated FliA due to nonsense mutation and frameshift alteration. FliA controls the transcription of some late flagellar genes (class 3) including the flagellin subunit, flaA [51, 52]. Previous studies have reported that H. pylori fliA mutants have no detectable flaA transcript and have truncated flagella [51]. In the conditional H. pylori urease mutants generated in this study, the transcription of tetR is driven by a flaA promoter. Therefore, it is reasonable to suggest that the fliA mutations identified in the tet-escape mutants likely affect the transcription of tetR from Pfla and consequently release the ureAB operon from tet-regulation. Additionally, some tet-escape mutants acquired non-synonymous mutations in fliE and flgE, genes that encode for components of the flagellum hook-basal body complex. Flagellar biosynthesis is a highly ordered and regulated process and transcription of late flagellar genes by fliA proceeds only once the hook-basal body complex is complete [53, 54]. Mutation of fliE has been shown in other studies to affect the regulation of fliA dependent genes, leading to a two-fold reduction in flaA transcript levels [54]. Thus mutation of fliE and flgE may indirectly impact on the transcription of tetR from PflaA through the negative control of FliA [51, 53]. Motility is essential for H. pylori colonization [19, 55, 56], and therefore the isolation of tet-escape mutants with mutations in genes involved in flagellar biosynthesis (fliA, fliE and flgE) raises interesting questions. During the chronic stage of infection, once the bacteria are established in their environmental niche and have adhered to gastric epithelial cells [57], can motility be compromised in preference for improved urease expression? Further investigations with the appropriate conditional mutants are necessary to better understand the escape and in particular to test whether motility is still required once colonization is established. Another question is whether there are other mutations acquired by the tet-escape mutants that could compensate for mutations in fliA, fliE and flgE? In-depth analysis of the whole genome sequencing data together with mutational studies needs to be undertaken to discern how these tet-escape mutants are able to survive in the host.
The sequencing results underline the importance of urease for H. pylori to maintain persistence infection and reveal the high selective pressure for continuous expression of urease even after colonisation is successfully established. The mutations identified in the tet-escape mutants add support to the hypothesis that the genomic plasticity of H. pylori is an important mechanism for adaptation to new and changing environments [58]. Furthermore these findings also highlight the potential of using tet-based genetic tools together with whole genome sequencing to study H. pylori genetic plasticity and adaptation in a changing biological environment when the bacterium is put under duress.
Using the conditional mutant, we have demonstrated that urease is essential for chronic infection and that repression of urease expression results in the loss of bacterial load within 5 to 7 days. The availability of these tools now allows for new questions to be asked regarding the reason behind the relatively rapid loss in colonization. One reason may be that the loss of urease activity results in the bacteria being more susceptible to clearance by phagocytic cells [24]. Another possibility is that the loss of urease activity negatively affects H. pylori ability to swim through gastric mucus, as the bacteria would no longer have the ability to decrease the viscosity of mucus through the elevation of local pH [59], and consequently are cleared due to turnover of the mucus lining. The conditional urease mutant will serve as a valuable tool in further studies that pursue this line of investigation.
In this study, H. pylori conditional urease mutants were generated by placing the expression of the urease subunits, UreA and UreB, under tet-control and have permitted the first direct testing of the hypothesis that urease is required by H. pylori for chronic infection. Furthermore, eventual escape from tet-regulated urease expression by H. pylori demonstrates that there is a very strong selective pressure on the bacterium to maintain urease expression during infection. Our data validates urease as a good target for therapeutic intervention. The conditional urease mutants generated here can also be used to gain more detailed insight into the role of urease in the persistence stage of infection including its interactions with MHC class II molecules [25], induction of proinflammatory cytokine [27] and its potential role in motility [59]. Furthermore, this study demonstrates the need for conditional mutants, generated by using genetic tools such as the tet-system, to study H. pylori virulence factors, persistence and the bacterium’s influence on the host microbiota.
H. pylori X47 strains used in this study are listed in S2 Table. Bacteria were grown at 37°C under microaerobic conditions on Columbia blood agar (CBA) plates containing 5% horse blood and Dent’s antibiotic supplement (Oxoid). When appropriate, antibiotic selection was carried out by supplementing media with chloramphenicol or streptomycin at a final concentration of 10 μg/ml. Microaerobic conditions were established in sealed jars using the Anoxomat MarkII system (Mart Microbiology B.V., the Netherlands) after one atmosphere replacement with the following gas composition N2:H2:CO2, 85:5:10.
All genetic manipulation of H. pylori strains was done using genomic insertion and replacement of a counter-selectable rpsL-cat cassette [60]. The use of the counterselectable streptomycin susceptibility (rpsL-based) system requires a host strain that possesses a streptomycin-resistant phenotype [61]. The H. pylori X47 host strain is naturally streptomycin-resistant and no modifications to this strain were required. The genotype of all mutants was confirmed by PCR and/or DNA sequencing.
Oligonucleotides used in this study are listed in S3 Table.
To place ureA and ureB under tet control, wild-type nucleotide sequences flanking the -35 and -10 promoter regions of the ureA promoter, PureA, were replaced with tetO sequences to generate five derivatives of PureA, urePtetO(-I through -V) (Fig 1A and 1B). These promoter constructs were used to replace the native urease promoter, using the two-step rpsL-cat based transformation approach.
A construct composed of the counterselection cassette flanked by DNAs homologous to regions of the ureA locus, ureA::rpsL-cat, was made by SOE PCR [62, 63] (S3 Fig) and used to generate recipient strains in which PureA and ureA were replaced with rpsL-cat. Two 1 kb regions flanking PureA and ureA (HP0073), were amplified from 26695 genomic DNA using primers ureArcat1 and ureArcat2, and ureArcat3 and ureArcat4 respectively. The rpsL-cat selection cassette was amplified using primers ureArcat5 and ureArcat6. Nested primers ureArcat7 and ureArcat8, were used to generate and amplify a final 3.4 kb PCR product, ureA::rpsL-cat. Natural transformation of the H. pylori strains with the ureA::rpsL-cat PCR construct was performed to obtain the recipient strain OND2017. Transformants isolated on chloramphenicol plates were urease negative.
Five tetO modified ureA promoter constructs urePtetO(I-V), containing up to three tetO sites, were constructed by SOE PCR (S4 Fig). The primer pairs used to make each urePtetO construct are listed in S4 Table. Briefly, a 1 kb fragment upstream, arm I, and a 1.5 kb fragment downstream, arm II, of PureA were amplified using 26695 genomic DNA as a template. Long primer tails were used to reconstruct the ureA promoter region upon fusion of arms I and II by SOE PCR. Primers ureArcat7 and ureArcat8 were used to amplify the final 2.5 kb products, urePtetO(I-V), and sequencing confirmed that the modified ureA promoters were reconstructed correctly. Natural transformation of the recipient strain OND2017 with urePtetO PCR constructs resulted in replacement of the rpsL-cat with urePtetO and restoration of ureA, generating strains X47 urePtetOI through X47 urePtetOV (OND2018—OND2022). Correct allelic replacement of the resulting Strr transformants was confirmed by colony PCR using primers ureAP1 and ureArcat8 and by sequencing using primer urePseq.
Conditional urease strains were generated by transforming the TetR expressing H. pylori strain, X47 mdaB::ptetR4 (OND1987) [32], with the ureA::rpsL-cat PCR construct to generate the recipient strain OND2026. This urease negative strain was then transformed with each of the five urePtetO constructs to generate conditional urease mutant strains X47 mdaB::ptetR4; urePtetO(I-V) (OND1954—OND1958). Transformants were first screened for tetracycline dependent urease expression using the urease phenotype assay and additional characterization was done using the urease activity assay and immunoblot analysis.
Urea culture plates (Brucella broth, 7% NCS, 1 mM urea, phenol red 100 mg/l, vancomycin 6 mg/l, pH 6) were used to assay the urease phenotype of H. pylori clones. The pH of the media was adjusted with 1 M HCl before the addition of NCS and vancomycin. The pH was low enough to observe the colourimetric change of phenol red, from yellow to red, due to the catalytic activity of urease on urea, but not acidic enough to inhibit the growth of urease negative strains. To screen for tet-regulated urease activity, transformants and colonies re-isolated from infected animals were replica plated onto CBA plates with or without 50 ng/ml of ATc and cultured for 48 h. Bacteria were then patched onto urea plates and incubated under microaerobic conditions. Urea plates were examined after 16 h of incubation to identify clones that had switched urease phenotype upon exposure to ATc. Localized changes in colour around each growing colony identified urease positive clones. Conditional urease mutant strains grown on CBA plates without ATc remained urease negative, while strains grown on CBA plates with ATc became urease positive (Example S5 Fig).
The urease activity assay used in this study was adapted from the protocol previously described [28]. Strains were grown on CBA plates without or with 50 ng/ml ATc for two successive passages. Bacteria from 24 h plate cultures were collected and resuspended in cold buffer A (25 mM phosphate buffer, pH 6.8) and standardized to an OD600 = 4.0. A 50 μl aliquot of the standardized bacterial suspension was then diluted with 50 μl of buffer B (25 mM phosphate buffer, pH 6.8, 0.2% Tween-20). A 25 μl aliquot of this diluted bacterial suspension was transferred into one well of a 96 well plate, diluted with 150 μl of buffer C (25 mM phosphate buffer, pH 6.8, 250 μM phenol red) and incubated for 5 min at 37°C. A 75 μl aliquot of urea solution (0.5 M) was then added to the well and the absorbance at 560 nm was measured every 72 s for 75 cycles using a POLARstar Omega (BGM Labtech) plate reader. Activity was measured as the rate of change in absorbance over time and expressed as percent of urease activity of the wild-type X47 strain. All urease activity measurements were carried out in triplicate and experiments were repeated at least three times.
Bacteria were grown in Heart Infusion (HI) medium supplemented with 10% Newborn Calf Serum (NCS) and vancomycin (6 μg/ml). Cultures were inoculated with bacteria suspended in PBS to give a starting OD600 = 0.05, and grown under microaerobic conditions at 37°C and 120 rpm. For induction, H. pylori cultures were grown to mid-log phase in 10 ml of media. Cultures were induced with 200 ng/ml ATc and bacteria were incubated for another 12 h, with aliquots were taken at indicated time points. For gene silencing, conditional strains were cultured in the presence of 200 ng/ml ATc to mid-log phase. Fresh HI media, with or without 200 ng/ml ATc, was inoculated with pre-induced bacteria (OD600 = 0.5) and grown for 12 h, with aliquots taken at indicated time points. Bacterial cells were collected by centrifugation and washed twice with PBS before processing for immunoblot analysis.
Bacterial whole cell lysates were prepared as previously described [32]. The protein concentration of bacterial cell whole cell lysate samples was determined using the Micro BCA protein assay reagent kit (Pierce) with bovine serum albumin as the standard. Equal amounts of protein for each sample were mixed with 3x SDS-PAGE sample loading buffer, incubated at 95°C for 10 min, and proteins were separated by 10% SDS-PAGE and electrotransferred to a PVDF membrane. For detection of the UreB subunit of urease, mouse anti-UreB (Austral biologicals) was used at a dilution of 1:8000. Secondary antibody rabbit anti-mouse-HRP (Jackson ImmunoResearch Laboratories) was used at a dilution of 1:10,000 and detection of the secondary HRP conjugate was accomplished by chemiluminescence (Sigma) using LAS 3000 (Fujifilm) (software Image reader LAS 3000 V2.2). For loading controls, duplicate gels were run in parallel and stained with Coomaise Brilliant Blue R-250 (S1B Fig, S6 Fig and S7 Fig).
Mouse procedures were reviewed and approved by the Institutional Animal Care and the Animal Ethics Committee of the University of Western Australia. 6–7 week old C57BL/6J female mice were challenged once by oral gavage with 200 μl of 1 x 109 CFU/ml of bacteria suspended in HI broth. Groups of infected mice received doxycycline (Dox), anhydrotetracycline (ATc) or no supplement in drinking water containing 5% sucrose. Water was kept in light-protected bottles and changed every three days. Mice were sacrificed at indicated time points and stomachs were removed and homogenized in 1 ml HI using a tissue lyser (Retch). Homogenates were serially diluted and plated out on H. pylori selective plates (CBA containing 5% Horse blood, Dent, nalidixic acid 10 mg/l and Bacitracin 100 mg/l) to determine the bacterial burden. Were appropriate, re-isolated clones were assayed for tet-responsive gene expression.
Preparation of MiSeq library was performed using Illumina Nextera XT DNA sample preparation kit (Illumina, San Diego, CA, USA) as previously described with minor modifications [64]. In brief, 1 ng of genomic DNA was fragmented in 5 μl of Amplicon Tagment Mix and 10 μl of Tagment DNA buffer. Tagmentation reaction was performed by incubation at 55°C for 10 min followed by neutralisation with 5 μl of Neutralise Tagment Buffer for 5 min. Tagmented DNA (25 μl) was indexed in a 50 μl limited-cycle PCR (12 cycles) as outlined in the Nextera XT protocol and subsequently purified using 25 μl of AMPure XP beads (Beckman Coulter Inc, Australia). The fragment size distribution of the purified DNA was analysed utilising a LabChip GXII 2100 Bioanalyser. DNA libraries were adjusted to 2 nM, pooled in equal volumes and then denatured with 0.2 N NaOH according to the Nextera protocol. The libraries were sequenced using 2 × 300 paired-end protocols on an Illumina MiSeq instrument (MiSeq Reagent Kit v3 for 600 cycles). The draft genome sequence of H. pylori OND1954 has been deposited at DDBJ/ENA/GenBank under the accession MVFB00000000. The version described in this paper is version MVFB01000000. All raw sequence data generated in this study have been submitted to Sequence Read Archives (SRA) database with accession numbers listed in S5 Table.
The generated MiSeq reads of H. pylori strain OND1954 was assembled using SPAdes genome assembler (version 3.8.2) with careful option [65]. The draft genome sequence was subsequently annotated using Prokka (version 1.11) with Swiss-Prot, Pfam (release 30.0), TIGRFAMs (release 15.0) and Superfamily (version 1.75) databases [66–70]. The annotation features are available in S1 File. The raw reads of OND1954-derivative strains were trimmed and mapped against the annotated draft genome using Bowtie2 on Geneious R7 platform [71, 72]. Variants were called using the following parameters: minimum coverage = 10 and minimum variant frequency = 0.7.
For mouse colonization assays (where n ≥ 5) the Mann-Whitney unpaired two-tailed test was used to compare colonization loads and the two sided Fisher’s exact test was used to compare infection rates. Bonferroni correction was used for multiple pairwise testing. Statistical analysis was performed using GraphPad Prism version 7 for Windows, (GraphPad Software) and Stata Statistical Software (StataCorp. 2015. Release 14. College Station, TX: StataCorp LP.).
|
10.1371/journal.pbio.1000426 | Structure of the CaMKIIδ/Calmodulin Complex Reveals the Molecular Mechanism of CaMKII Kinase Activation | Long-term potentiation (LTP), a long-lasting enhancement in communication between neurons, is considered to be the major cellular mechanism underlying learning and memory. LTP triggers high-frequency calcium pulses that result in the activation of Calcium/Calmodulin (CaM)-dependent kinase II (CaMKII). CaMKII acts as a molecular switch because it remains active for a long time after the return to basal calcium levels, which is a unique property required for CaMKII function. Here we describe the crystal structure of the human CaMKIIδ/Ca2+/CaM complex, structures of all four human CaMKII catalytic domains in their autoinhibited states, as well as structures of human CaMKII oligomerization domains in their tetradecameric and physiological dodecameric states. All four autoinhibited human CaMKIIs were monomeric in the determined crystal structures but associated weakly in solution. In the CaMKIIδ/Ca2+/CaM complex, the inhibitory region adopted an extended conformation and interacted with an adjacent catalytic domain positioning T287 into the active site of the interacting protomer. Comparisons with autoinhibited CaMKII structures showed that binding of calmodulin leads to the rearrangement of residues in the active site to a conformation suitable for ATP binding and to the closure of the binding groove for the autoinhibitory helix by helix αD. The structural data, together with biophysical interaction studies, reveals the mechanism of CaMKII activation by calmodulin and explains many of the unique regulatory properties of these two essential signaling molecules.
This article can also be viewed as an enhanced version in which the text of the article is integrated with interactive 3-D representations and animated transitions. Please note that a web plugin is required to access this enhanced functionality. Instructions for the installation and use of the Web plugin are available in Text S1.
| CaMKII enzymes transmit calcium ion (Ca2+) signals released inside the cell by regulating signal transduction pathways through phosphorylation: Ca2+ first binds to the small regulatory protein CaM; this Ca2+/CaM complex then binds to and activates the kinase, which phosphorylates other proteins in the cell. Since CaMKs remain active long after rapid Ca2+ pulses have dropped they function as molecular switches that turn on or off crucial cell functions in response to Ca2+ levels. The multifunctional CaMKII forms of this enzyme – of which there are four in human – are important in many processes including signaling in neurons and controlling of the heart rate. They are particularly abundant in the brain where they probably play a role in memory. CaMKII forms an exceptionally large, dodecameric complex. Here, we describe the crystal structure of this complex for each of the four human CaMKII catalytic domains in their autoinhibited states, a complex of CaMKII with Ca2+/CaM, as well as the structure of the oligomerization domain (the part of the protein that mediates complex formation) in its physiological dodecameric state and in a tetradecameric state. Detailed comparison of this large body of structural data together with biophysical studies has allowed us to better understand the structural mechanisms of CaMKII activation by CaM and to explain many of the complex regulatory features of these essential enzymes.
| Calcium/Calmodulin (Ca2+/CaM)-dependent serine/threonine kinases (CaMKs) constitute a family of 81 proteins in the human proteome that play a central role in cellular signaling by transmitting Ca2+ signals [1]. Kinases in this protein family are activated through binding of Ca2+/CaM to regulatory regions that either flank the catalytic domain or are located in regulatory molecules [2]. Four CaMKII isozymes (α, β, γ, and δ), in addition to about 30 splice variants, are expressed in humans. The α and β isoforms are brain specific and together make up approximately 1% of total brain protein in rodents and up to 2% of total protein in their hippocampus [3]. The γ and δ isoforms are expressed in most tissues, but in comparison have much lower expression levels [4],[5]. The unique switch-like properties of CaMKII activation and its extremely high abundance in the brain identified CaMKII as a key regulator of cellular memory and learning [6]. CaMKII is essential for the induction of long-term potentiation (LTP), a long-lasting increase in the efficiency of synaptic transmission between neurons that is believed to be a cellular correlate of memory [7],[8]. Stimuli that induce LTP lead to autophosphorylation at T286 in CaMKIIα (T287 in the β, γ, and δ isoforms), thereby resulting in sustained CaMKII activation [9]; mice expressing the CaMKIIα T286A mutant were severely impaired in learning [10].
Several CaMKIIδ variants are highly abundant in myocardial tissue [11],[12]. Increased CaMKII activity has been observed in patients with structural heart disease and arrhythmias, where prolonged action potential duration leads to sustained hyperactivation of CaMKII and heart failure [11].
CaMKII proteins form large oligomeric structures. The N-terminal kinase domain is tethered via an autoinhibitory helix and a calmodulin binding site to a C-terminal oligomerization domain that organizes the enzyme into ring-shaped oligomers. Three-dimensional reconstruction of single-particle electron microscopy images revealed dodecameric assemblies for all purified homogeneous full-length CaMKII isozymes [13],[14]. In contrast, tetradecamers were detected in the crystal structures of isolated oligomerization domains. This non-physiological oligomerization state has been attributed to the absence of the kinase domain [15],[16].
The cellular regulation of CaMKII activity is the outcome of a complex interplay between protein localization, heterooligomerization, local Ca2+/CaM concentrations, CaMKII autophosphorylation, and dephosphorylation of CaMKII by phosphatases [2],[17]. The structure of the isolated Caenorhabditis elegans CaMKII (CeCaMKII) kinase domain in its autoinhibited state provided the first insight into the molecular mechanism of CaMKII regulation [18]. In this structure the inhibitory domain forms a helix that binds tightly to the substrate binding pocket preventing access of substrates. Interestingly, the regulatory domains of two catalytic domains interacted as antiparallel coiled-coils in the CeCaMKII structure suggesting that the inhibitory helix mediates self association in the inactive state of the enzyme. This association model has also been evoked to explain cooperativity of CaMKII activation by Ca2+/CaM observed in enzyme kinetic assays and “pairing” of kinase domains in autoinhibited holoenzymes [19],[20]. In the inactive state the autophosphorylation site within the regulatory domains is not accessible [18]. It has been speculated that this inhibitory “block” of the regulatory domain is released by structural changes induced upon Ca2+/CaM binding. Once phosphorylated at the regulatory T286 site (CaMKIIα numbering) by catalytic domains present in the same holoenzyme, steric constraints prevent rebinding of the autoinhibitory domain to the catalytic domain [4],[21],[22].
In addition, CaMKII can be made insensitive to Ca2+/CaM by autophosphorylation at T305/T306 located within the Ca2+/CaM binding site [23],[24], a process that is facilitated by interaction with the membrane associated guanylate kinase (MAGUK/CASK) [25],[26]. The balance between the Ca2+/CaM-sensitive and -insensitive CaMKII pool is critical for the regulation of post-synaptic plasticity [27],[28].
In CaMKIIα, autophosphorylation of T306 but not of T305 was observed in vitro, leading to a strong reduction of Ca2+/CaM binding [29]. The region flanking this autophosphorylation site represents a non-consensus substrate site for CaMKII, which raises the question of how this motif would be efficiently recognized as a substrate.
To date, our structural knowledge of how CaMKIIs are activated is based solely on structures of isolated kinase domains and peptide complexes of either catalytic domains with their substrates or Ca2+/CaM with calmodulin binding sites [18],[20]. We were interested in describing the molecular mechanisms that govern CaMKII activation in an intact catalytic domain/Ca2+/CaM complex. The structure of the CaMKIIδ/Ca2+/CaM presented here captures the kinase in a state where the inhibitory helix is dislodged from the substrate binding site, thereby making it available for autophosphorylation by an adjacent kinase molecule. Analysis of this co-crystal structure, structures of all human isozymes in their autoinhibited state, and in-solution association studies showed that binding of Ca2+/CaM triggers large structural changes in the kinase domain as well as in the CaMKII regulatory domain that together lead to allosteric kinase activation. Furthermore, we also describe the structure of an oligomerization domain in its physiological, dodecameric state. Based on the comparison of this large body of structural information and biochemical characterization we propose a model that explains the substrate recognition leading to Ca2+/CaM-dependent allosteric activation of human CaMKIIs.
To date, our understanding of the molecular mechanisms that define the CaMKII autoinhibited state are based on the structural model of the C. elegans CaMKII orthologue (CeCaMKII). This crystal structure shows an occluded substrate binding site, rearrangements in the ATP binding site that disturb co-factor binding and a remarkable dimeric assembly involving the inhibitory helix and the CaM binding motif (corresponding to residues K293-F313 in CaMKIIα) [18]. CeCaMKII and human CaMKIIα share 77% sequence identity. We were interested in determining whether regulatory mechanisms suggested based on the crystal structure of CeCaMKII would be conserved in human CaMKII isozymes. To address this, we determined the structures of all human CaMKII isozymes in their autoinhibited state. The structures were refined at resolutions ranging from 2.25 Å (CaMKIIγ) to 2.4 Å (CaMKIIβ). Details of the diffraction data statistics and refinement have been summarized in Table S1. Importantly, whereas the crystallized constructs of the α and β isozymes contained the catalytic domain and the inhibitory region but only a part of the Ca2+/CaM binding motif, the constructs of both CaMKIIγ and CaMKIIδ additionally contained the entire regulatory region as well as a part of the unstructured linker to the association domain. The boundaries used for the crystallized proteins are shown in the boxed sequence inserts in Figure 1A and are indicated in the sequence alignment in Figure S1. As expected, based on the high sequence homology, all structures exhibited a high degree of structural similarity. The activation segments were all well-ordered and helix αC was correctly positioned for catalysis as indicated by formation of the conserved salt bridge between E60 located in αC and lysine K41, which is a hallmark of the active kinase conformation [30] (Figure S2).
Similar to the C. elegans orthologue, the substrate binding site is blocked by the regulatory domain in all human CaMKIIs (Figure 1B). However, dimeric association as in CeCaMKII, i.e., mediated by the regulatory domain, was not observed for any of the autoinhibited human isozymes. As mentioned above, the C termini of the crystallized γ and δ isozymes extended well beyond the Ca2+/CaM binding site (in CaMKIIδ, including the linker region up to S333). However, the autoinhibitory helices in all isozymes were structured only up to residue 302. Different crystal forms were observed, containing one (CaMKIIγ), two (CaMKIIα and CaMKIIδ), or four molecules (CaMKIIβ) in the asymmetric unit. However, no indication of a conserved dimer interface was evident within the crystals with packing contact regions typically involving small and diverse surface areas. However, evidence that dimerization occurs in solution was found using analytical ultracentrifugation (AUC) sedimentation velocity experiments for all CaMKII catalytic domains (exemplified by CaMKIIδ in Figure 1C), thus supporting studies that identified inactive CaMKII as paired (dimers) in cells [19]. The affinities, as estimated from the proportions between the areas of the peaks for monomeric and dimeric species, were weak (KD of 200–600 µM), but due to the high effective concentrations of catalytic domains in the context of the holoenzyme, the observed interactions are likely to be biologically relevant (Table S2). Similar association constants were observed for all human CaMKIIs independent of the construct length, suggesting that dimerization is not mediated by the regulatory domain in human CaMKIIs. To test this hypothesis, we repeated the AUC experiments in the presence of an isolated regulatory domain peptide that spans the autoinhibitory region as well as the Ca2+/CaM binding site (CaMKIIδ residues 282–310). Interestingly, no change in the proportion between the peaks was observed (red trace in Figure 1C), suggesting that binding of the peptide did not interfere with dimerization in human CaMKIIs. Binding of the peptide led to a significant shift towards smaller sedimentation coefficients, suggesting that interaction with the peptide induces a conformational change that leads to increased friction. Based on these data, it is tempting to speculate that binding of the peptide displaces and subsequently causes unfolding of the inhibitory helix and the Ca2+/CaM binding site, as observed in the structure of the Ca2+/CaM complex.
We were interested in exploring the structural consequences of Ca2+/CaM binding on CaMKII and determined the structure of the CaMKIIδ/Ca2+/CaM complex (Figure 2B). The structure of the complex comprised the catalytic and the regulatory domain of CaMKII (residues 1–333) and full length human calmodulin, and was refined at 1.9 Å resolution to an R/Rfree of 16.1 and 19.9%, respectively (Table S1). In the complex, the regulatory region no longer interacted with its corresponding catalytic domain (Figure 2B). Instead, the conformation of the inhibitory region adopted an extended conformation, in sharp contrast to its helical secondary structure in autoinhibited CaMKII (Figure 2A). The observed extended conformation allowed interaction between the inhibitory region and the substrate binding site of an adjacent catalytic domain. Most notably, T287 was aligned in a position suitable for phosphoryl transfer (Figure 2B). Thus, the structure effectively “captures” CaMKIIδ in the process of transphosphorylation by a neighboring kinase molecule. The Ca2+/CaM binding region exhibits equally significant structural changes compared to the autoinhibited kinase. Although this region displays either an extended or partially disordered conformation in all human autoinhibited CaMKIIs, it adopts an entirely helical secondary structure in the Ca2+/CaM complex (Figure 2C).
The CaMKIIδ/Ca2+/CaM co-crystal structure revealed that phosphorylation at T287 is not the only mechanism that prevents the regulatory region from rebinding to the lower kinase lobe. In the complex, helix αD blocked the access to the binding site for the inhibitory helix by a significant reorientation of this helix with respect to the autoinhibited kinase; E106 and Y107, for instance, are displaced by more than 10 Å from their position (Cα) in the autoinhibited kinase and block the binding groove for the inhibitory helix (Figure 2D). The movement of αD has another important consequence which results in structural changes within the kinase active site: E97, which is oriented away from the ATP binding site in autoinhibited CaMKIIs, was positioned in a conformation that enables coordination of the ATP co-factor in the CaMKIIδ/Ca2+/CaM complex. This transition has been proposed previously based on in silico molecular dynamics simulations using the autoinhibited CeCaMKII structure in which the regulatory region was deleted [18]. It is well known that the affinity of CaMKII for ATP is significantly reduced in the absence of Ca2+/CaM [29],[31],[32]–[34]. E97 is highly conserved in kinases and plays a major role in recruiting ATP by forming interactions with the sugar moiety [35]. Similar conformations involving altered orientations of E97 that lead to kinase inactivation have been observed in autoinhibited structures of CaMK1 [36] and twitchin [37], suggesting that reorientation of αD is a common regulatory mechanism in CaMKs. An animation that illustrates the conformational changes that take place during Ca2+/CaM-dependent CaMKII activation has been embedded in the enhanced version of the manuscript (Datapack S1).
Transphosphorylation of T287 (T286 in CaMKIIα) by a catalytic domain present in the same holoenzyme has been described as the molecular switch that leads to constitutive and Ca2+/CaM-independent CaMKII activity [4],[21],[22]. The sequence flanking T287 represents a typical CaMK consensus substrate site [38]. The structure of the CaMKIIδ/Ca2+/CaM complex revealed how the regulatory region flanking T287 is recognized as a substrate: The arginine residue in position −3 (R284)—a hallmark of CaMK substrate recognition—exhibits two conformations and forms multiple polar interactions with the αD residues E100 and E97 (Figure 3A). The conformations of residues interacting within the substrate binding site were well defined in the electron density (Figure 3B). The structure of this substrate complex is similar to a CeCaMKII/peptide complex published recently [20].
We used analytical ultracentrifugation (AUC) to determine whether the CaMKII/Ca2+/CaM heterodimer forms in solution and to estimate affinities for the T287-substrate interaction. In agreement with our structural data, sedimentation velocity experiments measured on the CaMKIIδ/Ca2+/CaM complex revealed apparent molecular weights that correspond well to the calculated mass of the complex. The experiments also revealed the presence of oligomers containing two or more copies of the CaMKII/Ca2+/CaM heterodimer (Figure 3C). We estimated dissociation constants (KD) of 50 µM and 120 µM, for this self-association for the δ and α isozymes, respectively (Table S2). To further investigate whether the substrate binding pocket was indeed the interacting interface, we performed sedimentation velocity experiments in the presence of a substrate-competitive peptide. When a peptide derived from CaMKIIδ residues 282–310 was included, we observed only free Ca2+/CaM and the CaMKIIδ/Ca2+/CaM heterodimer in solution, but no higher association species. This experiment confirmed that the observed association is mediated by the substrate binding pocket recapitulating what was observed in the crystal structure of this complex.
In inactive synapses, slow autophosphorylation at T306 (T307 in CaMKIIδ) accumulates a CaMKII pool of subunits that cannot be activated by Ca2+/CaM and requires phosphatase activity for reactivation [39],[40]. The sequence flanking T307 represents a site that is incompatible with CaMKII consensus substrate requirements, raising the question as to how this regulatory phosphorylation site would be recognized as a kinase substrate. In the CaMKIIδ structure, T307 was bound to the substrate binding site and oriented in an identical fashion to that observed for T287 in the transphosphorylating complex. The T307-containing region adopts an unusual turn conformation not previously seen in kinase substrate complexes. This unusual binding mode was stabilized by hydrophobic interactions of the conserved residues I304 and L305 that bound inside a deep cavity formed by residues located in the P1 loop and in helix αG (Figure 4A). This hydrophobic anchor allows recognition of this non-consensus substrate site, thus providing insight into how T307 is recognized as a cis-autophosphorylation site.
Ca2+/CaM has the ability to bind to a large number of distinct proteins by adjusting the relative orientation of its EF hands [41]–[43]. Residues in the interface were well-defined and the recognition motif bound tightly within the central cavity of the Ca2+/CaM structure while no stable contacts were made with the kinase domain. The Ca2+/CaM interaction involved the CaMKII residues 296–316. Comparison with the autoinhibited structures of human CaMKII isozymes revealed that a helical secondary structure was induced upon Ca2+/CaM binding for the region C-terminal to L300. The interactions between CaMKII and Ca2+/CaM were largely mediated by hydrophobic contacts in the interface of Ca2+/CaM and the CaMKII binding helix. Phosphorylation of T306 in CaMKIIα (T307 in other CaMKII isoforms), located within the Ca2+/CaM binding domain, has been shown to inhibit Ca2+/CaM binding [29],[25]. Moreover, Ca2+/CaM binding to CaMKII with unphosphorylated T306 effectively prevents phosphorylation of this residue [26]. These data are in agreement with our co-crystal structure that showed that both threonine residues (T306/T307) were deeply buried within the Ca2+/CaM complex (Figure 4B).
We used isothermal titration calorimetry (ITC) to compare the affinities of Ca2+/CaM to autoinhibited CaMKII catalytic domains with those of the isolated regulatory domain and catalytic domains of CaMKI and CaMKIV: Human CaMKII kinase domains bound Ca2+/CaM with affinities between 1.6 and 3.4×106 Mol−1 (KD: 0.6–0.3 µM) (Figure 4C, Table 1). The determined binding affinities were in agreement with affinities determined for the full-length enzyme [33]. Ca2+/CaM bound to CaMKI and CaMKIV with >20-times higher affinity. Interestingly, these values compared well with the affinity of the isolated Ca2+/CaM binding domain (CaMKIIδ residues 296–315) (97.3×106 Mol−1; KD: 10.2nM). Comparison with the affinity of the isolated regulatory domains suggests that an energy barrier of ∼2.3 kcal/mol is associated with the release of the inhibitory helix and the Ca2+/CaM trapping mechanism [33]. A unique feature of the CaMKII interaction is the unfavorable (positive) binding enthalpy. The observation that this thermodynamic fingerprint of the binding to Ca2+/CaM is shared by all human CaMKIIs, but neither by CaMKI nor CaMKIV, underlines the mechanistic differences between these classes of CaMK.
Oligomerization into large ring-like structures is a unique feature of CaMKIIs and the oligomeric state is crucial for rapid autophosphorylation in trans by catalytic domains present in the same oligomer [44]. The discrepancy between the dodecameric (12-mer) structures determined by electron microscopy [13],[14] and the tetradecameric (14-mer) assembly revealed by the crystal structure of the isolated oligomerization domain of the C. elegans orthologue [15] prompted us to crystallize the oligomerizaton domains of human CaMKII isozymes. Here we present the oligomerization domains of the human γ and δ isozymes that were refined at 2.7 and 2.8 Å resolution, respectively. The quality of the final model was substantially improved by averaging the electron density maps using non-crystallographic symmetry. While isolated domains of the δ isozyme were tetradecameric, the CaMKIIγ oligomerization domain crystallized in its dodecameric state, thus providing a model for the oligomerization state observed in full-length CaMKIIs. The oligomerization domain formed a hexameric structure with a diameter of 120 Å and a height of 60 Å surrounding a central cavity of only 17 Å (Figure 5A). Thus, the main consequence of the insertion of an additional subunit per ring in the tetradecameric assembly is a considerable widening of the central cavity, to about 33 Å.
Each oligomerization domain in the ring also contains a deep cavity that has been suggested to represent a peptide binding site, based on structural homology with peptide-binding domains present in nuclear transport factors and scytalone dehydratase [15]. In support of this hypothesis is the fact that a glycine or an acetate molecule, respectively, originating from the crystallization solution were partially occupying this putative binding site in the structures of the CaMKIIγ and CaMKIIδ oligomerization domains.
We used the structures of the dodecameric oligomerization domain, the CaMKIIδ/Ca2+/CaM complex and the structures of the inactive catalytic domains, to construct models for the full-length enzyme (Figure 5B). In our model of the autoinhibited protein, catalytic domains located in both hexameric rings associate. The pairing was chosen arbitrarily whilst based on the orientation of the helices in the dodecameric oligomerization domain. Binding of Ca2+/CaM triggers unfolding of the inhibitory helix and releases the kinase domain. This structural reorganization allows T287 to bind to an adjacent kinase domain, leading to autophosphorylation and restructuring of the kinase domain to form a conformation with high affinity for ATP due to reorientation of αD. Once T287 has been phosphorylated, the kinase domain is released fully active and independent of Ca2+/CaM, and with its substrate binding site accessible.
Four regulatory features distinguish the regulation of CaMKII isozymes from other CaMKs. Firstly, CaMKIIs form large oligomers bringing catalytic domains into close proximity to facilitate rapid autoactivation. Secondly, autophosphorylation of T287 generates Ca2+/CaM-independent sustained activity. Thirdly, phosphorylation of T287 increases the affinity of CaMKII for Ca2+/CaM by more than 10,000-fold, an effect known as CaM-trapping [45]. Fourthly, phosphorylation of T306/T307 leads to prevention of CaMKII activation, due to an interference with Ca2+/CaM binding. Comparison of the active CaMKII with its inactive autoinhibited structures provides a structural model for the unique switch-like mechanism of CaMKII autoactivation, as well as for its inactivation by autophosphorylation of T307. The different molecular mechanisms of CaMKII activation and inactivation are depicted in Figure 6, and we discuss here the structural background of these regulatory events in the light of the determined structures.
Autophosphorylation of T307 and its neighbor T306 prevents binding of Ca2+/CaM, resulting in inhibitory autophosphorylation [44]. Moreover, knock-in mice expressing mutants of CaMKIIα incapable of phosphorylation of T306 (e.g. T306A, corresponding to T307 in CaMKIIδ) identified this residue as an important site regulating synaptic plasticity [46]. However, if T287 is phosphorylated prior to T307, Ca2+/CaM-independent activity is induced, whereas prior phosphorylation of T307 prevents activation and autophosphorylation of T287. This illustrates that the sequence of autophosphorylation events is a critical component of CaMKII regulation [47]. In agreement with the observation that ATP binding is impaired in autoinhibited CaMKII, we observed only slow autophosphorylation activity of CaMKIIδ in vitro minimizing premature phosphorylation of T307 (unpublished data). This phosphorylation is stimulated by interaction with CASK [25],[26]. Based on our model of inactive CaMKIIs, it is likely that binding of CASK leads to an allosteric rearrangement similar to the one observed upon release of the inhibitory helix that results in CaMKII activation and autophosphorylation of T306/T307 by the described non-canonical binding mode of the non-consensus substrate site at T307 in inactive CaMKIIδ.
A recent report showed that Angiotensin II-induced oxidation of the methionines M281/M282 leads to sustained activity of CaMKII in the absence of Ca2+/CaM, inducing apoptosis in cardiomyocytes [48]. The structure of inactive CaMKII suggests that oxidation of M282 would lead to steric clashes with the lower lobe, resulting in destabilization of the autoinhibited state and Ca2+/CaM-independent activation. In the CaMKIIδ/Ca2+/CaM complex, M282 binds into a tight hydrophobic pocket in the substrate binding site, suggesting that oxidation of this residue also interferes with substrate binding and autophosphorylation of T287 (see Figure 3A and 4A). Taken together, our structural data suggest that Met oxidation interferes with both the CaMKII inactive state and autophosphorylation of T287, thus supporting a role of M282 oxidation in CaMKII regulation [48]. The key discoveries of this study were conformational changes that take place upon binding of Ca2+/CaM, resulting in activation of the kinase. These structural rearrangements were identified by detailed comparison between the structures of all autoinhibited CaMKIIs and the CaMKII/Ca2+/CaM co-crystal structure. Interestingly, CaM did not interact with any other region of the catalytic domain of CaMKII outside the helical recognition motif. The absence of a specific docking site for Ca2+/CaM on the catalytic domain proper presumably allows this versatile modulator to interact with a large number of highly diverse proteins by inducing structural rearrangements in its target enzymes. Recently, the structure of a complex of Ca2+/CaM with DAPK1 (Death Associated Protein Kinase 1) revealed multiple interactions of Ca2+/CaM with the upper and lower kinase lobe and binding of Ca2+/CaM in an extended conformation [49]. In addition, the substrate binding site in the DAPK1/Ca2+/CaM complex was occluded by Ca2+/CaM, suggesting that a conformational change would need to take place to fully activate DAPK1. CaMKIIs are only distantly related to DAPKs, and it seems that the interaction with Ca2+/CaM and the mechanism of activation of these two CaMKs are fundamentally different.
All four CaMKII isozymes were crystallized with ATP-competitive inhibitors that are relatively non-selective. However, their binding modes (see Figure S3) may nonetheless provide valuable chemical starting points for structure-based design of selective and potent inhibitors for the treatment of diseases. CaMKIIs have been implicated in heart failure [50], arthritis [51], and certain types of cancer [52]–[54]. The detailed comparison of the large body of structural information presented here provides the first insight into how an intact CaMKII catalytic/regulatory domain interacts with Ca2+/CaM. However, issues such as how catalytic domains pair together or how activation resulting from trans-phosphorylation is propagated in the holoenzyme would be best addressed by structures of full length CaMKII, an effort that is ongoing in our laboratory.
Expression constructs comprised the following residues of human CaMKs: CaMKIδ 1–333, CaMKIIα 13–301, CaMKIIβ 11–302 and 358–498, CaMKIIδ11–335 and 334–475, CaMKIIγ 5–317 and 387–527, CaMKIV 15–340, and calmodulin 1–152 were cloned into the T7 expression vector pNIC28-Bsa4 by ligation-independent cloning. Proteins were expressed in Escherichia coli BL21(DE3)R3 as fusions to a Tobacco Etch Virus (TEV)–cleavable N-terminal His6 affinity tag. Cells were re-suspended in lysis buffer (50 mM HEPES pH 7.5 at 25°C, 0.3 M NaCl, 20 mM imidazole) in the presence of a protease inhibitor mix (Complete, EDTA-free Protease Inhibitor Cocktail, Roche Diagnostics Ltd.) and lysed using an EmulsiFlex-C5 high pressure homogenizer (Avestin) or, alternatively, by sonication at 4°C. The lysate was bound to a Ni-NTA column, extensively washed (50 mM HEPES pH 7.5, 300 mM NaCl, 20 mM imidazole) and eluted using the same buffer containing 200 mM imidazole. Proteins were dephosphorylated in vitro with the addition of 50 mM MnCl2 and λ-phosphatase overnight at 4°C. The eluted protein was pooled, concentrated and applied to either a Superdex 75 or 200 16/60 HiLoad gel filtration column equilibrated in 50 mM HEPES, pH 7.5, 300 mM NaCl, 10 mM dithiothreitol (DTT). Additional purification by ion-exchange chromatography (HiTrap Q in 50 mM Tris pH 8.8, using 0.1 to 0.7 M NaCl gradients) was used where purification was insufficient. Purity was monitored by SDS-polyacrylamide gel electrophoresis and final samples were concentrated to 10–15 mg/ml. Dephosphorylation.of CaMKIIγ was inefficient, preventing TEV-cleavage due to phosphorylation of the TEV recognition motif. The alpha and gamma isozymes were crystallized in fusion with the His6-TEV tag.
Crystals were obtained by the sitting drop vapour diffusion method at 4°C using conditions included in Table S1. Inhibitors used for co-crystallization were added to the protein solutions prior to crystallization at a final concentration of 1 mM. Diffraction data were collected on beam-line X10SA at the Swiss Light Source (SLS, Paul-Scherrer Institute, Villigen, Switzerland) from crystals flash-frozen in liquid nitrogen.
Data were indexed and integrated using MOSFLM [55] and were scaled using SCALA [56]. Structures were phased by molecular replacement using PHASER [57] with the coordinates of either the C. elegans model of inactive catalytic domain [18] or the refined CaMKIIγ, respectively, as search models. Refinement was carried out using REFMAC5 [58] employing appropriate geometric and non-crystallographic restraints. The models and structure factors have been deposited with PDB accession codes listed in Table S1.
Sedimentation velocity experiments were carried out on an Optima XL-I Analytical Ultracentrifuge (Beckman Instruments, Palo Alto, CA) equipped with a Ti-50 rotor. Protein samples were studied at a various concentrations in 10 mM HEPES pH 7.5, containing 300 mM NaCl, 1 mM CaCl2 and 5 mM DTT at 4°C, employing a rotor speed of 50,000 rpm. Radial absorbance scans were collected in one-minute intervals using a double-sector cell. 300 µl aliquots were loaded into the sample channels of double-channel 12-mm centerpieces and 310 µl of buffer into the reference channels. Data were analyzed using SEDFIT [59],[60] to calculate c(s) distributions. SEDNTERP was used to normalize the obtained sedimentation coefficient values to the corresponding values in water at 20°C, . Translational frictional ratios were calculated from the s20,w values, using:where M is the molecular weight, is the partial specific volume, NA is Avogadro's number and is the sedimentation coefficient corrected to the standard conditions of density, ρ0, and viscosity, η0, of water at 20.0°C, and extrapolated to infinite dilution. Sedimentation equilibrium experiments were performed at 4°C and at a number of protein concentrations. Dissociation constants, Kd, were calculated, respectively, from the fitted apparent association constants, Ka,obs, according to the equationwhere d is the optical pathlength and ε290 the extinction coefficient at 290 nm.
ITC data were measured using a VP-ITC titration microcalorimeter from MicroCal, LLC (Northampton, MA). For all experiments the proteins used were dialyzed against 20 mM HEPES pH 7.5, containing 150 mM NaCl, 5 mM DTT and 1 mM CaCl2. Data were measured at 10°C by titrating Ca2+/CaM into CaMKII catalytic domains or peptides into Ca2+/CaM, respectively. The inhibitory peptide sequence used for titrations was ARRKLGAILTTMLATRNF, corresponding to Ca2+/CaM binding domain residues 296–315 in CaMKIIδ. Each experiment consisted of a first injection of 2 µl followed by 29 injections of 10 µl injected during 20 s and a spacing of 280 s between injections. Blank titrations (Ca2+/CaM into buffer) were subtracted from the titration data. Data were normalized and evaluated using ORIGIN with a single binding site model. Thermodynamic parameters were calculated using: ΔG = ΔH−TΔS = −RTlnKB, where ΔG, ΔH and ΔS are the changes in free energy, enthalpy and entropy of binding, respectively.
|
10.1371/journal.pntd.0000308 | Schistosoma mansoni Tegument Protein Sm29 Is Able to Induce a Th1-Type of Immune Response and Protection against Parasite Infection | Schistosomiasis continues to be a significant public health problem. This disease affects 200 million people worldwide and almost 800 million people are at risk of acquiring the infection. Although vaccine development against this disease has experienced more failures than successes, encouraging results have recently been obtained using membrane-spanning protein antigens from the tegument of Schistosoma mansoni. Our group recently identified Sm29, another antigen that is present at the adult worm tegument surface. In this study, we investigated murine cellular immune responses to recombinant (r) Sm29 and tested this protein as a vaccine candidate.
We first show that Sm29 is located on the surface of adult worms and lung-stage schistosomula through confocal microscopy. Next, immunization of mice with rSm29 engendered 51%, 60% and 50% reduction in adult worm burdens, in intestinal eggs and in liver granuloma counts, respectively (p<0.05). Protective immunity in mice was associated with high titers of specific anti-Sm29 IgG1 and IgG2a and elevated production of IFN-γ, TNF-α and IL-12, a typical Th1 response. Gene expression analysis of worms recovered from rSm29 vaccinated mice relative to worms from control mice revealed a significant (q<0.01) down-regulation of 495 genes and up-regulation of only 22 genes. Among down-regulated genes, many of them encode surface antigens and proteins associated with immune signals, suggesting that under immune attack schistosomes reduce the expression of critical surface proteins.
This study demonstrates that Sm29 surface protein is a new vaccine candidate against schistosomiasis and suggests that Sm29 vaccination associated with other protective critical surface antigens is the next logical strategy for improving protection.
| Schistosomiasis is the most important human helminth infection in terms of morbidity and mortality. Although the efforts to develop a vaccine against this disease have experienced failures, a new generation of surface antigens revealed by proteomic studies changed this scenario. Our group has characterized the protein Sm29 described previously as one of the most exposed and expressed antigens in the outer tegument of Schistosoma mansoni. Studies in patients living in endemic areas for schistosomiasis revealed high levels of IgG1 and IgG3 anti-Sm29 in resistant individuals. In this study, confocal microscope analysis showed Sm29 present in the surface of lung-stage schistosoluma and adult worms. Recombinant Sm29, when used as vaccine candidate, induced high levels of protection in mice. This protection was associated with a typical Th1 immune response and reduction of worm burden, liver granulomas and in intestinal eggs. Further, microarray analysis of worms recovered from vaccinated mice showed significant down-regulation of several genes encoding previously characterized vaccine candidates and/or molecules exposed on the surface, suggesting an immune evasion strategy of schistosomes under immune attack. These results demonstrated that Sm29 as one of the important antigens with potential to compose a vaccine against schistosomiasis.
| Schistosomiasis mainly occurs in developing countries and is the most important human helminth infection in terms of global mortality. This parasitic disease affects more than 200 million people worldwide causing more than 250,000 deaths per year [1]. Furthermore, schistosomiasis causes up to 4.5 million DALY (disability adjusted life year) losses annually [2]. Recently, King et al [3] have associated schistosomiasis with anemia, pain, diarrhea, exercise intolerance, and under-nutrition that results from chronic infection. Current schistosomiasis control strategies are mainly based on chemotherapy but, in spite of decades of mass treatment, the number of infected people remains constant [4]. Extensive endemic areas and constant reinfection of individuals together with poor sanitary conditions in developing countries make drug treatment alone inefficient [5]. Many consider that the best long-term strategy to control schistosomiasis is through immunization with an anti-schistosomiasis vaccine combined with drug treatment [6]. A vaccine that induces even a partial reduction in worm burdens could considerably reduce pathology and limit parasite transmission [7].
Sm29 protein has been characterized by our group [8], proving to be a membrane-bound antigen on adult worms that is strongly recognized by IgG1 and IgG3 antibodies of naturally resistant individuals and patients resistant to re-infection living in endemic areas for schistosomiasis in Brazil. Recent studies, using microarrays and RT-PCR, showed that Sm29 is among the 16% most highly expressed genes in S. mansoni, and probably serves an important function in the surface biology of this parasite [9],[10]. Proteomic analysis of the S. mansoni tegument composition identified Sm29 as one of the integral proteins to be consistently found in the outer surface [11]. Therefore, the next step would be to investigate humoral and cellular immune responses induced by Sm29 in vaccinated mice and protection studies.
Herein, we determined that Sm29 is present on the tegument of lung-stage and male and female adult worms of S. mansoni by confocal microscopy. Besides, rSm29 induced a Th1-type of immune response in mice and reduction in worm burden and liver pathology.
C57BL/6 and TLR4 KO female mice, 6–8 weeks old, were obtained from the Federal University of Minas Gerais (UFMG) animal facility. All procedures involving animals were approved by the local Ethics Committee on Animal Care (CETEA-UFMG). Cercariae of S. mansoni (LE strain) were maintained routinely in Biomphalaria glabrata snails at Rene Rachou Research Center (Fiocruz, Brazil) and prepared by exposing infected snails to light for 2 hrs to induce shedding of parasites. Cercarial numbers and viability were determined using a light microscope prior to infection. The protocols involving animals used in this study were approved by the Federal University of Minas Gerais Ethics Committee in Animal Experimentation (CETEA No. 023/2005).
The recombinant Sm29 was produced and purified as described previously [8]. Briefly, the Sm29 cDNA fused with a C-terminal 6x histidine was produced in E.coli using the pET21a expression vector (Novagen, NJ, USA). The recombinant Sm29 was purified in an affinity column and dialyzed against PBS pH 7.0.
Adult worms used in confocal microscopy studies were recovered from perfused mice and lung-stage schistosomula were prepared according to the method described by Harrop & Wilson [12]. Parasites fixed in Omnifix II (Ancon Genetics, St Petersburg, FL, USA) were used in whole mount or in section assays. For sections assays, 7 μm slices were deparaffinized with xylol series. Parasites were blocked with 1% BSA in PBST (Tween 20 0.05%) for 1 hr and incubated with anti-rSm29 serum diluted 1∶20 in blocking buffer. Serum from non-immunized mice was used as a negative control. Samples were washed three times with PBST and incubated with anti-mouse IgG antibody conjugated to Alexa Fluor 594 (Molecular Probes, CA, USA) diluted 1∶100 in blocking buffer containing Phalloidin Alexa Fluor 488 (Molecular Probes) to stain actin microfilaments. The samples were washed four times and mounted in antifade reagent (Prolong gold–Molecular Probes). For whole mount assays, parasites were blocked with 1% BSA, 0.1% Triton X-100, 0.1% azide in PBS for two hours at 4°C under agitation. Blocked parasites were incubated with anti-Sm29 serum diluted 1∶80 in the blocking buffer during 16hrs at 4°C under agitation. Samples were washed six times with PBST and incubated with anti-mouse and phalloidin, both diluted 1∶300, during 4 hours at 4°C under agitation. The samples were washed five times and mounted in 90% glycerol and 10% TRIS, pH 9.0. The parasites were visualized in a Zeiss 510 Meta confocal microscope using an immersion objective. All the parameters and microscope settings used were maintained throughout the process.
Cercariae obtained from Biomphalaria glabrata snails exposed to light for two hours were transformed mechanically in skin-stage schistosomula according to Ramalho-Pinto et al [13]. First, cercariae were incubated on ice for 30 minutes and centrifuged for 3 minutes, 1000 rpm, 4°C. The cercariae were resuspended in 1ml of cold ELAC (Earle's salts plus lactalbumin hydrolysate) contain 0.5% lacto albumin, 1% penicillin/streptomycin and 0.17% glucose. The tails were broken by vortex in high speed during 2 minutes. After this, the tails were removed through six washes with ELAC. The schistosomula were incubated for 1 hour and 30 minutes at 37°C in ELAC and washed with apyrogenic physiologic saline. For the tegument removal, the schistosomula were submitted to vortex using high speed, two times of eight minutes each, in a CaCl2 0.3 M solution. The sample was centrifuged at 1000 rpm for 1 minute. The supernatant was collected and centrifuged at 26500 rpm for 1 hour at 4°C. The pellet was dialyzed against physiologic saline 1.7%.
Two micrograms of purified skin-stage schistosomula tegument were submitted to SDS- PAGE 12%, transferred to a nitrocellulose membrane (Millipore) and blocked 16 hours at room temperature with TBST (0.5M NaCl, 0.02M Tris pH 7.5, 0.05% tween, 5% skim milk). The membranes were washed tree times with TBST and probed with anti-Sm29 or naive mice serum diluted 1∶500 in TBST for 4 hours at room temperature. The membranes were washed six times and probed one hour with anti-mouse IgG conjugated to alkaline phosphatase (Invitrogen) diluted 1∶10000 in TBST. After six washes, the membranes were developed using ECF subtract (GE Healthcare) and visualized in a Storm apparatus (GE Healthcare).
Six to eight week old female C57BL/6 mice were divided into two groups of ten mice each. Mice were subcutaneously injected in the nape of the neck with 25 μg rSm29 on days 0, 15 and 30. Sm29 protein concentration for vaccination was determined as previously described [14]. The recombinant protein was formulated with Freund's adjuvant (complete Freund's adjuvant/CFA for the first immunization and incomplete Freund's adjuvant/IFA for the boosters). In the control group, PBS with Freund's adjuvant was administered using the same immunization protocol.
Fifteen days after the last boost, mice were challenged through percutaneous exposure of abdominal skin for 1 h in water containing 100 cercariae (LE strain). Forty-five days after challenge, adult worms were perfused from the portal veins. Two independent experiments were performed to determine protection levels. The protection was calculated by comparing the number of worm recovered from each vaccinated group with its respective control group, using the formula:where PL = protection level, WRCG = worms recovered from control group, and WREG = worms recovered from experimental group.
Following perfusion for the recovery of the schistosomes, fragments of the intestine (terminal ileum) from each animal were separated and transferred to the Petri dishes containing saline as previously demonstrated [15]. The intestines were opened lengthwise and the excess mucus was removed. One-centimeter fragments were weighed and placed between a glass slide and a plastic cover. The preparation was inverted and pressed on a rubber surface padded with filter paper. Then, the slices were examined with a microscope (100 x) and all the eggs were counted. Quantitative oograms were obtained, using the formula: number of eggs/grams of tissue = number of eggs in intestinal fragment/weight of fragment in mg. A qualitative oogram evaluation was performed, in which the developmental stages of the eggs were classified as described previously.
The liver fragments were collected from the same animals analyzed in oogram studies and fixed in 10% paraformaldehyde. The fragments were processed for paraffin embedding and histopathological sections performed using microtome at 6–7 μm and stained in a slide with hematoxilin-eosin (HE). The number of granulomas was obtained from the liver sections using 10x objective in a microscope. The area from each liver section was calculated using capture in scanner followed by analysis in the KS300 software connected to a Carl Zeiss image analyzer.
Following immunization, sera of ten mice from each experimental group were collected at two-week intervals. Measurement of specific anti-Sm29 antibodies was performed using indirect ELISA. Maxisorp 96-well microtiter plates (Nunc, Denmark) were coated with 5 μg/ml rSm29 in carbonate-bicarbonate buffer, pH 9.6 for 16 h at 4°C, then blocked for 2 at room temperature with 200 μl/well PBST (phosphate buffer saline, pH 7.2 with 0.05% Tween-20) plus 10% FBS (fetal bovine sera). One hundred microliters of each serum diluted 1∶100 in PBST was added per well and incubated for 1 h at room temperature. Plate-bound antibody was detected by peroxidase-conjugated anti-mouse IgG, IgG1 and IgG2a (Southern Biotechnology, CA, USA) diluted in PBST 1∶10000, 1∶5000 and 1∶2000, respectively. Color reaction was developed by addition of 100 μl per well of 200 pmol OPD (o-phenylenediamine, Sigma) in citrate buffer, pH 5.0 plus 0.04% H2O2 for 10 min and stopped with 50 μl of 5% sulfuric acid per well. The plates were read at 495 nm in an ELISA plate reader (BioRad, Hercules, CA).
Cytokine experiments were performed using splenocyte cultures from individual mice immunized with rSm29 plus CFA/IFA (n = 5 for each group). Splenocytes were isolated from macerated spleens of individual mice 10 days after the third immunization and washed twice with sterile PBS. After washing, the cells were adjusted to 1×106 cells per well for IL-4, IL-10, IFN-γ and TNF-α assays in RPMI 1640 medium (Gibco, CA, USA) supplemented with 10% FBS, 100 U/ml penicillin G sodium, 100 μg/ml streptomycin sulfate, 250 ng/ml amphotericin B. Splenocytes were maintained in culture with medium alone or stimulated with rSm29 (25 μg/ml) or concanavalin A (ConA) (5 μg/ml) as previously described [14],[16]. The 96-well plates (Nunc) were maintained in an incubator at 37°C with 5% CO2. For cytokine assays polymyxin B (30 μg/ml) was added to the cultures and this treatment completely abrogate the cytokine response to LPS as previously described [17]. Culture supernatants were collected after 24 h of ConA stimulation, 48 h of rSm29 stimulation for IL-4 and TNF-α analysis and 72 h of rSm29 stimulation for IL-10 and IFN-γ. The assays for measurement of IL-4, IL-10, IFN-γ and TNF-α were performed using the Duoset ELISA kit (R&D Diagnostic) according to the manufacturer's directions.
Innate immune responses were assessed using peritoneal macrophages culture obtained from C57BL/6 and TLR4 KO mice that had received an intra-peritoneal injection of thioglycolate four days earlier. The macrophages were recovered through peritoneal washes with PBS and adjusted to 1×106 cells per well for IL-12p40 assays in RPMI 1640 medium (Gibco) supplemented with 5% FBS, 100U/ml penicillin G sodium and 100 μg/ml streptomycin sulfate. Macrophages were maintained in culture with medium alone or stimulated with rSm29 (25 μg/ml) or LPS purified from E. coli (1 μg/ml). Culture supernatants were collected after 24 hrs of stimulation. The assays for measurement of IL-12p40 were performed using the Duoset ELISA kit (R&D Diagnostic) according to the manufacturer's directions.
Total RNA was obtained using Trizol reagent (Invitrogen) according to the manufacturer's instructions. RNA was quantified using the Nanodrop ND-1000 UV/Vis spectrophotometer and RNA integrity was checked by electrophoresis with an Agilent 2100 Bioanalyzer. One microgram of total RNA from each sample was amplified using the T7-RNA polymerase-based linear amplification protocol. Three micrograms of aminoallyl-modified amplified RNA was labeled with Cy3 or Cy5 by indirect labeling (GE Healthcare).
The Cy3- and Cy5-labeled RNA samples were combined and hybridized overnight to cDNA microarray slides using ASP hybridization chambers (GE Healthcare) at 42°C and the protocols were followed as recommended by manufacturer [18]. Slides were scanned at 5 μm resolution, 100% laser power (GenePix 4000B, Molecular Devices, USA) and PMT levels were adjusted in order to obtain similar average intensities of red and green signals. Each array contained 4,000 different elements, spotted in duplicate on the left and the right sides of a glass slide (GEO accession: GPL3929). The cDNA clones spotted on the arrays were selected to represent 4,000 unique transcripts that were identified in the large-scale S. mansoni transcriptome project [19].
RNA from the pool of adult worms recovered from mice that had been immunized with the rSm29 vaccine (test sample) was hybridized against RNA from control adult worms recovered from mice treated only with adjuvant (control sample). Each assay consisted of two separate hybridizations, the first with Cy5-labeled test sample and Cy3-labeled control, and the second with dye-swap in order to account for dye bias; for each probe, the intensity of the test sample was computed as the average intensity of the Cy5 and Cy3 measurements; the same was computed for the control sample. Each assay provided two average values for the test and two for the control, arising from the duplicated copies of each gene that are spotted on the left and right halves of the glass slides [18]. A total of four hybridizations were performed, corresponding to two sets of technical replicate assays. Thus, a total of four different average intensity values for the test and four for the control sample were obtained for each probe on the array.
Data was extracted from images using the Array Vision 8.0 program. We used a locally weighted linear regression (LOWESS) algorithm to correct for systematic bias due to small differences in the labeling and/or detection efficiencies between the fluorescent dyes [20]. Normalized log2 ratios of the intensities of test/control were used for the statistical analysis with the Significance Analysis of Microarrays tool (SAM) using a 0.14% false discovery rate (FDR), which corresponds to a q-value (analogous to the t-test p-value) of q<0.01, to find significant differentially expressed genes [21]. Genes determined in the previous step were filtered using a median fold-change ≥1.5 as cutoff, where fold-change = 2ˆ|log2 (test/control)|. The fold-change filter was applied after identification of significant differentially expressed genes with the use of SAM, therefore avoiding the bias caused by the use of arbitrary thresholds to trim datasets before significance testing [22].
Regulated genes were manually annotated by similarity using BLASTx and the GenBank nr database with a cut-off e-value = 10−10. Sequences were placed into four categories according to the similarity matches, as follows: tegument proteins, membrane proteins, receptor interacting proteins, lipid metabolism and other functions.
Total RNA was extracted with Trizol reagent (Invitrogen), according to the manufacturer's instructions. RNA samples were treated with DNAse, purified and concentrated using RNeasy Micro Kit (QIAGEN), following the manufacturer instructions and 1.5 μg of each sample was reverse transcribed using the SuperScript® III First-Strand Synthesis SuperMix (Invitrogen). Specific primer pairs (Table S2) were designed by Primer Express Program using default parameters (Applied Biosystem) and arbitrarily named primers 1 and 2. Real-time RT-PCRs were run in triplicates in a volume of 20 μl containing 10 μl of Sybr Green PCR Master Mix (Applied Biosystem), 160 nmol of each primer (primers 1 and 2), 0.30 μl cDNA from reverse transcription. Real-time RT-PCR was performed with the 7300 Real-Time PCR System (Applied Biosystem) using the following cycling parameters: 60°C for 10 min, 95°C for 10 min, 40 cycles of 95°C for 15 sec and 60°C for 1 min, and a dissociation stage of 95°C for 15 sec, 60°C for 1 min, 95°C for 15 sec, 60°C for 15 sec. Real time data was normalized in relation to the level of expression of actin. p-values were determined for the triplicates with Student's t-test, using one tail distribution and heteroscedastic variance.
All microarray data were deposited in the GEO public database under Accession No. GSE10777.
Statistical analysis was performed with Student's t-test for comparison between two experimental groups using the software package GraphPad Prism.
In the present study, we demonstrated the localization of Sm29 in the male and female adult worms and also in the lung-stage schistosomula of S. mansoni using specific antibodies to rSm29 produced in E. coli. The native Sm29 was located exclusively on the surface of lung-stage schistosomula and male adult worms of S. mansoni (Fig. 1A, B, C, D, and Video S1). The female adult worm showed Sm29 located on the surface and also in internal tissues (Fig. 1E). Confocal microscopy analysis also revealed that Sm29 is not present in the cercariae stage of S. mansoni (Figure S1), as expected from the absence of mRNA coding for this antigen in cercariae cDNA library [8]. Additionally, we detected Sm29 in the purified tegument fraction of the skin-stage schistosomula by western blot analysis using specific antibodies to rSm29 (Fig. 2).
To evaluate the level of specific IgG, IgG1 and IgG2a antibodies to Sm29 sera from ten vaccinated animals of each group were tested by ELISA. Significant titers of specific anti-Sm29 IgG antibodies were detected at all time points studied after the first immunization (Figure 3). In order to determine the IgG isotype profile induced by vaccination, specific anti-Sm29 IgG1 and IgG2a antibodies levels were also evaluated. The levels of specific IgG1 and IgG2a increased at 15, 30 and 45 days after the first immunization (Table 1). Furthermore, IgG1/IgG2a ratio was reduced at days 30 and 45 that parallels with elevated anti-Sm29 IgG2a production. The IgG1/IgG2a ratio observed in mice immunized with rSm29 can lead us to speculate that a Th1 type of immune response was induced following vaccination.
Mice vaccinated with rSm29 showed elevated production of IFN-γ, TNF-α and IL-10 and absence of IL-4 (Table 2). Cultures used for measuring the cytokines were treated with polymyxin B to avoid non-specific stimulation due to eventual LPS contamination in the purified recombinant protein [17]. Additionally, rSm29 has stimulated peritoneal macrophages from C57BL/6 (2016±423 pg/ml) or TLR4 KO (1108±197 pg/ml) mice to produce IL-12 (Fig. 4). It is worth to emphasize that rSm29 has activated macrophages to secrete IL-12 which drives Th1 cell development. Protective immunity in mice against S. mansoni infection, before the onset of egg production, is thought to be the result of a Th1 pattern of immune response [23].
To test the vaccine potential of rSm29 against schistosomiasis, we investigated the protection induced by this recombinant antigen in the murine model of S. mansoni infection. C57BL/6 mice were immunized three times with adjuvant-formulated rSm29, and then challenged with 100 S. mansoni cercariae. Control groups received adjuvanted phosphate-buffered saline. Mice vaccinated with rSm29 showed 51% reduction in adult worm burden, 60% reduction in intestinal eggs and 50% reduction in liver granuloma counts compared to control mice (Table 2). The most critical points to be considered about an efficient anti-schistosomiasis vaccine are the reduction in pathology and transmission [7], aspects that are present in the protection profile induced by rSm29 vaccine.
In order to investigate the impact of rSm29 vaccination on S. mansoni, worms that were recovered from vaccinated mice had their gene expression profile analyzed using microarrays. This analysis revealed significant (q<0.01) regulation of 517 genes (Table S1) in worms recovered from mice vaccinated with adjuvanted rSm29 when compared to control mice injected with adjuvanted phosphate-buffered saline; of these, 495 genes (96%) were down-regulated and only 22 genes (4%) were up-regulated. Among down-regulated genes, several of them caught our attention and they fell into four major categories: tegument proteins (Sm23 and CD36-like class B scavenger receptor), membrane proteins (tetraspanin TE736), lipid metabolism (Sm14), receptor interacting proteins (B-cell receptor-associated protein and TGF-beta receptor interacting protein) and others (superoxide dismutase and Sm65) (Table 3). To validate the microarray data, we selected six genes (Sm23, Sm14, superoxide dismutase, CD36-like class B scavenger receptor, B-cell receptor-associated protein and TGF-β receptor interacting protein) and performed real-time RT-PCR. As demonstrated in Figure 5, real-time RT-PCR of all six tested genes confirmed the reduced expression previously determined by microarray analysis. Based on the data shown in Table 3 and Figure 5, we hypothesize that in rSm29 protected mice a reduced expression of these critical surface antigens enables the parasites to adapt to the host in the face of an ongoing antiparasite immune response developed in vaccinated animals.
Schistosomiasis is one of the most important neglected tropical diseases (NTDs), and an effective control is unlikely in the absence of improved sanitation and a vaccine. Recently, the tegument proteome of S. mansoni was characterized, and one study in particular revealed the major proteins that are exposed on live adult parasites such as SmTSP-2 and Sm29 [11]. In this study, we investigated murine humoral and cellular immune responses to rSm29 and tested this molecule as a vaccine candidate.
The strategic localization of Sm29 in the mammalian skin-stage and lung-stage schistosomula and its absence in cercariae suggests that this antigen has an important role in the adaptation of the parasite to the new environment when S. mansoni enters the mammalian host. Previous studies using mice vaccinated with irradiated cercariae showed that the lung is the key site of elimination of this parasite [24]. Sm29 localization on the surface of schistosomula is an important aspect of this antigen regarding its protective properties.
In the mouse model, rSm29 induced high levels of anti-Sm29 IgG after the second immunization and showed reduced IgG1/IgG2a ratio at 45 days after the first immunization (Table 1). This finding suggests a tendency of a Th1 type of immune response induced by rSm29 vaccination. Further, we confirmed by cytokine analysis that rSm29 immunization elicited a Th1-type of immune response characterized by high levels of IFN-γ and no IL-4 and 51% of worm burden reduction. The involvement of IFN-γ in protective immunity to schistosomiasis is well documented in the murine model [25]. In the irradiated cercariae vaccination model, which induces high levels of protection, treatment with monoclonal anti-IFN-γ antibody totally abrogated the protective immunity achieved [26]. Similar results were obtained using IFN-γ knockout mice exposed to the radiation-attenuated vaccine confirming the essential role of IFN-γ in protective immunity against murine schistosomiasis [27]. Further, we tested the ability of rSm29 to induce IL-12 in peritoneal macrophages of TLR4 KO mice. TLR4 is the major receptor for macrophage activation by bacterial LPS [28]. Therefore, using TLR4 KO we obviated any possible effect of LPS contamination in IL-12 synthesis. In TLR4 KO mice, rSm29 has induced the production of large amounts of IL-12 that is the key cytokine involved in Th1 cell development.
Pathology which results from granuloma formation around the eggs in murine schistosomiasis is characterized by Th2-type of immune response and the granuloma size can be reduced by neutralization of IL-4 [29]. Thus, morbidity and mortality in murine schistosomiasis were hypothesized to be developed as a direct consequence of the egg-induced Th2 type of immune response. Herein, the intestinal egg numbers was reduced (60%) following rSm29 vaccination compared to control group. Additionally, the immune response elicited by Sm29 was associated with 50% reduction in granuloma counts. Since we detected IL-10 production following splenocyte activation with rSm19, we hypothesize here that this cytokine might be regulating Th2 responses and/or preventing the development of highly polarized Th1 responses and therefore, reducing inflammation and liver pathology [30],[31].
In order to investigate the impact of rSm29 vaccination on S. mansoni, worms that were recovered from vaccinated mice had their gene expression profile analyzed using microarrays. In worms recovered from mice vaccinated with adjuvanted rSm29 when compared to a control group injected with adjuvanted phosphate-buffered saline, 495 genes (96%) were down-regulated and only 22 genes (4%) were up-regulated. Among down-regulated genes, many of them encode surface antigens and proteins associated with immune signals suggesting that under immune attack schistosomes reduce the expression of critical surface proteins. We propose here that down-regulation of expression of important surface antigens is an important strategy of the parasite resulting in escape from host immune surveillance.
As observed in Table 3, Sm23, tetraspanin TE736 (a paralog of Sm-TSP-2), Sm14, Sm65 and superoxide dismutase are significantly (q<0.01) down-regulated in adult-worms recovered from rSm29 protected mice; it is noteworthy that these genes encode important schistosome antigens that are recognized by immune mice and humans. Sm14 is able to induce cellular immune responses in individuals from endemic areas for schistosomiasis [32]. Sm23 and Sm65 are antigens recognized by sera from schistosome infected patients [33],[34]. Moreover, Sm14 and Sm23 induce partial protection in the murine model for schistosomiasis [35],[36]. Regarding tetraspanins, they belong to a family of membrane proteins, such as Sm-TSP-2, that is currently one of the most important vaccine candidates for schistosomiasis [37]. Further, vaccination of mice with naked DNA construct containing Cu/Zn cytosolic superoxide dismutase showed significant levels of protection compared to a control group [38].
It is noteworthy that no significant up-regulation of heat shock or chaperone genes was detected in schistosomes recovered from rSm29 vaccinated mice (Table S1), therefore excluding a common stress response. Rather, lowering the expression of critical surface antigens seems to be an important and specific strategy developed by schistosomes to reduce the damage caused by the host immune system. Therefore, a combined antigen vaccination with Sm29 and one or more of the proteins encoded by these down-regulated critical surface antigens might be the next logical step, thus circumventing the adaptive response of the parasite and eventually boosting the protective effect of vaccination.
Among the down-regulated genes (Table 3) were those encoding protein domains from TGF-β receptor interacting protein, B-cell receptor-associated protein, CD36-like class B scavenger receptor, all of which are associated with host immune response. These parasite gene products could interact with or modulate the host immune response. For example, the scavenger receptors are present on different cells such as macrophages, platelets and neutrophils where they are involved with innate immunity by facilitating phagocytosis, cell adhesion and pathogen recognition. Recently, it was demonstrated that schistosome CD36-like class B scavenger receptor binds to host low density lipoproteins [39]. From the parasite perspective, it would not be difficult to envisage that a reduced expression of S. mansoni scavenger receptor-like molecules would decrease pathogen recognition and cell adhesion by host cells.
Additionally, we found down-regulated genes involved with parasite carbohydrate and lipid metabolism, protein biosynthesis, intracellular signaling cascade, among others (Table S1). Down-regulated genes include SmINSIG (an ortholog of the mammal Insulin Induced Gene), a recently described gene in S. mansoni [40] that has a key role in HMG-CoA reductase degradation. HMG-CoA reductase is vital for egg production by S. mansoni [41]. An alternative hypothesis to our findings is that down-regulation of these genes may affect the development or survival of parasites in vaccinated animals, thus reducing worm burden, although the worms recovered from mice vaccinated with adjuvanted rSm29 are morphologically similar to those recovered from control mice. Additionally, haemoglobinases and Sm29 gene expression were unaltered in worms recovered from rSm29 immunized animals. Recently, Dillon et al [42] in an attempt to identify antigens responsible for protection induced by irradiated-cercariae vaccination performed microarray expression analysis of irradiated and normal worms. These investigators detected down-regulation of genes involved in neuromuscular activity and cell cycle. Their hypothesis is that attenuated parasite might have an extended stay in the host enhancing immune priming against exposed antigens [42]. Their findings may explain why irradiated-cercariae can elicit protective immunity when normal cercariae do not.
Recent advances in antigen discovery with preclinical studies showing promising efficacy have reinvigorated the case for a schistosomiasis vaccine. The irradiated-cercariae vaccine and the use of recombinant antigens, such as Sm-TSP-2 and Sm29, that induce approximately 50% worm burden reduction, have demonstrated that a vaccine to schistosomes is achievable [37]. Furthermore, we identified an ortholog of Sm29 in the ESTs of the Asian schistosome S. japonicum that shares more than 50% identity in the amino acid sequence, indicating that a vaccine based on Sm29 might also be effective against S. japonicum [43]. Another important issue is the selection of a suitable adjuvant and/or delivery system to induce the appropriate immune responses. Herein, we used rSm29 with Freund´s adjuvant, however, it is not suitable for human application. Experiments are underway testing rSm29 with CpG or CpG plus alum as suggested by McManus & Loukas [44]. As a final conclusion of this work, we believe that Sm29 is an efficient vaccine candidate against schistosomiasis in light of the data obtained from murine studies. The best long-term strategy to control schistosomiasis might be immunization with anti-Sm29 vaccine associated with other protective surface antigens and possibly combined with drug treatment.
|
10.1371/journal.pbio.1001784 | Low Frequency of ESRRA–C11orf20 Fusion Gene in Ovarian Carcinomas | The identification of recurrent gene fusions in common epithelial cancers—for example, TMPRSS2/ERG in prostate cancer and EML4/ALK in nonsmall cell lung carcinomas—has raised the question of whether fusion genes are pathogenetically important also in ovarian carcinomas. The first recurrent fusion transcript in serous ovarian carcinomas was reported by Salzman et al. in 2011, who used deep paired-end sequencing to detect the fusion gene ESRRA–C11orf20 in 10 out of 67 (15%) serous ovarian carcinomas examined, a finding that holds great promise for our understanding of ovarian tumorigenesis as well as, potentially, for new treatment strategies. We wanted to test how frequent the ESRRA/C11orf20 fusion is in ovarian carcinomas of all subtypes, and therefore examined a series of 230 ovarian carcinomas of which 197 were of the serous subtype and 163 of the 197 were of stages III and IV—that is, the very same carcinoma subset where the fusion transcript had been found. We performed PCR and high-throughput sequencing analyses in search of the fusion transcript. We used the same primers described previously for the detection of the fusion and the same primer combination, but found no ESRRA/C11orf20 fusion in our series. A synthetic DNA plasmid containing the reported ESRRA/C11orf20 fusion was included as a positive control for our PCR experiments. Data from high-throughput sequencing of 23 ovarian carcinomas were screened in search of alternative partner(s) for the ESRRA and/or C11orf20 gene, but none was found. We conclude that the frequency of the ESRRA/C11orf20 gene fusion in serous ovarian carcinomas of stages III and IV must be considerable less than that reported previously (0/163 in our experience compared with 10/67 in the previous study). At the very least, it seems clear that the said fusion cannot be a common pathogenetic event in this tumor type.
| The identification of characteristic fusion genes in cancer helps us to understand how a particular cancer arises and also to improve classification and diagnosis, with a view to develop specific medical treatments that target exactly those aberrant molecules that trigger the disease. A fusion transcript presumed to arise from a chromosomal rearrangement involving the ESRRA and C11orf20 genes has previously been described to be present in 15% of serous ovarian carcinomas—the first fusion transcript to be associated with this common and often fatal cancer. We assessed 163 similar ovarian carcinomas for the presence of the ESRRA–C11orf20 fusion transcript, plus a further 67 ovarian carcinomas of different histologic subtypes/grades, to see if these tumors were characterized by the same fusion. Surprisingly, we found no ESRRA–C11orf20 transcripts in any of the 230 carcinomas. The question as to whether fusion genes contribute to the pathogenesis of ovarian carcinoma therefore remains open.
| Cancer of the ovary makes up 30% of all malignant diseases of the female genital tract. Prognosis is poor, with a mean 5-year survival rate in Europe of 32%. This unfavorable outcome is largely attributable to a lack of early warning symptoms and signs and also a lack of diagnostic tests that allow early detection. As a result, approximately 70% of patients present with advanced stage, metastatic disease [1].
A number of specific genes have been identified as playing a role in ovarian carcinogenesis; the ones that have received the most attention are BRCA1 and BRCA2 followed by TP53. In addition, integrated genomic analysis of ovarian carcinomas has identified four ovarian cancer transcriptional subtypes, three microRNA subtypes, four promoter methylation subtypes, and a transcriptional signature associated with survival duration [2], attesting to the genetic complexity of these tumors.
The identification of recurrent gene fusions in common epithelial cancers—for example, TMPRSS2/ERG in prostate cancer [3] and EML4/ALK in nonsmall cell lung carcinomas [4],[5]—has raised the question of whether fusion genes are pathogenetically important also in ovarian carcinomas. Salzman et al. [6] reported the first recurrent fusion transcript in serous ovarian carcinomas. They used deep paired-end sequencing to detect the fusion gene ESRRA–C11orf20 in 10 out of 67 (15%) serous ovarian carcinomas examined, a finding that holds great promise for our understanding of ovarian tumorigenesis as well as, potentially, for new treatment strategies. The fusion was brought about by rearrangements in the long arm of chromosome 11, in subband 11q13.1. The gene ESRRA (estrogen-related receptor alpha) encodes a nuclear receptor that is closely related to the estrogen receptor, whereas its partner is but an open reading frame sequence. Because ESRRA and C11orf20 (also known as TEX40) normally lie only 11 kb apart, it is possible that the rearrangement leading to their fusion is an incidental consequence of another functionally important genetic event or that it is merely a “passenger” to other structural rearrangements.
To test how frequent ESRRA/C11orf20 fusion is in ovarian carcinomas of all subtypes, we performed PCR analysis of 230 ovarian carcinomas, of which 197 were of the serous subtype and 163 of the 197 were of stages III and IV—that is, the very same carcinoma subset examined by Salzman et al. [6].
The PCR analysis of the 230 ovarian carcinomas showed no fusion transcript for the ESRRA/C11orf20. A synthetic DNA plasmid containing the reported ESRRA/C11orf20 fusion was included as a positive control for our PCR experiments and was the only sample showing the transcript and demonstrating, at the same time, the validity of the experiments (Figure 1).
We also performed high-throughput sequencing of 23 ovarian carcinomas (already tested by PCR analysis), of which 10 were serous, five endometrioid, four clear cell, three mucinous, and one of a mixed endometrioid and undifferentiated subtype. Each sample was sequenced to yield about 60∼70 million reads using the Illumina HiSeq 2000 instrument. We extracted from the raw data all sequences containing the last 20 bp before the putative break of the ESRRA exon 2 gene sequence, getting 2,705 reads in total. We also found 58, 59, and 49 reads containing the first 20 bp of the C11orf20 exon 3, exon 4, and exon 5 gene sequences, respectively (Table 1). From the extracted ESRRA- and C11orf20-specific sequences, none contained sequences of both ESRRA and C11orf20. The comparison was performed by investigating if the ESRRA-specific sequences contained C11orf20 exon 3, 4, or 5 sequences and vice versa. It is possible to argue that the fusion gene, if present, should be driven by the ESRRA promoter, and therefore that the fusion gene read counts should be more similar to the high ESRRA ones than to the low C11orf20 ones. As a result, assuming the presence of the fusion, the C11orf20 reads should have been totally dominated by the fusion, something that was not seen (Table 1). Furthermore, all 2,705 sequences were used in a Blast search to verify their identity. The Blast search identified specific ESRRA and C110rf20 sequences but revealed no sequences containing both ESRRA and C11orf20 gene sequences. When searching in the same series of sequenced carcinomas (n = 23) for involvement of either the ESRRA or C11orf20 in alternative fusions—that is, with other partner(s)—none was found.
We therefore conclude that the frequency of the ESRRA/C11orf20 gene fusion in serous ovarian carcinomas of stages III and IV must be considerable less than that reported by Salzman et al. (0/163 in our experience, compared with 10/67 in their study) [6]. We have no explanation for the frequency differences observed. It is important to note that the difference in frequency calculated above is based only on adenocarcinomas of stages III and IV—that is, 163 tumors—as the remaining 67 tumors were of different histological subtypes or serous adenocarcinomas of grades I and II, which would not necessarily be expected to carry the same genetic fusion.
Looking into the possible mechanism—that is, chromosomal rearrangement(s)—by which the ESRRA/C11orf20 fusion could have originated, it seems that a simple deletion or inversion could not alone have produced it. Both genes are located 5′ to 3′ from centromere to telomere on 11q, with C11orf20 proximal to ESRRA (Figure 1); therefore, to get a fusion in which ESRRA is 5′ in a chimeric transcript would require a tandem duplication with a breakpoint in the central region. Regardless of how it may have been generated, however, it seems clear that the said fusion cannot be a common pathogenetic event in this tumor type.
The tumors were surgically removed at The Norwegian Radium Hospital from 1999 to 2010. The RNA was extracted using Trizol reagent according to the manufacturer's instructions (Invitrogen, Grand Island, NY), and its quality was checked by Experion Automated Electrophoresis System (Bio-Rad Laboratories, Hercules, CA). cDNA was synthesized using the Iscript advanced cDNA synthesis kit for RT-qPCR (Bio-Rad). Quality was checked using the TaqMan Gene Expression Assays for actin B (ACTB) and Glyceraldehyde 3-phosphate dehydrogenase (GAPDH). To measure the expression of ACTB and GAPDH, the assays Hs99999903_m1 and Hs99999905_m1, obtained from Applied Biosystems (Life Technologies, Carlsbad, CA), were used and run on a CFX96 Real-Time System (Bio-Rad).
For the first RT-PCR reaction we used the G1P1-FWD (5′-GGCATTGAGCCTCTCTACATCA-3′) mapping between 240 and 261 bp in the ESRRA gene (accession number NM_004451 version 4) and REV_pair3 (5′-GGGTCAGGCTTGGGTCTG -3′) located between 681 and 698 bp of the C11orf20 (accession number NM_001039496 version 1) combination of primers—that is, the same primers as Salzman et al. [6]. The PCR cycles were as follows: initial denaturation at 94°C for 30 s, followed by 30 cycles, 15 s at 94°C, 30 s at 55°C; and 60 s at 70°C [6]. For the nested RT-PCR, the primers were G1P2-FWD (5′-AAAGGGTTCCTCGGAGACAGAGA-3′) located between 290 and 312 base pairs in the ESRRA gene (accession number NM_004451 version 4) and F1-REV (5′-TAATTCACGTACAGCCTCTTGCTCCG-3′) mapping between 597 and 622 bp of the C11orf20 gene (accession number NM_001039496 version 1) [6]. The cycles were as follows: 15 s at 94°C, 30 s at 55°C, and 60 s at 72°C [6]. The nested PCR was run for 30 cycles. A synthetic DNA plasmid containing the reported ESRRA/C11orf20 fusion was included as a positive control in our PCR experiments.
High-throughput sequencing was performed on 23 ovarian carcinomas, of which 10 were serous, five endometrioid, four clear cell, three mucinous, and one of a mixed endometrioid and undifferentiated subtype. A total of 3 µg of RNA was sent for high-throughput pair-end RNA-sequencing to the Norwegian Sequencing Centre at Ullevål Hospital (www.sequencing.uio.no). We used paired-end HTS with an average sequence read of 60–70 millions. We analyzed the sequences only with respect to the ESRRA and C11orf20 genes. The last 20 bp of the ESRRA exon 2 gene sequence before the putative break and the first 20 bp of the C11orf20 exon 3, exon 4, and exon 5 gene sequences have been extracted from the raw data (fastq-files) and further analyzed for putative gene fusions.
|
10.1371/journal.pntd.0006487 | 2,4-Diaminothieno[3,2-d]pyrimidines, a new class of anthelmintic with activity against adult and egg stages of whipworm | The human whipworm Trichuris trichiura is a parasite that infects around 500 million people globally, with consequences including damage to physical growth and educational performance. Current drugs such as mebendazole have a notable lack of efficacy against whipworm, compared to other soil-transmitted helminths. Mass drug administration programs are therefore unlikely to achieve eradication and new treatments for trichuriasis are desperately needed. All current drug control strategies focus on post-infection eradication, targeting the parasite in vivo. Here we propose developing novel anthelmintics which target the egg stage of the parasite in the soil as an adjunct environmental strategy. As evidence in support of such an approach we describe the actions of a new class of anthelmintic compounds, the 2,4-diaminothieno[3,2-d]pyrimidines (DATPs). This compound class has found broad utility in medicinal chemistry, but has not previously been described as having anthelmintic activity. Importantly, these compounds show efficacy against not only the adult parasite, but also both the embryonated and unembryonated egg stages and thereby may enable a break in the parasite lifecycle.
| The human whipworm, Trichuris trichiura, infects around 500 million people globally, impacting on their physical growth and educational performance. There are currently huge mass drug administration (MDA) programs aiming to control whipworm, along with the other major soil transmitted helminths, Ascaris and hookworm. However single doses of albendazole and mebendazole, which are used in MDA, have particularly poor effectiveness against whipworm, with cure rates less than 40%. This means that MDA may not be able to control and eliminate whipworm infection, and risks the spread of resistance to albendazole and mebendazole in the parasite population. We are attempting to develop new treatments for parasitic worm infection, particularly focused on whipworm. We report the identification of a class of compounds, diaminothienopyrimidines (DATPs), which have not previously been described as anthelmintics. These compounds are effective against adult stages of whipworm, and also block the development of the model nematode C. elegans. Our DATP compounds reduce the ability of treated eggs to successfully establish infection in a mouse model of human whipworm. These results support a potential environmental spray to control whipworm by targeting the infectious egg stage in environmental hotspots.
| The benzimidazole anthelmintics albendazole and mebendazole are typically used to treat human whipworm infection but are compromised by lack of single-dose efficacy and the risk of resistance. Thus, existing drugs lack sufficient efficacy in mass drug administration (MDA) programs to adequately control or potentially eradicate whipworm. This is a major stumbling block in the WHO target to eliminate morbidity from soil transmitted helminthiases in children by 2020. The current approach for controlling soil-transmitted helminths such as Trichuris is mass drug administration of a single-dose of albendazole or mebendazole, typically repeated annually [1]. However for infection with T. trichiura, single doses of benzimidazoles lead to low cure rates, only 28% and 36% for albendazole and mebendazole respectively [2]. These cure rates are much lower than those of other major human soil-transmitted helminths, Ascaris lumbricoides and hookworm, demonstrating the need for improvements to therapy specifically targeting Trichuris. Indeed modelling studies have demonstrated that, due to these low cure rates, MDA with benzimidazoles does not interrupt whipworm transmission and thus cannot achieve eradication in many settings [3].
Furthermore, the experience from studies on veterinary parasites is that widespread usage of anthelmintics can lead to rapid development of resistance. The discovery of isolates of two species of gastrointestinal nematodes resistant to monepantel only four years after its introduction [4] underlies the real threat to control programmes imposed by emerging drug resistance. Indeed, the combination of MDA programs and low single-dose cure rates may facilitate the development of drug resistance in populations of human parasites. For example, resistance to benzimidazole drugs is caused by point mutations in β-tubulin. Such resistance mutations have been found in T. trichiura after mass drug administration [5], and have been found to increase in frequency after MDA. High frequency of resistance mutations in a population may be associated with lower egg-reduction rates after MDA [6]. Whilst there is no clear evidence yet of widespread anthelmintic resistance in human populations, identification of new drugs with novel mechanisms of actions is warranted to slow the development of drug resistance.
A T. trichiura infection becomes patent when adult female worms, embedded in the gut of the host, start to lay eggs. A single female worm can lay up to 20,000 eggs per day and these unembryonated eggs pass out with the faeces and embryonate in the soil. Development only proceeds further if the embryonated eggs are accidentally consumed via contact of the next host with contaminated food, water or soil. Once ingested, signals for hatching are received when the eggs reach the large intestine [7,8], the newly emerged first stage larvae invade the mucosal epithelium and development to the adult stage of the parasite occurs through a succession of larval moults. Importantly, even when active infections are successfully treated, hosts are constantly re-infected due to high levels of infective eggs present within the water and soil, which can remain viable for years.
Current anthelmintic programmes, including those targeting Trichuris, focus on post-infection eradication of existing infections. However, lifecycle stages outside of the host are also potential viable targets for small molecule drugs. Thus, both preventing egg embryonation and reducing the infectivity of embryonated eggs prior to ingestion offer targets that would break the parasite lifecycle.
The mouse whipworm, T. muris, is a convenient model of the human whipworm as it can be grown routinely in the laboratory via infection of severe combined immune deficiency (SCID) mice. Screening ex vivo adult T. muris has been used to test the anthelmintic activity of a variety of compounds, including approved drugs with the potential for repurposing, and also plant extracts [9–11]. We recently reported a small molecule screen utilising an automated assay for assessment of the motility of ex vivo T. muris adults. This screen led to the identification of a class of molecules termed dihydrobenzoxazepinone (DHB) which demonstrated encouraging activity in this assay, as well as the ability to reduce in vivo infectivity of treated eggs [12]. Most of the active molecules identified from that screen belonged to the dihydrobenz[e][1,4]oxazepin-2(3H)-one chemotype, but interestingly one additional active was from a completely different structural class. Here we report the identification, synthesis and characterisation of a series of compounds belonging to this second chemotype, which has not previously been described as having anthelmintic activity, the 2,4-diamino thieno[3,2-d]pyrimidines (henceforth called diaminothienopyrimidines or DATPs).
All animal experiments were approved by the University of Manchester Animal Welfare and Ethical Review Board and performed under the regulation of the Home Office Scientific Procedures Act (1986) and the Home Office project licence 70/8127.
T. muris worms were cultured using severe combined immune deficiency (SCID) mice, at the Biological Services Facility at the University of Manchester. Male and female mice were infected with 200 infective embryonated T. muris eggs via oral gavage. Thirty-five days later, the mice were sacrificed. Adult T. muris were obtained from the intestine as previously described [12]. Worms were maintained in Roswell Park Memorial Institute (RPMI) 1640 media supplemented with penicillin (500 U/mL) and streptomycin (500 μg/mL) at approximately 37°C and studied on the same day.
Individual adult worms were added to wells containing 75 μL of RPMI-1640 medium, penicillin (500 U/mL), streptomycin (500 μg/mL) plus 1% v/v final concentration of dimethylsulfoxide (DMSO) or compound dissolved in DMSO. Plates were incubated at 37°C, 5% CO2. Motility was determined after 24 hours.
An automated system was used to quantify worm movement. An earlier version of this system has been previously described [13,14]. Two hundred frame movies of the whole plate were recorded at 10 frames per second and then motility determined by an algorithm based on thresholding pixel variance over time [15]. For the hit confirmation and expansion assays, library material was used at a final concentration of 100μM. Dose-response curves were calculated with the four factor log-logistic model using the R package drc [16] or using GraphPad Prism.
Thin layer chromatography (TLC) was performed on aluminium sheets coated with 60 F254 silica. All solvents are used anhydrous unless stated otherwise. NMR spectra were recorded on Bruker AV400 (400 MHz), Bruker AVII 500 (500 MHz) or AVIIIHD 600 (600 MHz) instruments in the deuterated solvent stated. All chemical shifts (δ) are quoted in ppm and coupling constants (J), which are not averaged, in Hz. Residual signals from the solvents were used as an internal reference using the stated deuterated solvent. Infrared spectra were recorded on a Perkin-Elmer 1750 IR Fourier Transform spectrophotometer using thin films on a diamond ATR surface (thin film). Only the characteristic peaks are quoted. Melting points were determined using a Stanford Research Systems EZ-Melt. Low resolution mass spectra (m/z) were recorded on an Agilent 6120 spectrometer and high resolution mass spectra (HRMS m/z) on a Bruker microTOF mass analyzer using electrospray ionization (ESI). Compounds were synthesised from commercially available starting materials, and fully characterised by Infrared (IR) Spectroscopy, Mass Spectrometry (ESI-MS, HRMS-ESI) and Nuclear Magnetic Resonance (1H and 13C NMR). Spectra supporting the synthesis of these compounds are provided in the S1 File.
A mixed-stage C. elegans N2 population was obtained by liquid culture (20°C) according to standard methods [17]. It was then bleached to obtain an egg population with 1.5 mL 4M NaOH, 2.4 mL NaOCl, 2.1 mL water, washed three times, and allowed to hatch in 50 mL S-basal buffer at 20°C overnight to obtain a synchronised L1 population. For the growth assay, 49 μL of S-complete buffer and 1 μL of DMSO or DMSO plus compound were added to each well of 96-well plates. 50 μL of a worm suspension (approximately 20 synchronised L1 worms, 1% w/v E. coli HB101 in S-complete buffer) were then added to each well. Plates were incubated at 20°C before imaging 5 days later. Worm movement was stimulated by inserting and removing a 96-well PCR plate into/from the wells of the assay plate, and then whole plate 200 frame movies were recorded at 30 frames per second. Growth was quantified as a correlate of movement using the same automated system described earlier, which estimates movement for each well by categorising pixels as imaging movement if their variance is greater the mean plus one standard deviation of the variances of all the pixels on the plate [15].
The mouse rectal epithelial cell line CMT-93 (LGC Promochem, Teddington, United Kingdom) was used for these studies. The WST-8 and neutral red cytotoxicity assays were performed as described [12]. Briefly, cells were cultured with test compounds, chlorpromazine positive control or DMSO alone (final compound concentrations of 0 to 100 μM) for 72 hours. The WST-8 assay was then carried out using the Cell Counting Kit– 8 (Sigma Aldrich # 96992) with an incubation time of 2 hours. This time was chosen according to the manufacturer’s instructions and was such that the absorbance of the WST-8 formazan dye was within the linear range of the microplate reader. Following this assay, the medium was exchanged, and the ability of the cells to take up the dye neutral red (concentration 33 μg/mL, incubation time 2 hours) was determined using a microplate reader (absorbance at 540 nm). Results were analysed using GraphPad Prism and fitted using a log-logistic model.
100 infective embryonated eggs were incubated in deionised water with 1% v/v DMSO or test compounds at a final concentration of 100 μM in 1% v/v DMSO for 14 days at room temperature in the dark. Eggs were then washed and resuspended in deionised water. For in vitro hatching assays 100 eggs were added to 1 mL of E. coli bacterial culture grown in LB broth overnight at 37°C shaking at 200 rpm. Egg-bacterial cultures were incubated for 24 hours at 37°C, 5% CO2 and hatching determined following blinding by visual examination under a dissecting microscope. For in vivo hatching assays, 40 eggs were counted under a dissecting microscope and given to a SCID mouse in 200 μL water. At day 15 post-infection mice were culled and the number of L2 larvae present in the caecae and colon enumerated in a blinded manner under a dissecting microscope.
The experiment was conducted in two ‘experimental batches’. For batch one there were 5 mice in each of the DMSO and OX02926 groups. For batch two there were 9 mice in each of the DMSO and OX02926 groups. The raw data (number of worms that established infection in each mouse) are shown separated by batch and treatment in the S1 Fig.
To analyse the data we used a two-way ANOVA (worm number ~ treatment * batch). This showed a significant effect of treatment [F(1,24) = 8.520, P = 0.00752]. It also showed a significant effect of batch [F(1,24) = 10.956, P = 0.00294]. There was no significant interaction between treatment and batch [F(1,24) = 0.296, P = 0.59153]. The significant effect of batch reflected that in both DMSO- and OX02926-treated groups, the number of worms that established infection was generally lower in mouse batch 1 than in batch 2 (S1 Fig). Variation in control worm establishment, which is commonplace in Trichuris infections due to natural variation in egg infectivity from a standardised egg number, was within expected ranges. We therefore took the approach of normalising each data point by dividing by the mean of the DMSO-treated group for that batch. This yielded the % batch normalised infection establishment.
We used a two-way ANOVA (% batch normalised infection establishment ~ treatment * batch) to analyse the data. There was a significant effect of treatment [F(1,24) = 9.569, P = 0.00497] but no effect of batch [F(1,24) = 0.083, P = 0.77618] or interaction [F(1,24) = 0.083 0.77618]. We therefore conducted a post-hoc Tukey HSD test which showed that infection establishment in the OX02926-treated group was significantly different from the DMSO-treated control group (P = 0.0050).
One hundred unembryonated eggs were treated with water, 1% v/v DMSO in water or test compounds at a final concentration of 100 μM (unless stated) with 1% v/v DMSO, in the dark at 26°C, either for 56 days or for shorter periods as described. Images were collected on an Olympus BX63 upright microscope using a 60x / 1.42 PlanApo N (Oil) objective and captured and white-balanced using an DP80 camera (Olympus) in monochrome mode through CellSens Dimension v1.16 (Olympus). Images were then processed and analysed using the image analysis platform Fiji [18].
We have recently described a small molecule screen for new anthelmintics, which used reduction or loss of motility of adult ex vivo T. muris as an endpoint for screening [12]. This screen was designed to identify compounds active on Trichuris as existing drugs are notably less efficacious against this nematode, and it is comparatively evolutionarily distant to nematodes typically screened in anthelmintic-discovery efforts, such as H. contortus, M. incognita and C. elegans. From this primary screen, we found 13 members of the dihydrobenzoxazepinone chemotype, which had not previously been shown to have anthelmintic activity.
In this report we describe the identification of a second new anthelmintic chemotype from this screen. A single 2,4-diaminothieno[3,2-d]pyrimidine (DATP) compound was found in the primary screen. This has been given the identifier OX02926 (Fig 1A). We confirmed this activity in a secondary screen using the same source sample (DMSO solution containing 10 mM compound), and also tested a number of structurally-related compounds from our small molecule collection using the same assay (Fig 1B). The rationale for this was to gain greater confidence in the screening hit and also to explore the activity of “near-neighbour” molecules with the same core 2,4-diaminothieno[3,2-d]pyrimidine structure, which could support the early development of the series. The hit expansion process led to the identification of three further active molecules in this series OX02925, OX03143 and OX03147 (Fig 1C). Two structurally-related compounds were however not active in this assay (Fig 1D).
Having identified promising active DATPs from testing of DMSO solution samples of compounds, these were then resynthesised to obtain authentic, unambiguously characterised samples from which confirmatory screening could take place. Compound resynthesis is important since DMSO solution samples can degrade over time, and this often leads to so-called ‘false positive’ hits [19]. These compounds could be readily prepared in two steps from commercially available 2,4-dichlorothieno[3,2-d]pyrimidine 1, via two sequential nucleophilic aromatic substitution reactions. Treatment of 1 with 2-(2-chlorophenoxy)ethylamine or 2-phenoxyethylamine gave exclusively monosubstitution affording 2a and 2b as a single regioisomer in 64% and 80% yield respectively. Subsequent displacement reaction at C4 gave authentic samples of OX02925, OX02926, OX03143 and OX03147 in 57–91% yield (Fig 2).
The resynthesized hits were then tested in this screen and a concentration-response curve constructed, thereby confirming the anthelmintic activity of several examples of this structural class (Fig 3, Fig 4).
This class has ‘lead-like’ or ‘drug-like’ chemical properties [22], although it is important to note that in the contemporary medicinal chemistry literature this term is usually applied in the context of imparting oral bioavailability characteristics (Fig 4). For agents targeting the gastrointestinal located Trichuris, minimal systemic exposure of the host is desirable and therefore it is critical to differentiate between the conventionally used terminology and parameters for ‘drug-like’ molecules, which affect solubility and permeability, compared to properties that would be relevant to agents targeting other body compartments. Recent literature has described this important caveat for non-peripheral CNS drugs [23], and indeed for anti-parasitic drug development [24]. Importantly, there is considerable scope for generating the large number of structural variants of the DATPs needed for the iterative improvement of compound properties during the downstream lead optimisation process.
Although we are focused on developing an anthelmintic with improved efficacy over existing drugs against Trichuris, activity across the nematode phylum is valuable, particularly as efficacy against economically significant agricultural animal parasites would make further development more economically viable.
We therefore wanted to test the activity of the DATP chemotype against the clade V nematode Caenorhabditis elegans. Using a quantitative development assay to measure the growth of synchronised L1 stage worms, we tested varying concentrations of the compounds to determine the concentration-response effects. As shown in Fig 5, all four DATP compounds were active in this assay with EC50 values from 7–87 μM.
Interestingly, the DATPs display differing trends in activity between the Trichuris and C. elegans assays. At this stage we do not know whether this reflects different potency at the target or different patterns of drug access between the species, but the findings highlight the importance of screening against Trichuris in the search for novel anthelmintic agents targeting whipworm. The data from each of these assays as well as structural descriptors and Lipinski rule assessment for the four DATP compounds and other anthelmintics are summarised in Fig 4. The leading member of the dihydrobenzoxazepinone class OX02983 is shown in Fig 4 for comparison. EC50 values for the two series are currently in a similar range.
It was critical to ensure that this series of compounds showed minimal cytotoxicity towards mammalian cells, and showed selective activity against the parasite. For example, gut cytotoxicity may result in the compounds having too narrow a therapeutic window. Selected examples of the DATPs were assessed for cytotoxicity using the mouse gut epithelial cell line CMT-93 (Table 1). Although, the DATPs exhibited increased in vitro cytotoxicity in these assays compared to the previously reported DHB series [12], an encouraging overall profile was exhibited for these early stage molecules. Furthermore, the nematode cuticle often limits drug access which reduces target engagement by small drug-like molecules [25,26]. This means that compound optimisation to improve uptake through the cuticle may be a fruitful route to improved anti-nematode selectivity, as well as improving the cytotoxicity profile.
It is interesting to note that the activity against Trichuris did not correlate with cytotoxicity, with the most cytotoxic compound (OX03143) showing the lowest activity in the T. muris adult paralysis assay, with an EC50 > 80μM. This suggests that either anti-Trichuris activity is distinct from cytotoxic action, or that differential drug access can be exploited to achieve differential host-parasite activity. Either possibility is encouraging and suggests that continued exploration and iterative improvement of the DATP structure might be anticipated to deliver a more potent anthelmintic with acceptable host toxicity.
Developing novel anthelmintics to disrupt the T. trichiura life cycle at the egg stage represents an exciting and complementary strategy to an oral therapy and is particularly attractive as T. trichiura eggs are highly resistant to extreme temperature changes and ultraviolet radiation, thereby remaining viable in the environment for many years [27]. We assessed whether the DATP derivatives were capable of affecting either infection establishment or embryonation of eggs. We first explored whether the compounds could alter the establishment of infection by soaking embryonated T. muris eggs in the test compounds for 14 days, washing the eggs and then determining infectivity both in vitro and in vivo (Fig 6A).
To determine effects on in vitro hatching, a protocol modified from that previously described [8] was established whereby eggs were induced to hatch when incubated in a culture of Escherichia coli at 37°C. The results are summarised in Fig 6B. Strikingly all DATPs were capable of significantly reducing in vitro hatching compared to the DMSO control.
To extend this finding, we selected OX02926 to test in an in vivo hatching and infection establishment assay, as this compound showed both a significant decrease in in vitro hatching and a small standard deviation between samples. The eggs were soaked as for the in vitro experiment and SCID mice were infected with 40 treated eggs (OX02926 or DMSO) by oral gavage. Egg infectivity was quantified at day 15 post infection by culling the mice and counting the number of established L2 larvae in the gut. All L2 larvae counted had a normal morphology as viewed under a dissecting microscope.
This experiment was carried out in two batches and the raw data are shown in the S1 Fig. Because variation in control worm establishment is commonplace in Trichuris infections due to natural variation in egg infectivity from a standardised egg number, we took the approach of normalising data for each batch relative to the mean of the DMSO-only control group for that batch. This allowed us to determine the effects of OX02926 treatment (a full statistical description is given in the Methods section). The results are shown in Fig 6C. We used a two-way ANOVA (% batch normalised infection establishment ~ treatment * batch) to analyse the data. There was a significant effect of treatment [F(1,24) = 9.569, P = 0.00497] but no effect of batch [F(1,24) = 0.083, P = 0.77618] or interaction [F(1,24) = 0.083 0.77618]. We therefore conducted a post-hoc Tukey HSD test which showed that infection establishment in the OX02926-treated group was significantly different from the DMSO-treated control group (P = 0.0050). Treatment of eggs with OX02926 was able to significantly reduce the burden of worms in vivo by an estimated 40%. This is likely to reflect reduced infectivity of DATP-treated eggs.
The ability of the DATPs to alter the embryonation of T. muris eggs was investigated by soaking unembryonated T. muris eggs collected overnight from live adult T. muris in the test compounds at 26°C for the duration of the embryonation process (56–60 days). During embryonation the first larval stage of the parasite develops within the egg shell (Fig 7A) from a ball of cells (Fig 7B). Treatment with the DATPs OX02925 and OX03147 resulted in a significant increase in the percentage of unembryonated eggs present compared to the DMSO control (Fig 7C). Importantly, although the other DATPs did not alter the percentage of eggs unable to undergo the embryonation process, the larvae that developed were atypical (Fig 7D–7I). These atypical larvae were morphologically altered with the granules present within the larvae appearing less distinct.
As OX03147 had the clearest phenotype with a significant increase in the number of unembryonated eggs, a concentration response study was performed to determine if an effect could be seen at lower treatment doses. Additionally, we repeated the experiment at room temperature to allow for more physiological conditions rather than the constant 26°C utilised in the initial study to standardise conditions across experiments. Although the increased number of unembryonated eggs was only detected at the highest drug dose tested (100 μM) at both 26°C and room temperature (Fig 7C and 7J) striking effects on egg morphology was detectable at concentrations as low as 1 μM with significant larval stunting observed (Fig 7K).
To determine if an effect on embryonation could be observed following a shortened drug exposure we soaked unembryonated eggs in 100 μM OX03147 at 26°C for weeks 0–2, 0–3, 2–4 or 4–6 of embryonation. Although there was no increase in the proportion of unembryonated eggs observed in any treatment group (S2F Fig) there were clear morphological alterations in the L1 larvae within the egg following exposure to OX03147 during weeks 0–3, 2–4 or 4–6 of the embryonation process (S3A–S3E Fig). The most striking observation was the clear larval stunting observed following drug soaking from weeks 0–3 (S2C and S2G Fig). A one-way ANOVA test showed a significant effect of treatment on larval length, F(4, 21) = 3.984, P = 0.0147. A post-hoc Dunnett’s test showed a significant difference in the Weeks 0–3 treatment group compared to the DMSO-only control group (P = 0.0076). This appeared to phenocopy the effect OX03147 had when treated for the duration of the embryonation process at 1 μM (Fig 7K). Additionally, in the 2–4 week and 4–6 week groups, although larval length was not affected, there was evidence of structural alterations in the L1 larvae with a less distinct structure present and altered granulation within the larvae (S2D and S2E Fig).
To the known range of applications of DATPs in medicinal chemistry we can now add anthelmintic activity. This study suggests they have significant potential for further development into dual-acting therapeutic agents for both the reduction of Trichuris egg infectivity, and embryonation in the environment. Thus, their actions on both the embryonated and unembryonated egg stages may enable a break in the parasite lifecycle.
Gastrointestinal nematode parasites remain a significant human health burden. Current anthelmintics lack efficacy and achieve low cure rates, threatening the targets set by the World Health Organisation for control of soil-transmitted helminths [2,28]. In particular, existing drugs have notably low efficacy against T. trichiura, the human whipworm. T. trichiura may be especially difficult to target as it inhabits the large intestine and is in part intracellular [29]. The metabolically active anterior of the worm, the stichosome, is buried in the host epithelial cells lining the gut, affording some protection from orally delivered anthelmintics.
We recently reported a small molecule screen for new anthelmintics targeting the gastro-intestinal (GI) nematode parasite Trichuris muris that identified the dihydrobenzoxazepinone (DHB) chemotype. The DHBs had not previously been ascribed anthelmintic activity [10]. Here, we describe a second class of novel anthelmintic, the diaminothienopyrimidines (DATPs). The potential for this early stage series is significant; their chemical synthesis is facile and lends itself to iterative optimisation, which will facilitate structural modifications aiming, for example, to increase local epithelial penetrance and hence improve efficacy during future development. Furthermore, their straightforward production imparts a favourable cost benefit aspect to the series.
Thienopyrimidines have received much interest in medicinal chemistry as they are bioisosteres for purines, such as the nucleic acid components adenine and guanine. They are also related to quinazolines, an important class of kinase inhibitors, including gefitinib and erlotinib, which act by recognizing the ATP-binding site of the enzyme [30]. Thieno[2,3-d]pyrimidines are a particularly important scaffold, with many reported examples of protein kinase inhibitors, as well as inhibitors of dihydrofolate reductase, kainate receptor agonists, and α1-adrenoreceptor antagonists [31].
The thieno[3,2-d]pyrimidine scaffold found in the compounds reported in this study, has also been investigated. A series of 2-aryl 4-morpholino derivatives have been identified as phosphatidylinositol-3-kinase inhibitors [32], leading to the discovery of the PI3K inhibitor GDC-0941 (pictilisib) [33] and the dual PI3K/mTOR inhibitor DGC-0980 (apitolisib) [34]. The structures of these compounds are shown in the S3 Fig, in comparison with the 2,4-diaminothieno[3,2-d]pyrimidine OX02926. Pictilisib and apitolisib are under development as anti-cancer agents, have been tolerated in Phase I trials for solid tumors, and Phase II trials have commenced [35,36].
A series of 2,4-diaminothieno[3,2-d]pyrimidines have been described as orally active antimalarial agents [37], with activity in the low nanomolar range against Plasmodium falciparum. The structures of these compounds are shown in the S3 Fig in comparison with OX02926. This anti-malarial series was later improved by systematic modification giving improved antimalarial activity, but unfortunately continued hERG inhibition [38]. Whilst our DATP compounds have the same core scaffold as the anti-malarial series, they have different substituents, and in particular lack the 6-aryl substituent that is critical for anti-malarial activity and found in all compounds tested for hERG activity. However, the authors were able to demonstrate that hERG activity could be removed through modification of the C1 substitutuent, suggesting that this potential liability is not instrinsic to the 2,4-diaminothieno[3,2-d]pyrimidine core. We have not yet performed hERG assessment of our compounds, but this will form an important part of the future development of this series.
A series of 2,4-diaminothieno[3,2-d]pyrimidines has also recently been reported as active against the endosymbiotic bacterium Wolbachia, with potential use against filarial nematodes [39]. In neither the anti-malarial or anti-Wolbachia case is the molecular target of the compounds known.
The major goal of our research is to develop a new oral therapy for trichuriasis, which could be widely used in mass drug administration programs leading to the eradication of human whipworm. Such an agent should have a substantially higher single-dose cure rate than the current drugs used in mass drug administration, albendazole and mebendazole. Drug development is long process, and recent work has defined a set of criteria, tailored to neglected infectious diseases, for progression in the hit to lead and lead optimisation stages [40,41]. Our DATP series members are early-stage compounds in the development process. The compounds meet almost all of the criteria for hit selection in neglected diseases, including confirmed activity with resynthesized material, dose-dependent in vitro activity, a tractable chemotype that passes drug-likeness filters such as the Lipinksi rule of five, and an established synthetic route of only two steps [40]. The most pressing weakness of the series is the small selectivity window for their activity against the parasite compared to cytotoxicity in a mammalian cell line. Improving this property for these early stage compounds must be a priority for future development. The DATP compounds also meet some of the milestones in the hit to lead process, particular in terms of drug-likeness and the exploitability of the structure, giving the ability to generate variants and establish the structure-activity relationship and hence improve potency and selectivity [41]. The in vitro activity of OX02926 in the adult whipworm motility assay (EC50 = 27μM, equivalent to 10.2μg/ml) also reaches the activity threshold for lead compounds that has been determined for drug development against the microfilarial nematode Brugia malayi [41]. In summary the DATP series are promising early-stage compounds with a number of lead-like features. Improvement of potency, together with an understanding of parasite/host selectivity and pharmacokinetic properties will be the focus of the next steps of development.
In addition to activity against the adult stage of whipworm, the DATPs were also able to significantly reduce egg hatching, both in vitro and in vivo. These data are in keeping with members of the DHB series, which also were able to inhibit parasite egg hatching. However, unlike the DHB series, we identified members of the DATPs that also significantly reduced the percentage of eggs embryonating ex vivo, with other members of the DATP series appearing to disrupt the embryonation process, resulting in defects in embryonic elongation and abnormal egg shape. Trichuris egg embryonation occurs gradually and the mechanism by which it occurs is currently a poorly understood process. A detailed characterisation of the morphological changes which occur with the Trichuris suis egg during embryonation has been described and other Trichuris species appear to undergo the same process. Once the unembryonated, unsegmented eggs are deposited, the two clear, nuclei-like areas move together and fuse. Cellular division then begins, initially occurring asymmetrically with two blastomeres of unequal size. The larger blastomere then divides again and then subsequently each blastomere divides in two until a blastula formed of many small blastomeres develops. The initial larval differentiation then occurs with the appearance of a motile cylindrical embryo, which gradually turns into an infective larva with its characteristic oral spear. The fully developed larva is no longer motile and is thought to be an L1 larva as no moult is observed within the egg [42]. The embryonation process is temperature sensitive. The effect of temperature on egg embryonation has been characterised in detail in recent years for T. suis eggs with the embryonation process accelerated at 30–32°C compared to 18°C, with degeneration of the eggs rather than embryonation observed at higher temperatures (40°C). At low temperatures (5–10°C) no embryonation occurs, however once these eggs are then transferred to optimal embryonation temperatures normal embryonation proceeds [43]. Similar temperature sensitivity has been described for other Trichuris species including Trichuris trichiura with different species embryonating with different kinetics [44,45]. More research is required to understand the mechanisms behind this embryonation process, which may then allow an even more targeted approach to breaking the life cycle.
Humans become infected with Trichuris via a faecal oral route. Adult parasites in the intestine shed unembryonated eggs, which pass out with the faeces and embryonate in the external environment over a period of five weeks. Eggs can remain viable in the environment for many months [46]. Parasite eggs are only infective if fully embryonated upon ingestion. Thus, the ability of the DATPs to disrupt both the infectivity of embryonated eggs and the embryonation process itself suggests a potential environmental control to decrease Trichuris infection rates in the field without the need to develop and administer a new oral anthelmintic to the infected population.
In particular, it has been noted that the environmental pool of infectious eggs makes those individuals successfully treated, typically once or twice per year, in mass drug administration programs at risk of reinfection [47]. It has therefore been proposed that improvements in sanitation are required in addition to anthelmintic MDA. We suggest that an environmentally-acting, egg-targeting agent, potentially developed from our DATP series compounds, could play a complementary role to help break transmission in parallel with MDA and santitation improvements.
Clearly it is not possible to widely treat large areas of endemic regions with such an environmental control. Instead, we envisage the targeted use of DATPs in the environment at sites of high parasite egg density; these might include for example focusing treatment around pit latrines, as it is known that pit latrines may be a focal point of infection with a high concentration of eggs of soil-transmitted helminths [48]. In a study in Ethiopia, Trichuris trichiura prevalence was higher in communities with greater latrine usage (compared to field or yard defecation), suggesting that basic pit latrines may in some circumstances be ineffective at reducing infection [49]. However improved sanitation facilities generally, including pit latrines, ventilated improved pit latrines, and flush toilets, do reduce STH infection rates [47,50].
Such an egg-targeted agent should have a limited negative effect on the environment, have a suitable formulation for practical delivery, and be able to block egg viability at low concentration in the environment. The DATP series, which damage egg development and infectivity when applied at fairly high concentrations (1 to 100μM) for quite long periods of time (from 2 to 3 weeks to 60 days) show potential for developing such an agent. However these properties need to be improved during future development, while achieving an appropriate safety and environmental profile.
In summary we report the discovery of a new class of anthelmintic, the DATPs, which possesses activity directed against adult stage T. muris parasites and the egg stage. Importantly, as a chemical series the DATPS are notable, since they are relatively facile to produce synthetically thereby presenting considerable scope for structural modifications to improve efficacy and deliver an optimised agent.
|
10.1371/journal.pgen.1008144 | A long noncoding RNA distributed in both nucleus and cytoplasm operates in the PYCARD-regulated apoptosis by coordinating the epigenetic and translational regulation | Long noncoding RNAs (lncRNAs) participate in various biological processes such as apoptosis. The function of lncRNAs is closely correlated with their localization within the cell. While regulatory potential of many lncRNAs has been revealed at specific subcellular location, the biological significance of discrete distribution of an lncRNA in different cellular compartments remains largely unexplored. Here, we identified an lncRNA antisense to the pro-apoptotic gene PYCARD, named PYCARD-AS1, which exhibits a dual nuclear and cytoplasmic distribution and is required for the PYCARD silencing in breast cancer cells. The PYCARD-regulated apoptosis is controlled by PYCARD-AS1; moreover, PYCARD-AS1 regulates apoptosis in a PYCARD-dependent manner, indicating that PYCARD is a critical downstream target of PYCARD-AS1. Mechanistically, PYCARD-AS1 can localize to the PYCARD promoter, where it facilitates DNA methylation and H3K9me2 modification by recruiting the chromatin-suppressor proteins DNMT1 and G9a. Moreover, PYCARD-AS1 and PYCARD mRNA can interact with each other via their 5′ overlapping region, leading to inhibition of ribosome assembly in the cytoplasm for PYCARD translation. This study reveals a mechanism whereby an lncRNA works at different cellular compartments to regulate the pro-apoptotic gene PYCARD at both the epigenetic and translational levels, contributing to the PYCARD-regulated apoptosis, and also sheds new light on the role of discretely distributed lncRNAs in diverse biological processes.
| As the reveal of tens of thousands of long noncoding RNAs (lncRNAs) from mammalian genomes, there is increasing interest to understand how these transcripts function in physiological and pathological processes. The regulatory potential of many lncRNAs has been revealed at specific subcellular location. Nevertheless, many lncRNAs are discretely distributed in different cellular compartments, and the related biological significance remains largely unclear. PYCARD-AS1 is such a “discrete” lncRNA, which exhibits a dual nuclear and cytoplasmic localization. Here, we provide a comprehensive picture of how PYCARD-AS1 works at different cellular compartments to regulate the pro-apoptotic gene PYCARD at both the epigenetic and translational levels, thereby contributing to the apoptotic regulation. Our study adds a different layer to the lncRNA-mediated regulation of gene expression, and strongly suggests that the transcripts of a specific lncRNA may have distinct functionalities at different cellular compartments where they are located.
| The great majority of mammalian genomes are pervasively transcribed, giving rise to tens of thousands of noncoding transcripts, especially long noncoding RNAs (lncRNAs) [1]. LncRNAs participate in a large range of biological processes such as cell differentiation, apoptosis and proliferation, and most of them function by participating in the regulation of gene expression [2, 3]. Unlike mRNAs, which must be localized to the cytoplasm for protein synthesis, lncRNAs exhibit diverse subcellular distribution patterns, ranging from predominant nuclear foci to almost exclusively cytoplasmic localization, and exert distinct regulatory effects at their particular site of action [4, 5]. Thus, the subcellular localization of lncRNAs is critical for their biological function.
A number of studies have implicated the regulatory potential of lncRNAs at specific subcellular location. Determined by the site of action where they are located, lncRNAs may work in cis on neighboring genes or in trans to regulate distantly located genes or molecular targets in the nucleus or cytoplasm [5, 6]. Also, our related studies showed that several nuclear lncRNAs drive tumorigenesis by sequestering the activity of PSF protein in repression of proto-oncogene transcription [7, 8], and revealed that the OCC-1 transcripts localized in cytoplasm inhibit cell cycle transition by modulating the stability of HuR protein [9]. Current studies on lncRNAs have greatly advanced our knowledge of their physiological and pathological roles. Nevertheless, a substantial proportion of lncRNAs are revealed to exhibit a discrete subcellular distribution [10], and the biological significance of discrete distribution of an lncRNA in different cellular compartments remains largely unexplored.
The pro-apoptotic gene PYCARD encodes a signaling factor that consists of an N-terminal PYRIN-PAAD-DAPIN domain (PYD) and a C-terminal caspase-recruitment domain (CARD) and operates in the intrinsic and extrinsic cell death pathways [11, 12]. PYCARD was originally identified to undergo the DNA methyltransferase-1 (DNMT1)-mediated epigenetic silencing in a wide range of human tumors [13–17], and subsequent studies showed that its inactivation is also associated with other epigenetic events such as H3K9 dimethylation [18]. In this study, we further report an lncRNA antisense to PYCARD, named PYCARD-AS1, which exhibits a dual nuclear and cytoplasmic distribution and is required for the PYCARD silencing in breast cancer cells. PYCARAD-AS1 is functionally involved in the PYCARD-regulated apoptosis, and regulates apoptosis in a PYCARD-dependent manner, indicating that PYCARD is a critical function mediator of PYCARD-AS1. Mechanistically, PYCARD-AS1 not only acts in cis to facilitate the recruitment of DNMT1 and histone H3K9 methyltransferase G9a to the PYCARD locus, but also inhibits ribosome assembly in the cytoplasm for PYCARD translation, leading to coordinated regulation of PYCARD expression at the epigenetic and translational levels. Our findings highlight the notion that the transcripts of a specific lncRNA may operate in biological processes by exerting distinct regulatory effects at different cellular compartments.
Through a GenBank search, we identified the gene C16orf98 on the opposite strand of the PYCARD gene, which produces a transcript in a head-to-head orientation relative to the PYCARD mRNA (Fig 1A). Based on the sequence of C16orf98, the experiments of 5′- and 3′-RACE were initiated with total RNA from SKBR3 cells and resulted in a 1,095-nucleotide (nt) antisense transcript of PYCARD (Fig 1B), which is the same as the transcript annotated as PYCARD-AS1. PYCARD-AS1 originates in the second intron 892 nt downstream of the transcription start site (TSS) of PYACRD, ends at nt 676 upstream of the PYCARD TSS, and consists of two exons (Fig 1C). A “directional” RT-PCR assay, which was set up by using gene-specific reverse primers in RT reaction (AS-R and S-R, shown in Fig 1C), showed that neither PYCARD-AS1 nor PYCARD produces longer RNA species spanning their 3’ non-overlapping regions (Fig 1D). In addition, the RT-PCR detection initiated by different primers (S1A Fig), as well as the 3′-RACE experiment, indicated that PYCARD-AS1 is poly(A)-tailed, belonging to an mRNA-like transcript. We next examined the sensitivity of PYCARD-AS1 transcription to the Pol II inhibitor α-amanitin. Following α-amanitin treatment, the levels of PYCARD-AS1 and ACTB transcripts were reduced, whereas those of pre-tRNAtyr and 45S pre-rRNA showed no change (S1B Fig), indicating that PYCARD-AS1 is transcribed by Pol II.
PYCARD-AS1 is annotated as a noncoding transcript in GenBank. In addition, the Coding Potential Calculator tool [19] predicted that PYCARD-AS1 displays no protein-coding potentiality (S1C Fig). Although UniProt showed a putative protein prediction of 204 amino acids for PYCARD-AS1 (S1D Fig), we found that the putative open reading frame (ORF) of PYCARD-AS1 could not be expressed as an N-terminal enhanced green fluorescent protein (EGFP) fusion protein (S1E and S1F Fig). By RNA FISH assay, PYCARD-AS1 was revealed to exhibit a dual nuclear and cytoplasmic distribution (Fig 1E). Moreover, cellular fractionation assay showed that the distribution of PYCARD-AS1 is clearly distinct from that of the nuclear-localized U1 snRNA, the mitochondrially retained 12S rRNA and the protein-coding GAPDH mRNA (Fig 1F). We further separated the chromatin fraction, and found that approximately 25% of nuclear-localized PYCARD-AS1 transcripts were chromatin-enriched (S1G Fig).
Since antisense lncRNAs have been implicated as regulator of their sense counterparts [2, 5], we set out to analyze whether PYCARD-AS1 can regulate PYCARD expression. The transcription level of PYCARD-AS1 was first measured in four breast cancer cell lines, namely, SKBR3, MCF7, MDA-MB-231 and T47D. PYCARD-AS1 transcription was detectable in all cell lines, with SKBR3 expressing it most (S2A Fig). Next, shRNAs were designed to silence PYCARD-AS1 and detect the effect of PYCARD-AS1 knockdown on PYCARD expression in SKBR3 cells. The shRNAs were shown to reduce the levels of total and cytoplasmic PYCARD-AS1 and efficiently act on the nuclear-retained and chromatin-associated PYCARD-AS1 (Fig 2A–2F). Concomitantly, we observed that PYCARD-AS1 silencing increases the PYCARD mRNA and protein levels (Fig 2G and 2H and S2B Fig), indicating a negative regulation of PYCARD by PYCARD-AS1. In parallel, we knocked down PYCARD mRNA with specific shRNA, and found that PYCARD silencing didn’t change the PYCARD-AS1 level (S2C Fig). In addition, PYCARD-AS1 knockdown was shown to have no effect on the mRNA levels of FUS, TRIM72 and PYDC1, three genes neighboring PYCARD and PYCARD-AS1 (Fig 1A and S2D Fig), suggesting that PYCARD-AS1 specifically regulates PYCARD. Also, the negative PYCARD regulation by PYCARD-AS1 was confirmed in MCF7, MDA-MB-231 and T47D cells (S2E–S2G Fig).
The silencing of PYCARD is closely correlated with the defective apoptosis of tumor cells, and its reactivation was revealed to increase the susceptibility of tumor cells to cytotoxic agents [20]. Confirming the regulatory effect of PYCARD-AS1 on PYCARD expression prompted assays on the contribution of PYCARD-AS1 to PYCARD-regulated apoptosis. We tested whether PYCARD-AS1 regulates the sensitivity of SKBR3 cells to paclitaxel, which is also associated with the expression level of PYCARD [21]. Knockdown of PYCARD-AS1 increased PYCARD expression (S2H Fig), and the PYCARD-AS1-knockdown SKBR3 cells showed increased sensitivity to paclitaxel treatment compared with the control cells (Fig 2I and S2I Fig). Remarkably, simultaneous PYCARD knockdown, which neutralized the increased PYCARD expression (S2H Fig), was shown to largely weaken the sensitivity of SKBR3 cells to paclitaxel increased by PYCARD-AS1 silencing (Fig 2I and S2I Fig), suggesting that PYCARD-AS1 regulates apoptosis in a PYCARD-dependent manner.
SKBR3 cells included in the above apoptosis assay were also subjected to global gene expression analysis. Microarray data revealed hundreds of genes that were induced or suppressed more than twofold as a consequence of PYCARD-AS1 knockdown, and showed that the expression of a significant proportion of these differentially expressed genes (91 of 175 PYCARD-AS1 knockdown-induced genes and 37 of 59 PYCARD-AS1 knockdown-suppressed genes) could be reversed at least 1.5-fold in PYCARD-AS1 and PYCARD double-knockdown SKBR3 cells compared with that in PYCARD-AS1 knockdown SKBR3 cells (Fig 2J and S1 Table; GEO accession number: GSE85032). Several target genes identified by microarray were randomly selected for qRT-PCR confirmation (Fig 2K and S2J Fig). The above data indicate that PYCARD-AS1 is involved in the PYCARD-regulated apoptosis, and that PYCARD is a critical downstream target of PYCARD-AS1.
Following on from the above studies, we attempted to determine how PYCARD-AS1 suppresses PYCARD by using SKBR3 cells. DNA methylation and histone H3K9 dimethylation are two types of epigenetic modifications contributing to the PYCARD silencing in tumor cells [13, 18]. Since shRNAs were shown to act on the PYCARD-AS1 in nucleus (Fig 2C and 2D), we first analyzed whether PYCARD-AS1 knockdown can impact the methylation status of the PYCARD promoter, with the PYCARD-AS1 promoter being detected as a control. The results from bisulfite sequencing showed that the PYCARD promoter, but not the PYCARD-AS1 promoter, was partially demethylated by PYCARD-AS1 knockdown (Fig 3A), leading to a relatively hypomethylated state. We also performed ChIP assay to examine the distribution of H3K9me2 across a genomic region of approximately 2.7 kb that covers the PYCARD and PYCARD-AS1 loci. The decrease of H3K9me2 occupancy, which occurred upon PYCARD-AS1 knockdown, was only detected in the region near the PYCARD TSS (Fig 3B, upper). Histone modification H3K27me3 was included as a ChIP control, and the result showed that it was rarely enriched across this genomic region compared with H3K9me2 (Fig 3B, lower).
As detected by Pol II ChIP assay, PYCARD-AS1 knockdown also resulted in an elevation in the initiating Pol II occupancy at the 5′ PYCARD region, as well as an elevation in the elongating Pol II occupancy at the 3′ PYCARD region and the total Pol II occupancy at the 5′ and 3′ PYCARD regions (Fig 3C), indicating an enhanced PYCARD transcription initiation. The effect of PYCARD-AS1 on PYCARD transcription was further dissected via nuclear run-on assay, and the results revealed an increased production of nascent transcript for PYCARD when PYCARD-AS1 was knocked down (Fig 3D). Altogether, the findings described above suggest that PYCARD-AS1 knockdown in SKBR3 cells changes the chromatin status from an inactive state to an active one, thereby inducing PYCARD transcription.
In addition to DNMT1, which had previously been identified [22], G9a, a histone methyltransferase responsible for H3K9 dimethylation, was also revealed to occupy the PYCARD promoter in SKBR3 cells (S3A Fig). PYCARD expression was induced by DNMT1 knockdown, G9a knockdown, or DNMT1 and G9a double-knockdown (S3B Fig). Moreover, PYCARD-AS1 knockdown was shown to cause simultaneous loss of the DNMT1 and G9a occupancy at the PYCARD promoter without influencing their expression levels (Fig 4A and 4B and S3C Fig). These findings indicate that PYCARD is under the control of DNMT1 and G9a, and that PYCARD-AS1 contributes to the recruitment of DNMT1 and G9a to the PYCARD promoter.
A native RIP assay using DNMT1- and G9a-specific antibodies retrieved a substantial amount of PYCARD-AS1 (Fig 4C), revealing a PYCARD-AS1 association with DNMT1 and G9a. Meanwhile, ChIRP assay using tiling oligos against PYCARD-AS1 (Fig 4D) specifically retrieved the PYCARD promoter DNA (Fig 4E), indicating that PYCARD-AS1 is also associated with the PYCARD locus. In accordance with the finding that shRNAs can efficiently knock down the nuclear-retained and chromatin-associated PYCARD-AS1 transcript (Fig 2C and 2D), PYCARD-AS1 knockdown was shown to decrease the PYCARD promoter DNA associated with PYCARD-AS1 transcript (Fig 4F). Together, the above results suggest that PYCARD-AS1 may recruit DNMT1 and G9a to PYCARD promoter via the intermolecular interactions.
We proceeded to elucidate the molecular mechanism whereby PYCARD-AS1 epigenetically regulates PYCARD. Although lncRNAs contain functionally redundant sequences, it is acknowledged that their core functionality generally depends on particular functional region [23]. To map the DNMT1- and G9a-interacting region within PYCARD-AS1, nuclear extract from SKBR3 cells was subjected to limited RNase T1 digestion, so that G residues protected by an RNA binding protein would remain preferentially uncleaved. After DNMT1 or G9a RIP, the enriched RNA fragments, which were associated with DNMT1 or G9a and protected from degradation, were identified by qRT-PCR analysis using primer sets that scanned the PYCARD-AS1 transcript in overlapping ~150 nt-long segments (Fig 5A). As shown in Fig 5B, the PYCARD-AS1 transcript from nt 631 to 1,095 was enriched by DNMT1- and G9a-specific antibodies after RNase T1 treatment, suggesting that the 3′ PYCARD-AS1 domain is the major region responsible for the DNMT1 and G9a interaction.
The above finding was confirmed by an RNA pull-down assay using an in vitro-generated biotinylated 3′ PYCARD-AS1 region, which was shown to bind DNMT1 and G9a with equal efficiency as full-length PYCARD-AS1 (Fig 5C). However, the effort to further narrow down and distinguish the DNMT1- and G9a-binding regions was failed because neither DNMT1 nor G9a could be retrieved by any of the biotinylated RNA segments derived from the 3′ PYCARD-AS1 region (Fig 5C). In addition to the consensus primary sequence, a proper secondary structure is also critical for the function of lncRNAs such as protein binding [24, 25]. Thus, further truncation of the 3′ PYCARD-AS1 region might have destroyed the necessary secondary structure and thus impair its property of association with DNMT1 and G9a. On the other hand, given that DNMT1 and G9a are reported to interact with each other directly [26], it could be speculated that DNMT1 and G9a form a unique complex prior to their binding to PYCARD-AS1. Consistently, our IP assay showed that RNase treatment didn′t destroy the DNMT1–G9a interaction (S4A Fig).
Since there is a substantial degree of reverse complementarity between PYCARD-AS1 and PYCARD sequences, we tested whether the association of PYCARD-AS1 with the PYCARD locus, as demonstrated by ChIRP assay (Fig 4E and 4F), is mediated by direct base pairing, which results in formation of an RNA/DNA hybrid (i.e. R-loop). Total RNA extracted from RNase H- or RNase inhibitor-treated SKBR3 cells was subjected to region-specific qRT-PCR as well as primer walking assay. These analyses indicated that the region of potential RNA–DNA interaction is likely to be contained in a 317-nt fragment within exon 2 of PYCARD-AS1 (nt 576/892 with respect to the PYCARD-AS1 TSS), which is complementary to a region of the first exon of PYCARD (nt 1/317 with respect to the PYCARD TSS), because this PYCARD-AS1 region is sensitive to the RNase H treatment, which degrades RNA in RNA/DNA hybrids (Fig 5D and 5E and S4B Fig).
We next established an RNase-ChIP assay to test whether R-loop formation contributes to the DNMT1/G9a occupancy at the PYCARD promoter, which was revealed to be PYCARD-AS1-dependent in SKBR3 cells (Fig 4A and 4B). RNase T1 treatment, which would degrade the single-stranded PYCARD-AS1 sequence lying outside the 3′ DNMT1/G9a-binding region, led to a separation of the 3′ PYCARD-AS1 region from the PYCARD DNA that abrogated the association between DNMT1/G9a complex and PYCARD promoter; if PYCARD-AS1 interacts with the PYCARD DNA sequence to recruit DNMT1 and G9a as expected, treatment with RNase H would result in the significant release of DNMT1 and G9a from the PYCARD promoter (Fig 5F). We next tested several other DNMT1- and G9a-targeted loci that include KCNQ1 and CDH1, two genes characterized to also associate with the lncRNA KCNQ1OT1 or NEAT1 [27–30]. The results showed that the R-loop-dependent genomic recruitment of DNMT1 and G9a is not confined to the PYCARD locus (S4C and S4D Fig).
As PYCARD-AS1 and PYCARD mRNA constitute a pair of head-to-head overlapping transcripts, we next sought to determine whether they can interact with each other. Affinity pull-down assay showed that the in vitro-generated biotinylated full-length PYCARD-AS1 retrieved a substantial amount of PYCARD mRNA compared with the negative controls including beads alone or EGFP transcript (Fig 6A). In parallel, the deletion mutant of PYCARD-AS1 that lacks the overlapping region (PYCARD-AS1ΔOS, shown in Fig 6C, upper) was tested, and the result showed that deletion of the overlapping region led to an impaired association with PYCARD mRNA (Fig 6A). The interaction between PYCARD-AS1 and PYCARD transcripts was also confirmed within the cellular context by a MS2-RIP assay (Fig 6B); furthermore, an RNase-A assay indicated that the overlapping region of this pair of transcripts was resistant to RNase degradation (Fig 6C, lower). PYCARD-AS1 was revealed to exhibit a dual nuclear and cytoplasmic distribution (Fig 1E and 1F). Interestingly, by cellular component-specific RNase-A assays, the PYCARD-AS1–PYCARD interaction was detectable in both the nucleus and the cytoplasm (S5A Fig).
The interaction between PYCARD-AS1 and PYCARD transcripts implies a post-transcriptional regulation of PYCARD by PYCARD-AS1. lncRNAs have been involved in the stability control of their paired mRNAs [31–37]. However, it seems unlikely that PYCARD-AS1 operates through this strategy because its depletion didn′t change the half-life of PYCARD mRNA in SKBR3 cells, which were treated with α-amanitin in advance to block new RNA synthesis (S5B Fig). In addition, PYCARD-AS1 was demonstrated to have no effect on the subcellular distribution of PYCARD mRNA (S5C Fig).
We next tested whether PYCARD-AS1 gets connected to the PYCARD translation by monitoring the association of PYCARD mRNA with polysomes, the translational entity, in SKBR3 cells with or without PYCARD-AS1 knockdown. SKBR3 cell lysates fractionated through sucrose gradients were subjected to RNA isolation (Fig 6D), and the isolated RNA samples were used to measure the recruitment of PYCARD mRNA on polysomes by qRT-PCR. As shown in Fig 6E (upper), PYCARD-AS1 knockdown drove a shift of PYCARD mRNA toward heavier polysomes, in keeping with increased translation. As a negative control, the distribution of GAPDH mRNA in separation fractions did not change (Fig 6E, lower). Given that the overlapping region of this pair of sense–antisense transcripts is localized at their 5′ ends, we reasoned that PYCARD-AS1 can influence the ribosome assembly on PYCARD mRNA when they interact with each other. To test this, we established an RIP assay to determine whether PYCARD-AS1 knockdown can affect the ratio of ribosome-occupied PYCARD to total PYCARD mRNA, which was expected to eliminate the "contaminating" effect of PYCARD-AS1 knockdown on PYCARD transcription. The results clearly showed that occupancy of RPS6 and L26, the integral components of ribosomal subunits, on PYCARD mRNA was increased by PYCARD-AS1 knockdown (Fig 6F, upper). In parallel, PYCARD-AS1 knockdown was shown to have no effect on the RPS6 and L26 occupancy on GAPDH mRNA (Fig 6F, lower). To further address this issue, we performed a compensation experiment by infecting the PYCARD-AS1-knockdown SKBR3 cells, which had increased mRNA and protein levels of PYCARD (Fig 6G and 6H), with lentivirus expressing the 5′ overlapping region of PYCARD-AS1 (PYCARD-AS1OS) or the PYCARD-AS1ΔOS, which contains several point mutations that disrupt the shRNA target. Neither of the transcripts could compensate for the effect of PYCARD-AS1 knockdown on PYCARD mRNA level (Fig 6G). However, the PYCARD-AS1OS, but not the PYCARD-AS1ΔOS, has the ability to neutralize the PYCARD translation increased by PYCARD-AS1 knockdown (Fig 6H). Collectively, the results suggest that PYCARD-AS1 also suppresses the translation of PYCARD in addition to its transcription.
LncRNAs have been implicated in a large range of biological processes such as apoptosis, and their function is closely correlated with the cellular compartments where they are located. It is interesting that a sizable proportion of lncRNAs are discretely distributed in different cellular compartments [10]. PYCARD-AS1 reported here is such a "discrete" lncRNA, which exhibits a dual nuclear and cytoplasmic distribution, and our current study describes a model whereby the discretely distributed PYCARD-AS1 transcripts link different effector mechanisms to simultaneously operate in the different aspects of PYCARD regulation (Fig 7), contributing to the PYCARD-regulated apoptosis.
We first showed that PYCARD-AS1 can regulate PYCARD at the epigenetic level. Current evidence indicates that interaction with DNA methyltransferase or histone modifier is the major mechanism through which lncRNAs function in epigenetic regulation [2, 38]. Nevertheless, PYCARD-AS1 represents one of the few lncRNAs, which include KCNQ1OT1 and NEAT1 [27, 28, 30], identified to simultaneously modulate DNA methylation and histone modification at the loci of regulated genes. Biochemical interactions between DNA and histone methyltransferases were thought to provide a molecular explanation for the combinatorial pattern of DNA and histone modification in chromatin [26, 39, 40]. The current data further indicate that PYCARD-AS1 interacts with the DNMT1/G9a complex and aids in its recruitment to the PYCARD promoter. The core functionality of lncRNAs relies heavily on the cooperative action of their dispersed functional regions [23, 41]. In the case of PYCARD-AS1-mediated epigenetic regulation of PYCARD, PYCARD-AS1 interacts with DNMT1/G9a complex via its 3′ region, and localizes at the PYCARD promoter via its 5′ region, thereby facilitating the location-specific DNMT1/G9a recruitment. While the detailed mechanism underlying the lncRNA-mediated R-loop formation is not fully understood [42], our results showed that the R-loop structure formed by the 5′ PYCARD-AS1 region substantially contributes to the PYCARD-AS1-mediated PYCARD regulation. On the other hand, the interactions between lncRNA and DNMT1 are documented to have either positive or negative effect on the activity of DNMT1 [43–45]. The "guiding" lncRNAs, including PYCARD-AS1 reported here, can facilitate the DNMT1-mediated DNA methylation by recruiting DNMT1, whereas some other lncRNAs, such as ecCEBPA and Dali, were found to sequester the DNMT1 activity as they compete with the DNA substrate of DNMT1 for DNMT1 binding [46, 47].
PYCARD-AS1 also exerts inhibitory effect on PYCARD translation. Recent studies have shown that through binding mRNAs, lncRNAs, especially antisense lncRNAs, can repress or promote mRNA translation [35, 48–50]. The distinct effects may depend on specific binding sequences and embedded elements in lncRNAs. In the case of PYCARD-AS1-mediated repression of PYCARD translation, PYCARD-AS1 activity depends on the 5′ overlapping sequence, which interferes with the ribosome assembly on PYCARD mRNA. The feature of 5′ overlapping sequence is shared by many other natural antisense transcripts. Nevertheless, the antisense Uchli RNA was reported to exert an opposite effect on the translation of its sense counterpart due to the presence of an embedded inverted SINEB2 element, which facilitates the ribosome binding to mRNA [49]. The paired antisense lncRNA may also affect certain other steps in protein translation. For instance, a binding of the PXN mRNA by its antisense lncRNA PXN-AS1-S can reduce PXN protein synthesis by inhibiting translational elongation [35]. In addition, a recent study reported that through competitive RNA–RNA interaction, an lncRNA is able to attenuate the activity of its paired antisense lncRNA in repression of mRNA translation [50], thereby constituting a finely tuned lncRNA/antisense lncRNA/mRNA translational regulatory axis. The cluster of natural antisense transcripts comprises a surprisingly large fraction of lncRNAs [51]; moreover, the lncRNA–mRNA gene pairs are prevalent in mammalian genomes [52]. Thus, effect on mRNA translation might be a common effector mechanism employed by lncRNAs, especially antisense lncRNAs, to function in biological processes.
Taken together, this study provides an example of how an lncRNA works at different cellular compartments to regulate a specific target gene at multiple levels, contributing to the regulation of apoptosis. The feature of discrete distribution in different cellular compartments is not exclusively exhibited by PYCARD-AS1; instead, it is extensively shared by other lncRNAs [10]. We propose that, as with PYCARD-AS1, many other discretely distributed lncRNAs should also be multifunctional within the cell, and that elucidating their different functionalities at distinct distribution sites will greatly broaden our knowledge of lncRNA biology and provide new insights into their physiological and pathological roles in depth.
PYCARD-AS1 cDNA was synthesized from SKBR3 cells by RT-PCR. For the test of protein-coding potentiality of PYCARD-AS1, the EGFP-coding sequence was inserted into the 3′ end of the putative PYCARD-AS1 ORF, and the fusion gene PYCARD-AS1-EGFP was cloned into the restriction sites Nhe I and Xho I of plasmid pcDNA3.1 (Invitrogen). For lentivirus-mediated RNA interference, complementary sense and antisense oligonucleotides encoding short hairpin RNAs (shRNAs) targeting PYCARD-AS1, PYCARD, DNMT1 and G9a transcripts were synthesized, annealed and cloned into the Age I and EcoR I sites of plasmid pLKO.1 (Addgene). For compensation experiment, cDNA corresponding to the PYCARD-AS1OS or PYCARD-AS1ΔOS was PCR-amplified from the PYCARD-AS1 cDNA and cloned into the Xho I and EcoR I sites of plasmid pLVX (BD Clontech). For MS2-RIP experiment, the MS2-6× fragment was synthesized and fused to the 5′ end of PYCARD-AS1 or PYCARD-AS1ΔOS cDNA, and the resulting fusion constructs (MS2-PYCARD-AS1 and MS2-PYCARD-AS1ΔOS) were cloned into the Xho I and EcoR I sites of plasmid pcDNA3.1; the sequence encoding MS2 coat protein was PCR-amplified from plasmid pMS2-GFP (Addgene) and fused to the 3′ end of FLAG tag-encoding sequence, and the resulting FLAG-MS2 construct was cloned into the Hind III and Xho I sites of plasmid pcDNA3.1. The primers and oligonucleotides used for plasmid construction are shown in S2 Table.
HEK293T, SKBR3, T47D, MDA-MB-231 and MCF7 cells were obtained from the ATCC and cultured in DMEM, MEM or RPMI1640 supplemented with 10% FBS in a 5% CO2 incubator at 37°C. To block cellular transcription by Pol II, SKBR3 cells in culture media were treated with 50 μM α-amanitin (Sigma-Aldrich). For the apoptosis analysis, SKBR3 cells were treated with 5 nM paclitaxel (Sigma-Aldrich) for 72 h. Plasmid transfections were performed using Lipofectamine 2000 (Invitrogen). For lentivirus infection, shRNA-encoding pLKO.1, or pLVX encoding specific PYCARD-AS1OS or PYCARD-AS1ΔOS, was co-transfected with psPAX2 and pMD2.G plasmids (Addgene) into HEK293 cells; the infectious lentivirus was harvested 2 days post-transfection, filtered through 0.45-μm PVDF filters and transduced into SKBR3, T47D, MDA-MB-231 or MCF7 cells. After lentivirus infection, the resulting cell population, but not the isolated single clones, were used for subsequent assays to avoid clone-specific effects.
The full-length PYCARD-AS1 was obtained using 5′- and 3′-RACE System for Rapid Amplification of cDNA Ends (Invitrogen) in accordance with the manufacturer′s instructions. RACE PCR products were separated on a 1.5% agarose gel. Gel products were extracted with a Gel Extraction kit (Foregene), cloned into pMD18-T vector and sequenced bidirectionally using M13 forward and reverse primers. The primers used in the RACE experiments are shown in S2 Table.
RNA FISH was conducted using QuantiGene ViewRNA ISH Cell Assay Kit (Invitrogen) in accordance with the manufacturer′s instructions. In brief, SKBR3 cells cultured on cover slips were fixed, permeabilized and digested by protease to allow target accessibility. A probe set specific for PYCARD-AS1 (designed and supplied by Invitrogen) was added to the cells and hybridization was performed at 40°C for 3 h. After a series of signal amplification with Pre-Amplifier Mix, Amplifier Mix and Label Probe Mix supplied in the kit, cells were counterstained with DAPI and then detected using a fluorescent microscope (Leica).
A total of 1 × 107 cells were washed twice in cold PBS and then incubated in hypotonic buffer (50 mM HEPES, pH 7.5, 10 mM KCl, 350 mM sucrose, 1 mM EDTA, 1 mM DTT and 0.1% Triton X-100) on ice for 10 min. After 5 min of centrifugation at 2,000 g, the supernatant was collected as the cytoplasmic fraction, and after additional washing, the remainder was considered as nuclear pellets, which could be resuspended in lysis buffer (10 mM HEPES, pH 7.0, 100 mM KCl, 5 mM MgCl2, 0.5% NP-40, 10 μM DTT and 1 mM PMSF) to prepare the nuclear lysate. To isolate the chromatin-enriched RNA, the chromatin pellets, as well as the soluble nucleoplasm, was prepared from the nuclear extract as described [53].
RNA samples were prepared from whole cell lysate or specific subcellular fractions using TRIzol reagent (Invitrogen). RNA levels for a specific gene were measured by qRT-PCR (starting with 50–100 ng RNA sample per reaction) using Real-Time PCR Easy (Foregene), in accordance with the manufacturer′s instructions. The qRT-PCR data were normalized to ACTB mRNA, 18S rRNA, mitochondrial-retained 12S rRNA, nuclear-localized U1 snRNA, or chromatin-associated XIST RNA, or presented as a percentage of the total amount of detected transcripts. The primers used in the qRT-PCR are shown in S2 Table.
For global gene expression analysis, 15 μg of biotinylated cDNA synthesized from total RNA was hybridized to the Affymetrix GeneChip® PrimeView™ Human Gene Expression Array at 45°C for 16 h with Affymetrix GeneChip Hybridization Oven 640. After washing and staining with Affymetrix Fluidics Station 450, GeneChips were scanned by the Affymetrix GeneChip Command Console installed in the GeneChip Scanner 3000 7G. Hybridization data were analyzed with the Robust Multichip Analysis (RMA) algorithm using the default Affymetrix settings. Values are presented as log2 RMA signal intensity.
A total of 5 × 105 cells treated with paclitaxel were harvested and stained with Alexa Fluor® 488 Annexin V/Dead Cell Apoptosis kit (Invitrogen), in accordance with the manufacturer′s instructions. Flow cytometry analysis was carried out using an Accuri C6 flow cytometer (BD Biosciences).
Genomic DNA was extracted with the Genomic DNA Isolation kit (Foregene) and the bisulfite conversion reaction was performed using CpGenome Turbo Bisulfite Modification kit (Millipore), in accordance with the manufacturer′s instructions. PCR amplification of bisulfite-treated DNA was carried out with PCR Easy (Foregene). The amplified products were cloned and sequenced. The primers used in the PCR amplification are shown in S2 Table.
The Abs used for immunoblotting were rabbit anti-PYCARD Ab (13833, Cell Signaling), mouse anti-actin Ab (sc-130301, Santa Cruz Biotechnology), mouse anti-DNMT1 Ab (ab13537, Abcam) and goat anti-G9a Ab (sc-22879, Santa Cruz Biotechnology). The Abs used for IP analysis were mouse anti-DNMT1 Ab (ab13537, Abcam) and normal mouse IgG (sc-2025, Santa Cruz Biotechnology). The Abs used for ChIP analysis were rabbit anti-H3K9me2 Ab (4658, Cell Signaling), rabbit anti-H3K27me3 Ab(9733, Cell Signaling), rabbit anti-RNA polymerase II CTD repeat YSPTSPS (S5P Pol II) Ab (ab5131, Abcam), rabbit anti-RNA polymerase II CTD repeat YSPTSPS (S2P Pol II) Ab (ab5095, Abcam), mouse anti-RNA pol II Ab (39097, Active Motif), mouse anti-DNMT1 Ab (ab13537, Abcam), rabbit anti-G9a Ab (ab40542, Abcam), normal mouse IgG (sc-2025, Santa Cruz Biotechnology) and normal rabbit IgG (sc-2027, Santa Cruz Biotechnology). The Abs used for RIP analysis were mouse anti-DNMT1 Ab (ab13537, Abcam), rabbit anti-G9a Ab (ab40542, Abcam), mouse anti-p53 Ab (P6874, Sigma-Aldrich), rabbit anti-RPS6 Ab (ab70227, Abcam), rabbit anti-L26 Ab (ab59567, Abcam), normal mouse IgG (sc-2025, Santa Cruz Biotechnology) and normal rabbit IgG (sc-2027, Santa Cruz Biotechnology). The Ab used for nuclear run-on assay was anti-BrdU Ab (ab1893, Abcam).
A total of 5 × 106 cells were washed twice in cold PBS and pelleted. The pellet was resuspended in lysis buffer (10 mM HEPES, pH 7.0, 100 mM KCl, 5 mM MgCl2, 0.5% NP-40, 10 μM DTT and 1 mM PMSF), incubated on ice with frequent vortexing for 15 min and then the lysate was obtained by centrifugation at 12,000 g for 10 min. Protein concentrations of the extracts were measured by the bicinchoninic acid assay (Pierce). Forty micrograms of the protein was used for immunoprecipitation, or was fractionated by SDS-PAGE, transferred onto PVDF membranes and then blotted.
For immunoprecipitation assays, protein samples were incubated with a specific antibody or control IgG overnight at 4°C. Subsequently, the samples were incubated with 50 μl of protein A agarose beads (Invitrogen) for 4 h at 4°C and then washed three times in washing buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM MgCl2 and 0.5% NP-40). Finally, protein complexes were eluted by SDS buffer (120 mM Tris-HCl, pH 6.8, 20% glycerol and 4% SDS) and then detected by immunoblotting.
For native RIP assays, RNase OUT (50 U/ml, Invitrogen) and a protease inhibitor cocktail (Roche) were added to the lysis buffer, and Ribonucleoside Vanadyl Complex (10 mM, NEB) was added to the washing buffer (the buffer mentioned here is the same as that used in immunoblotting and immunoprecipitation). Following the addition of antibody to the lysate, samples were incubated with 50 μl of protein A agarose bead (Invitrogen) for 4 h at 4°C and then washed three times in washing buffer. The beads were resuspended and treated with proteinase K at 45°C for 45 min. Coprecipitated RNAs were extracted using TRIzol reagent, ethanol-precipitated with Glycoblue (Invitrogen) as a carrier and then detected by qRT-PCR. The data of retrieved RNAs are presented as a percentage of the amount input.
For RIP-based mapping assays, lysates were first mixed with RNase T1 (1 U/ml, Thermo Fisher Scientific), after which standard native RIP assays were performed using an antibody against DNMT1 or G9a. Following extraction of the coprecipitated RNA, the PYCARD-AS1 segments bound by DNMT1 and G9a, and hence protected from RNase T1 digestion and immunoprecipitated, were identified by qRT-PCR analysis using primer sets that scanned the PYCARD-AS1 transcript at ~150-nt-long, overlapping intervals (S2 Table).
For MS2-RIP, pcDNA3.1-MS2-PYCARD-AS1, or pcDNA3.1-MS2-PYCARD-AS1ΔOS, was co-transfected with pcDNA3.1-FLAG-MS2 into SKBR3 cells. After 48 h, cells were subjected to RIP assay with Anti-FLAG M2 Magnetic beads (Sigma-Aldrich) in accordance with the manufacturer′s instructions.
A total of 1.5 × 107 cells were washed with cold PBS and harvested in cold douncing buffer (5 mM MgCl2, 0.5% NP-40, 10% glycerol, and 50 mM Tris-HCl, pH 7.4). After 10 min of incubation on ice, cells were disrupted with 30 strokes of a Dounce homogenizer, centrifuged at 3,300 g for 5 min and washed four times with 1 ml cold douncing buffer. Microscopic analysis indicated that samples contained intact nuclei, cellular debris and some intact cells. The crude nuclei were then resuspended in 100 μl nuclear run-on buffer (5 mM MgCl2, 150 mM KCl, 0.1% sarbyl, 10 mM DTT and 50 mM Tris-HCl, pH 7.4), and were mixed with 1 μl each of 10 mM ATP, GTP, CTP, 1 μl 10 mM BrUTP (Sigma-Aldrich) and 1 μl RNase inhibitor (Thermo Fisher Scientific). Reaction mixtures were pre-incubated on ice for 30 min, then at 28°C for 5 min. The RNA was isolated by TRIzol reagent (Invitrogen), and DNA was removed by DNase I (Promega) treatment. Nascent transcripts were immunoprecipitated with anti-BrdU antibody (Abcam) and subjected to qRT-PCR assays with primers listed in S2 Table.
ChIP analyses were performed as described [7]. For RNase-ChIP assays, 1 × 106 cells were collected by centrifugation, permeabilized in 1 ml of PBST (PBS containing 0.05% Tween 20), and treated with 1,000 U/ml RNase T1 (Thermo Fisher Scientific), 1,000 U/ml RNase H (Thermo Fisher Scientific) or 1,000 U/ml RNase inhibitor (Thermo Fisher Scientific) for 4 h at 25°C. The following procedures were carried out in accordance with the standard ChIP protocol. The genomic DNA in the precipitate was detected by qPCR using the primers shown in S2 Table, and the DNA precipitated by each antibody, including IgG, is presented as a percentage of the amount input.
The collected cells were subjected to permeabilization treatment and then treated with RNase H or RNase inhibitor as described in RNase-ChIP assay. RNA was extracted from the treated cells and subjected to region-specific qRT-PCR as well as primer walking assay using primers listed in S2 Table to test the abundance of specific PYCARD-AS1 regions.
Cells were cross-linked by 1% glutaraldehyde at room temperature for 10 min, followed by three washes in cold PBS. After being snap-frozen by liquid nitrogen and stored at −80°C, cross-linked cells were resuspended in nuclear lysis buffer (10 mM EDTA, 1% SDS and 50 mM Tris-HCl, pH 7.5) supplemented with a protease inhibitor cocktail (Roche), and sonicated until DNA was in the size range of 100~500 bp. Cell lysates and a set of biotin-labelled antisense probes (20 nt in length) were then incubated at 37°C for 4h, with the corresponding sense probes being included as a control. Streptavidin-coupled Dynabeads (Invitrogen) were added to pull down the probes. After washing, the retrieved DNA was isolated using the ChIP DNA Clean & Concentrator kit (Zymo Research) and subjected to qPCR analysis. The probes used for ChIRP are shown in S2 Table.
To synthesize biotin-labelled transcripts, PCR fragments were prepared using forward primers harboring the T7 RNA polymerase promoter. Following purification of the PCR products, biotinylated transcripts were synthesized using MaxiScript T7 kit (Ambion). Biotinylated RNA was heated to 85°C for 2 min, placed on ice for 2 min, supplied with RNA structure buffer (0.1 M KCl, 10 mM MgCl2 and 10 mM Tris-HCl, pH 7.0) and then incubated at room temperature for 20 min. Cell lysates were incubated with 10 pmol of biotinylated transcripts for 3 h at 25°C. Complexes were isolated with streptavidin-coupled Dynabeads (Invitrogen). The retrieved protein and RNA were detected by immunoblotting and qRT-PCR, respectively.
RNase-A assay was performed as described [31]. Briefly, cell lysates were treated with 20 ng/ml RNase A (Thermo Fisher Scientific) at 37°C for 30 min. RNA was extracted from the resultant sample, and subjected to qRT-PCR using primers shown in S2 Table to detect the association between PYCARD-AS1 and PYCARD transcripts.
Polysome analysis was performed as described [48]. A total of 5 × 106 cells were preincubated with 100 mg/ml cycloheximide (Sigma-Aldrich) for 15 min. Cytoplasmic lysates were prepared and then fractionated by ultracentrifugation through 15%–50% linear sucrose gradients. Twelve fractions were collected, and RNA extracted from each fraction was subjected to qRT-PCR detection for specific transcript.
Student′s t-test was performed to compare the differences between experimental groups relative to their paired controls. The data were presented as the mean ± SD and p-values of < 0.05 or below were considered statistically significant.
|
10.1371/journal.ppat.1003031 | Semaphorin-7A Is an Erythrocyte Receptor for P. falciparum Merozoite-Specific TRAP Homolog, MTRAP | The motility and invasion of Plasmodium parasites is believed to require a cytoplasmic actin-myosin motor associated with a cell surface ligand belonging to the TRAP (thrombospondin-related anonymous protein) family. Current models of invasion usually invoke the existence of specific receptors for the TRAP-family ligands on the surface of the host cell; however, the identities of these receptors remain largely unknown. Here, we identify the GPI-linked protein Semaphorin-7A (CD108) as an erythrocyte receptor for the P. falciparum merozoite-specific TRAP homolog (MTRAP) by using a systematic screening approach designed to detect extracellular protein interactions. The specificity of the interaction was demonstrated by showing that binding was saturable and by quantifying the equilibrium and kinetic biophysical binding parameters using surface plasmon resonance. We found that two MTRAP monomers interact via their tandem TSR domains with the Sema domains of a Semaphorin-7A homodimer. Known naturally-occurring polymorphisms in Semaphorin-7A did not quantitatively affect MTRAP binding nor did the presence of glycans on the receptor. Attempts to block the interaction during in vitro erythrocyte invasion assays using recombinant proteins and antibodies showed no significant inhibitory effect, suggesting the inaccessibility of the complex to proteinaceous blocking agents. These findings now provide important experimental evidence to support the model that parasite TRAP-family ligands interact with specific host receptors during cellular invasion.
| Apicomplexan parasites are one of the most significant groups of pathogens infecting humans and include Plasmodium falciparum, the parasite responsible for malaria. These parasites critically depend on their human host and must invade our cells to multiply; therefore, understanding this invasion process - with the eventual aim of therapeutically preventing it - has been a focus for scientific investigation. A key component of the invasion machinery is a family of proteins (the “TRAP” family) which traverse the membrane surrounding the parasite: the part remaining within the parasite connects to a molecular motor that powers invasion, whilst the surface-exposed region is thought to interact with proteins on the surface of the target host cell. One major question that remains unanswered is the identity of the host receptors for the TRAPs. In our paper, we use a method specifically designed to detect interactions that occur in the extracellular space between host and pathogen proteins to reveal a host receptor called Semaphorin-7A for the TRAP-family member used by the blood stage of the malarial parasite – a protein called MTRAP. The characterization of this host-parasite interaction may therefore lead to novel therapies based upon preventing parasite invasion.
| Plasmodium falciparum is the etiological agent of the most severe form of malaria causing over one million deaths annually, primarily in African children [1]. The parasite lifecycle is complex and involves distinct stages that can recognise and invade differentiated cell types of both the human host and the mosquito vector. These stages are characterised by different invasive properties: ookinetes must cross the epithelial cells of the mosquito gut; sporozoites target both the secretory cells of the mosquito salivary glands and the hepatocytes of the human host, which they can either traverse or invade; and merozoites invade human erythrocytes. The ability of each stage to invade their target cells is an obligatory step in the lifecycle of the parasites and therefore these events have been considered attractive points for therapeutic intervention.
Invasion and motility requires a single-headed class XIV myosin anchored to the inner membrane complex that unidirectionally propels short actin filaments to impart motive force [2]. The actin filaments are coupled via the glycolytic enzyme aldolase [3], [4] to parasite cell surface proteins or “invasins” belonging to the TRAP (thrombospondin-related anonymous protein) family, which in turn are thought to bind via their extracellular region to host cell surface receptors thereby coupling the actin-myosin power-stroke into forwards movement of the parasite (Figure 1 A).
Each different motile form of the parasite is distinguished by its own stage-specific cell surface TRAP-family member [5]. In Plasmodium species, the TRAP-family proteins include TRAP, S6 (also known as TREP), CTRP, MTRAP and TLP. TRAP and S6 are expressed on sporozoites [6], [7], [8], CTRP on ookinetes [9], MTRAP on merozoites [10] and TLP on both sporozoites and merozoites [11], [12]. Attempts to target the genes encoding these proteins have shown that most of them are essential for motility and invasion. TRAP is critical for sporozoite invasion of the salivary glands and for infection of mammalian liver as well as sporozoite gliding motility [13]; CTRP is essential for invasion of the mosquito midgut [9]; and S6 is important for sporozoite gliding motility and invasion of mosquito salivary glands [6], [8]. TLP deletion initially showed no effect indicating a redundant role for this protein [11]; however, recent studies indicate a role in sporozoite cell traversal [12], [14]. The TRAP-family can be extended to include other cell surface and secreted proteins that contain similar domains and include CSP, SPATR, TRSP, WARP and PTRAMP [5]; PTRAMP, like MTRAP, is expressed in merozoites [15]. To date, it has not been possible to genetically delete MTRAP, indicating it may be essential for parasite growth in blood stage culture [10].
Structurally, TRAP-family proteins are predicted type I cell surface proteins characterised by having one or more extracellular thrombospondin type-I repeats (TSR) domains, and/or von Willebrand factor (vWF)-like A-domain(s) and an acidic cytoplasmic tail with a sub-terminal tryptophan residue [5]. Studies of the individual domains have implicated distinct functions in motility and invasion. The cytoplasmic tail of TRAP, CTRP, TLP and MTRAP have all been shown to interact with aldolase [4], [10], [11], and the cytoplasmic tail of TRAP was shown to be essential for gliding motility and invasion of both salivary glands and hepatocytes [16]. The extracellular TSR and vWF A-domains have been implicated in host cell interaction and invasion; indeed, both the TSR and A-domain of TRAP have been shown to be essential for invasion into both mosquito salivary glands and mammalian hepatocytes [17], whereas for CTRP, only the A-domains are essential for infectivity [18]. In contrast to the cytoplasmic regions of these proteins, much less is known about the host binding partners for the extracellular regions of TRAP-like proteins and how they play a role in motility, invasion and host cell tropism. The human host extracellular molecules that have so far been identified as binding TRAP-family proteins are not restricted to particular cell types: TRAP ectodomains are known to bind sulphated glycoconjugates [7], [19] and heparin [20], [21], [22] which are both widely distributed molecules. Indeed, both biochemical and functional studies have suggested the presence of additional TRAP receptors on hepatocytes [17], [20], [23] but their identities remain unknown; very likely, this is due to the technical challenges of biochemically manipulating membrane proteins and the fact that their extracellular interactions are typified by highly transient interaction strengths [24]. The identification of the host cell surface receptor proteins for TRAP-like parasite ligands remains an important unanswered question towards a better understanding of their role in host cell recognition and invasion (Figure 1 A).
Here we report how we have used an assay that is specifically designed to circumvent the technical difficulties in identifying low affinity extracellular interactions called AVEXIS (for AVidity-based EXtracellular Interaction Screen) [25], [26] to identify Semaphorin-7A as an erythrocyte receptor for the P. falciparum merozoite-specific TRAP homolog protein, MTRAP. We report the biochemical characterisation of the interaction and examine its role in erythrocyte invasion.
To identify an erythrocyte receptor for P. falciparum MTRAP, we expressed the entire predicted extracellular region as a secreted recombinant protein in human embryonic kidney (HEK)293E cells. Given the known difficulties in expressing functional Plasmodium proteins [27], we codon-optimised the MTRAP gene for expression in mammalian cells, replaced the signal peptide with a high-scoring exogenous sequence from a mouse antibody, and mutated the predicted N-linked glycosylation sequons to prevent inappropriate glycan addition that might mask potential protein interaction interfaces. The ectodomain was expressed as both a monomeric and a pentameric his-tagged protein. Pentamerisation was achieved by using a peptide sequence derived from the cartilage oligomeric matrix protein (COMP) and was used to increase binding avidity so as to increase the likelihood of detecting transient binding events that are a common feature of extracellular receptor interactions. Both the monomeric and pentameric forms of MTRAP bound human erythrocytes (Figure 1 B) relative to controls, which confirmed that the proteins were biochemically active and that MTRAP binds an erythrocyte cell surface receptor. As expected, the binding of the more avid pentameric protein was more resistant to washing steps (Figure 1 B).
To determine the molecular identity of the human erythrocyte MTRAP receptor, we took a systematic approach by using the AVEXIS assay and a protein library that represents the cell surface receptor repertoire of the human erythrocyte. This approach has been successfully used to identify basigin as the erythrocyte receptor for P. falciparum RH5 [26]. The pentameric β-lactamase-tagged MTRAP ectodomain was screened against the library of 40 erythrocyte receptor baits used previously. A single interaction was observed (Figure 1 C, upper panel) corresponding to Semaphorin-7A (also known as CD108). The same single interaction was identified in the reciprocal bait-prey orientation (Figure 1 C, lower panel). Semaphorin-7A is a GPI-linked surface protein that is broadly expressed in several tissues [28], [29], and particularly on activated lymphocytes where it has been shown to be involved in regulating immune responses [30], [31] and neurons of both the central and peripheral nervous systems where it has documented roles in axon guidance [32]. Semaphorin-7A is the antigen for the John-Milton-Hagen blood group, although its function on erythrocytes isn't known.
To show that MTRAP and Semaphorin-7A interact directly and to quantify the biophysical parameters of the interaction, we used surface plasmon resonance (SPR). The entire ectodomain of Semaphorin-7A was expressed as a soluble recombinant protein and purified before serial dilutions were injected over MTRAP immobilised on a sensor chip. Clear saturable binding was observed (Figure 1 D) from which an equilibrium binding constant (KD) of 1.18±0.40 µM was derived. Independent kinetic parameters were in agreement with the equilibrium data (Table S1) and were within the expected range for a typical membrane-tethered receptor-ligand pair that have been shown to have physiological relevance [24], [26], [33]. Taken together, these data show that Semaphorin-7A is an erythrocyte receptor for the P. falciparum merozoite TRAP-family ligand, MTRAP.
The interaction interface on erythrocyte receptors bound by merozoite surface ligands have been shown, in several cases, to be dependent on the glycosylation state of their erythrocyte receptors [34], [35], [36]. Human Semaphorin-7A contains five predicted N-linked glycosylation sites and so to determine whether MTRAP binding was influenced by glycans, we treated recombinant Semaphorin-7A with PNGase F to remove N-linked glycans (Figure 2 A). PNGase F-treated Semaphorin-7A was indistinguishable from untreated Semaphorin-7A in its ability to bind MTRAP using either the AVEXIS assay (Figure 2 B) or more quantitative SPR (Figure 2 C). These data suggest that the interaction of MTRAP with Semaphorin-7A is not influenced by the presence of glycans on the receptor.
It is known that TRAP can bind sulphated glycoconjugates on hepatocytes [19]. To investigate whether MTRAP was able to bind sulphated glyconjugates, we tested a panel of natural and synthetic glycoconjugates and determined whether they could bind MTRAP using SPR. Chondroitin sulphate A, chondroitin sulphate C, dextran sulphate, heparin and heparan sulphate were each injected at high concentrations over MTRAP immobilised on a sensor chip. No detectable binding for any of the glycoconjugates was observed relative to the Semaphorin-7A positive control (Figure 2 D). We estimate that interactions as weak as 100 µM would have been detected using this approach and conclude that glycoconjugates are unlikely to be major MTRAP ligands.
Structural and biochemical studies have shown that semaphorins exist as homodimers [37], [38], [39]. Size exclusion chromatography (SEC) confirmed that our recombinant soluble monomeric Semaphorin-7A ectodomain eluted in a fraction consistent with it forming a homodimer (Figure 3 A; top panel), as has been shown before [39]. Surprisingly, the ectodomain of MTRAP also eluted at an increased size, perhaps suggesting it also forms a homodimer in solution (Figure 3 A; middle panel). To further investigate these findings, both proteins were subjected to SEC immediately followed by multiangle light scattering (MALS). This analysis demonstrated that while soluble Semaphorin-7A formed quasi-stable homodimers, the soluble MTRAP ectodomain was monomeric (Figure 3 B). The early elution behavior of MTRAP in SEC may therefore be caused by a large hydrodynamic shape possibly due to the protein being highly flexible or adopting a long rod-like shape, as has recently been suggested from atomic force microscopy studies [40]. To determine the stoichiometry of the Semaphorin-7A:MTRAP complex, both purified proteins were mixed at equimolar concentrations and allowed to form a complex prior to separation by SEC. As expected, the complex eluted at a higher mass than each protein individually (Figure 3 A; bottom panel) and both proteins were present in these fractions (Figure 3 C). The unusual behaviour of these proteins by SEC made determining the stoichiometry of the complex by this method difficult and so the fraction corresponding to the peak was subjected to amino acid analysis (Figure 3 D). The amino acid compositions determined experimentally were compared to expected theoretical stoichiometries of 1∶1, 2∶1 and 1∶2 (Semaphorin-7A∶MTRAP). The amino acids that were most characteristic of either Semaphorin-7A or MTRAP indicated that a 1∶1 ratio best fitted the data (Figure 3 D). Calculating the sum of the squared residuals for all amino acids gave values of 1×10−4, 14×10−4 and 17×10−4 for the 1∶1, 2∶1 and 1∶2 models, respectively; again, indicating that the 1∶1 ratio best fitted the data. These results therefore suggest that two MTRAP monomers bind one Semaphorin-7A homodimer.
The paired TSR domains are the most conserved region of the MTRAP ectodomain across different Plasmodium species, and the TSR domain of TRAP has previously been shown to contribute to receptor binding [19]. To investigate whether the TSR domains of MTRAP contain the Semaphorin-7A binding site, a 74 amino acid fragment that contained both predicted TSR domains (TSR1+2), and two additional fragments containing each TSR domain individually, were expressed as biotinylated bait proteins (Figure 4 A). The TSR1+2 MTRAP fragment bound to Semaphorin-7A indistinguishably from the entire MTRAP ectodomain using the AVEXIS assay demonstrating that the Semaphorin-7A binding site was localised to the two TSR domains (Figure 4 B). A quantitative analysis using SPR demonstrated a slightly weaker interaction affinity for TSR1+2 (KD = 1.96±0. 03 µM; Figure 4 C; Table S1) compared to the entire MTRAP ectodomain, suggesting that residues outside of the TSR domains make only minor contributions to the binding affinity. Supporting this, purified pentameric TSR1+2 was able to bind erythrocytes (Figure S1), and it has recently been shown that recombinant MTRAP lacking the TSR domains could not [40]. Neither of the two individual TSR domains bound Semaphorin-7A by AVEXIS (Figure 4 B) or SPR (Figure 4 C lower inset) demonstrating that the Semaphorin-7A binding site requires both TSR domains.
Similarly, to localise the MTRAP binding site on Semaphorin-7A, we expressed constructs containing each of the three recognisable domains: Sema, PSI and Ig-like (Figure 4 A). Only the entire ectodomain of Semaphorin-7A bound MTRAP using the AVEXIS assay, irrespective of the bait-prey orientation (Figure 4 B, Figure S2). By SPR, however, detectable binding to MTRAP was observed using the Sema domain alone (Figure 4 D; top graph) with similar binding parameters to the full-length ectodomain (KD = 0.83±0.43 µM). No binding was observed with the individual PSI or Ig-like domains (Figure 4 D; bottom graph). Taken together, these data demonstrate Semaphorin-7A and MTRAP directly interact via their Sema and tandem TSR domains respectively.
Malaria is thought to have been a powerful selective force in human evolutionary history and given the essential role of MTRAP in parasite blood stage culture we asked whether any naturally-occurring polymorphisms in human Semaphorin-7A would influence the binding of MTRAP. Eight variants in the extracellular region of human Semaphorin-7A are known (seven within the Sema domain and one within the PSI domain) and all were individually introduced by site-directed mutagenesis (Figure 5 A; Table S2). All variants were expressed (Figure 5 B) and the dissociation rate constants (kd) for MTRAP binding were determined using SPR (Figure 5 C, Table S3). No significant differences were observed in the interaction strengths for any of the variants suggesting that at least the known common variants in Semaphorin-7A have not been selected due to differences in the ability to bind P. falciparum MTRAP.
To examine the role of the MTRAP-Semaphorin-7A interaction in erythrocyte invasion, we attempted to block invasion in vitro using purified recombinant proteins and antibodies raised against either the parasite ligand or erythrocyte receptor. Addition of purified highly avid pentamerised Semaphorin-7A or MTRAP in increasing concentrations had no inhibitory effect on erythrocyte invasion (Figure 6 A), even at concentrations 10-fold higher than the measured interaction strength between monomeric proteins (Figure 1 D). Previous studies of the PfRH5-basigin interaction suggest that antibodies more potently block receptor-ligand interactions during erythrocyte invasion, presumably because their interaction affinities are much higher. We therefore tested an anti-Semaphorin-7a monoclonal antibody in P. falciparum invasion assays at increasing concentrations. No inhibitory activity was seen even at the highest concentrations, unlike a monoclonal directed against the PfRH5 receptor, basigin, which has a >80% invasion inhibitory effect at 10 µg/ml (Figure 6 B). We also raised rabbit polyclonal antibodies against purified, recombinant, monomeric MTRAP and Semaphorin-7A. Both antibodies were able to detect proteins of the expected size in parasite supernatants (MTRAP) and erythrocyte ghost preparations (Semaphorin-7A) by Western blot (Figure 6 C); we also showed that the anti-MTRAP antibodies were able to block binding of MTRAP to Semaphorin-7A by AVEXIS (Figure S3). When added to invasion assays, however, neither had an inhibitory effect on P. falciparum erythrocyte invasion, even at the highest concentration, in contrast to antibodies against AMA-1 (Figure 6 D). Other attempts to block invasion through antibodies targeting MTRAP have yielded similar results [10], [40], suggesting that the MTRAP-Semaphorin-7A interaction is either not accessible to blocking agents in in vitro assays, perhaps because it takes place at a late time point during the invasion process, or it is not essential for erythrocyte invasion.
In this study, we have successfully expressed a functional recombinant P. falciparum MTRAP protein and shown that it binds erythrocytes. This protein and a library of human erythrocyte receptor ectodomains were used to identify Semaphorin-7A as its erythrocyte receptor using a systematic screening assay (AVEXIS) that is specifically designed to detect low affinity extracellular protein interactions. Importantly, this represents the first example of a host cell surface receptor protein for the TRAP-like family of parasite ligands that provide the crucial link between the target host cell and the parasite's cytoplasmic actin-myosin motor that powers the invasion process in any Plasmodium species. Saturation binding behaviours showed that the MTRAP-Semaphorin-7A interaction was specific and, as expected, was of moderately low affinity as is typical of other measured extracellular receptor-ligand interactions [24] and is consistent with low recovery of bound recombinant MTRAP to erythrocytes performed by others [40].
The biochemical characterisation of the interaction suggests that two MTRAP monomers interact via their tandem TSR domains with the Sema domains of a Semaphorin-7A homodimer. This result is supported by the recent finding that a recombinant MTRAP protein lacking the TSR domains was unable to bind erythrocytes [40], whereas a protein containing just the two TSR domains could (Figure S1). This was not unexpected as the TSR domains contribute to binding of other TRAP family proteins to their host cells [5], [19] and this region is conserved across MTRAP orthologues in other Plasmodium species [10]. In contrast to the sporozoite TRAP protein, we could find no evidence that MTRAP bound sulphated glycoconjugates despite using a highly sensitive assay. This might be explained by the presence of the sequence “WSPCSVTC” in the TSR domains of TRAP, TLP and the related protein CSP which is believed to be a sulfatide binding motif (Muller et al., 1993) and is absent from MTRAP. Recent work by Uchime et al. confirmed that the TSR domains of MTRAP structurally differ from previously studied TSR domains based on its disulphide bonds suggesting a more compact structure, perhaps indicating that both TSR domains function together [40] and explaining the requirement for tandem TSR domains in Semaphorin-7A binding.
Our experiments to define the MTRAP binding site on Semaphorin-7A were complicated by the fact that Semaphorin-7A, like other semaphorins, is known to form a dimer with a large (2860 Å2) and primarily hydrophobic contact interface that involves the whole molecule, including the Ig domain [39]. Individually expressing each of the constituent domains to map the MTRAP binding site therefore disrupted this homodimeric structure. MTRAP, however, did interact with a construct containing the Sema domain alone immobilised at sufficient density on a Biacore chip (Figure 4 D) suggesting that the Sema domain contained the MTRAP interaction interface. We also established that the stoichiometry of binding is likely to be two MTRAP monomers binding to a single Semaphorin-7A homodimer. This binding model is also used by the endogenous semaphorin ligands, the plexins, as shown by crystallisation of the complex [39], [41]. It is possible, as for the plexins, that binding of MTRAP to a dimeric receptor triggers local MTRAP clustering which is then necessary for function by bringing the cytoplasmic regions into close proximity.
Attempts to genetically disrupt MTRAP in multiple P. falciparum strains were unsuccessful suggesting that it is essential for blood stage growth [10]. We therefore attempted to disrupt the MTRAP-Semaphorin-7A interaction; however, neither purified highly-avid pentameric proteins of both MTRAP or Semaphorin-7A nor polyclonal antibodies raised against either MTRAP or Semaphorin-7A showed any discernible effect on erythrocyte invasion in vitro. The inability of polyclonal antibodies raised against MTRAP to affect invasion is consistent with findings from other groups and suggests that MTRAP is unlikely to be a component of an effective subunit blood-stage vaccine [10], [40]. MTRAP may therefore have an important receptor-independent role similar to TRAP which is required not only for cellular invasion but also gliding motility [13]. One other possible explanation is that the interaction may occur in close physical proximity to the moving junction - an electron dense thickening formed at the nexus of the erythrocyte and merozoite plasma membranes, which would almost certainly be inaccessible to large soluble proteinaceous blocking reagents. This hypothesis can be supported by the lack of non-synonymous polymorphisms found in the MTRAP ectodomain, suggesting that unlike other merozoite ligands it is not exposed to strong selection pressure by the immune system [10]. In agreement with this, MTRAP does not appear to be a primary target of the adaptive immune system, with low anti-MTRAP reactivity in human sera from malaria endemic regions [40]. This is in clear contrast to the highly polymorphic sporozoite TRAP protein [19]. Similarly, Semaphorin-7A had very few known polymorphisms, all of which did not quantitatively affect its interaction with MTRAP, and to our knowledge, there is no evidence of selective pressure on this receptor in malaria endemic regions. Interestingly, loss of Semaphorin-7A expression on erythrocytes can be acquired, typically with increasing age, and levels of erythrocyte Semaphorin-7A expression have been observed to fluctuate significantly during pregnancy [42]. Our finding that Semaphorin-7A is a receptor for P. falciparum MTRAP makes correlating the levels of cell surface Semaphorin-7A with clinical parameters of P. falciparum infection an important area for future study.
Our finding that Semaphorin-7A is a receptor for MTRAP provides the first example of a host receptor protein for a member of the TRAP-like family. Semaphorin-7A, similar to MTRAP, is a member of a larger family of cell surface proteins, the semaphorins, that can be subdivided into eight different structural classes [43]. Whether the other members of the TRAP-like family will have identifiable receptors within the broader semaphorin family is a key area for future research.
Whether the essentiality of MTRAP simply lies in its function as a membrane-tethered link to the parasite cytoplasmic myosin-based motor or has additional roles in determining the cellular tropism of invasion is still not clear. However, the identification of an erythrocyte receptor for the extracellular region of MTRAP supports a mechanism whereby TRAP-family ligands directly interact with a protein displayed on the surface of the target cell. This interaction may therefore provide the traction required to couple the activity of the parasite myosin-based motor into a relative cellular movement that is necessary for invasion. We believe that this finding together with the successful demonstration of an experimental approach to identify host receptors for parasite TRAP-like ligands will stimulate further research into the challenging task of identifying receptors for this important class of parasite ligands.
Use of erythrocytes and serum from human donors for P. falciparum culture was approved by the NHS Cambridgeshire 4 Research Ethics Committee. All subjects provided written informed consent. The use of animals to raise antisera was performed according to UK Home Office governmental regulations and approved by the local Sanger Institute ethical review board.
A list of the erythrocyte receptor proteins used in this study and the numbering used in Figure 1 C can be found in Supplementary Table 1 in Crosnier et al., 2011. Proteins within the human erythrocyte protein library were produced as bait and prey constructs as previously described [26]. Briefly, for the proteins containing a signal peptide, each expression construct contained the entire extracellular region (including the native signal peptide) flanked by unique NotI and AscI sites to facilitate cloning into a vector containing a C-terminal rat Cd4d3+4-tag and either an enzymatically biotinylatable peptide (baits) or a peptide from the rat cartilage oligomeric matrix protein (COMP) which spontaneously forms pentamers followed by the enzyme beta-lactamase (preys). The ectodomain fragments of the four type II proteins (which lack a signal peptide) were expressed only as monomeric baits and not preys. Baits for the type II proteins were produced by flanking the predicted extracellular regions with NotI and AscI restriction enzymes and cloning them into a vector containing a mouse immunoglobulin kappa light chain signal peptide followed by the biotinylatable tag and Cd4d3+4 at the N-terminus of the insert. Bait proteins were enzymatically biotinylated during expression by cotransfection of a secreted form of the E.coli BirA protein biotin ligase [25]. The MTRAP ectodomain bait and prey constructs differed from the erythrocyte receptors in that the low-scoring endogenous signal peptide was replaced by a high-scoring signal peptide from a mouse immunoglobulin kappa light chain and the serines and threonines in the context of potential N-linked glycan sites were systematically mutated to alanine to prevent inappropriate glycosylation. All ectodomains were codon optimised for mammalian expression and chemically synthesized (Geneart AG, Regensburg, Germany). The constituent Sema, PSI and Ig-like domains of human Semaphorin-7A were produced by identifying the domain boundaries using the crystal structure of the Semaphorin-7A extracellular region as a guide [39]. The MTRAP TSR1+2, TSR 1 and TSR 2 domain boundaries were estimated based on the location of conserved cysteine residues identified by protein alignments of TSR repeats. The sequences corresponding to these domains were amplified using primers with flanking NotI and AscI cloning sites for cloning into the appropriate expression vectors. The PSI, Ig and all TSR domains were cloned into the same vectors as MTRAP to add an exogenous signal peptide required for protein secretion. Naturally-occurring variants of Semaphorin-7A were found in dbSNP (www.ncbi.nlm.nih.gov/projects/SNP/). Constructs containing these variants were produced by site directed mutagenesis (GeneArt AG). Variants were mapped onto the structure of Semaphorin-7A using PyMOL (www.pymol.org). Monomeric His-tagged proteins were prepared by subcloning the NotI/AscI flanked extracellular regions into a vector containing a C-terminal Cd4d3+4 tag followed by a hexa-His tag [25]. An additional monomeric His-tagged MTRAP lacking the Cd4d3+4-tag was produced by amplifying the MTRAP coding region with primers containing NotI and EcoRI sites and inserting into a NotI/EcoRI-digested His-vector using standard cloning procedures. Pentameric His-tagged proteins were similarly made by cloning the inserts into a NotI/EcoRI-digested prey vector where the COMP-beta-lactamase region had been replaced by a COMP-hexa-His tag. All proteins were expressed as secreted proteins by transient transfection of the human HEK293E cell line grown in suspension as described [25], [44].
His-tagged proteins were purified from supernatants from transient transfections on HisTrap HP columns (GE Healthcare) using an ÄKTAxpress (GE Healthcare) according to manufacturer's instructions. Size exclusion chromatography of nickel purified samples was carried out on a Superdex 200 Tricorn 10/600 column (GE Healthcare) in HBS-EP (GE Healthcare).
Amino acid analysis was performed by the PNAC Facility, University of Cambridge, Cambridge, UK.
For PNGase F treatment, biotinylated Semaphorin-7A was incubated with 50 U/µl of PNGase F (NEB) for 10, 30 and 60 min at 37°C for Western blot analysis, and 60 min at 37°C for AVEXIS and Biacore analysis.
The P. falciparum AMA-1 ectodomain was produced in a similar way to MTRAP, cloned into the vector containing a C-terminal Cd4d3+4 tag followed by a hexa-His tag, then expressed and purified as described above.
Interaction screening was carried out as previously described [25], [26]. Briefly, both bait and prey protein preparations were normalised to activities that have been previously shown to detect transient interactions (monomeric half-lives less than 0.1 second) with a low false positive rate [25]. Biotinylated baits that had been dialysed against HBS were immobilised in the wells of a streptavidin-coated 96-well microtitre plate (NUNC). Normalised preys were added, incubated for 1 hour at room temperature, washed three times in HBS plus 0.1% Tween- 20, and once in HBS, after which 125 µg/ml of nitrocefin was added and absorbance values measured at 485 nm on a Pherastar plus (BMG laboratories). For the screen, a positive control interaction using rat Cd200 as a bait and rat Cd200R as a prey, and a negative control interaction using rat Cd4d3+4 as a bait and rat Cd200R as a prey, was used (+ and − in Figure 1 C). Where AVEXIS was used for interaction site mapping and PNGase F experiments (Figure 2 B and 4 B), the Cd4d3+4 tag alone was used as a negative control bait, and a biotinylated anti-Cd4 antibody as positive control to capture the Cd4d3+4-tagged prey.
Erythrocyte binding assays were carried out as described previously [45] but with slight modifications. Briefly, 60 µg of purified proteins were mixed with 50 µl of packed fresh erythrocytes for 2 hours at 4°C. The erythrocytes were separated from supernatant by spinning through 400 µl of ice cold dibutyl phthalate (Sigma) at 12000 g for 30 seconds, after which the erythrocyte pellet was washed in ice cold PBS. Proteins bound to the erythrocytes were eluted by incubating with 20 µl of 1.5 M NaCl at room temperature for 45 minutes, and collected after 30 seconds of 12,000 g centrifugation. The unbound, wash and eluted material were analysed by Western blotting as described below.
Erythrocytes were pelleted then washed twice in 5 volumes of ice cold 2 mM HEPES, 154 mM NaCl, pH 7.1 and centrifuged for 15 mins at 500 g at 4°C. The pellet was transferred into 15 volumes of ice cold 10 mM Tris-HCl 1 mM EDTA pH 7.1 and left on ice for 30 mins. After centrifugation at 20,000 g for 15 mins at 4°C, the supernatant was discarded and the pellet gently resuspended whilst leaving behind the denser dark pellet of unlysed cells. The pellet was centrifuged at 20,000 g for 15 mins at 4°C then washed in 2 mM HEPES, 154 mM NaCl, pH 7.1 four times. The washed ghosts were resuspended in 10 mM Tris-HCl pH 7.1 and centrifuged at 20 000 g for 15 mins at 4°C, after which, the pellet was resuspended in 1 volume of 10 mM Tris-HCl pH 7.1, then stored at −20°C.
To make culture supernatants, synchronised schizonts were purified by centrifugation onto an 80% Percoll cushion, collected at the cushion interface, placed back into in vitro culture at 2.5×107 parasites/ml in the absence of additional erythrocytes and allowed to rupture overnight. Cells were removed by centrifugation and supernatants stored at −80°C.
To raise polyclonal antisera against Semaphorin-7A, MTRAP, and AMA-1, purified proteins were injected into rabbits (Cambridge Research Biochemicals, Billingham, UK). The sera were purified on Hi-Trap Protein G HP columns (GE Healthcare), and the mouse anti-human Semaphorin-7A, MEM-150 monoclonal antibody [46] was purified from mouse ascites on a HiTrap IgM Purification HP column (GE Healthcare), using an ÄKTA Xpress (GE Healthcare) according to the manufacturer's instructions.
Proteins were resolved by SDS-PAGE using NuPAGE 4–12% Bis Tris precast gels (Invitrogen). Where reducing conditions were required NuPAGE reducing agent and anti-oxidant (Invitrogen) were added to the sample and the running buffer, respectively. Proteins were blotted onto PVDF membranes (Amersham) and blocked in 2% BSA. Membranes were incubated with either peroxidase-conjugated streptavidin (Jackson Immuno Research) or anti-C-term His-HRP antibody (Invitrogen) as appropriate, and proteins detected using SuperSignal West Pico Chemiluminescent substrate (Thermo Scientific). When using rabbit-anti-Semaphorin-7A or anti-MTRAP, an anti-rabbit-IgG-HRP (Invitrogen) secondary antibody was used.
Surface plasmon resonance studies were performed using a Biacore T100 instrument (GE Healthcare). Biotinylated bait proteins were captured on a streptavidin-coated sensor chip (GE Healthcare). Approximately 150 RU of the negative control bait (biotinylated rat Cd4d3+4) were immobilised in the flow cell used as a reference and approximate molar equivalents of the query protein immobilized in other flow cells. Purified analyte proteins were separated by size exclusion chromatography on a Superdex 200 Tricorn 10/600 column (GE Healthcare) in HBS-EP (GE Healthcare) just prior to use in SPR experiments to remove any protein aggregates that might influence kinetic measurements. Increasing concentrations of purified proteins were injected at 100 µl/min to determine kinetic parameters, or at 20 µl/min for equilibrium measurements. The surface was regenerated with a pulse of 2 M NaCl at the end of each cycle. Duplicate injections of the same concentration in each experiment were super imposable demonstrating no loss of activity after regenerating the surface. Both kinetic and equilibrium binding data were analysed in the manufacturer's Biacore T100 evaluation software version 1.1.1 (GE Healthcare). Equilibrium binding measurements were taken once equilibrium had been reached using reference-subtracted sensorgrams. Both the kinetic and equilibrium binding were replicated using independent protein preparations of both ligand and analyte proteins. All experiments were performed at 37°C in HBS-EP.
Sulfated-glycoconjugates were obtained from Sigma and resuspended in 1× HBS-EP (Biacore, GE Healthcare) and used for surface plasmon resonance studies at 1 mg/ml.
Size exclusion chromatography was performed on a Superdex200 10/30 column (GE Healthcare) equilibrated in 50 mM Tris.HCl, pH 7.5, 150 mM NaCl at 0.4 ml/min. The column was followed in-line by a Dawn Heleos-II light scattering detector (Wyatt Technologies) and an Optilab-Rex refractive index monitor (Wyatt Technologies). Molecular mass calculations were performed using ASTRA 5.3.4.14 (Wyatt Technologies) assuming a dn/dc value of 0.186 ml/g.
The 3D7 P. falciparum parasite strain was cultured in human O+ erythrocytes at 5% hematocrit in complete medium (RPMI-1640 containing 10% human serum), under an atmosphere of 1% O2, 3% CO2, and 96% N2.
Invasion assays were carried out in round-bottom 96-well plates, with a culture volume of 100 µL per well at a hematocrit of 2%. Parasites were synchronized at early stages with 5% (w/v) D-sorbitol (Sigma), trophozoite stage parasites were mixed with the specified protein blocking reagent, and then incubated in the plates for 24 hours at 37°C inside a static incubator culture chamber (VWR), gassed with 1% O2, 3% CO2, and 96% N2. At the end of the incubation period, erythrocytes were harvested and parasitized erythrocytes were stained with 2 µM Hoechst 33342 (Invitrogen), as described previously [47]. Purified MEM-150, rabbit polyclonal antibodies and pentamerised MTRAP and Semaphorin-7A ectodomains were dialysed into RPMI (GIBCO) prior to use. A monoclonal antibody targeting basigin (MEM-M6/6, Abcam, Cambridge, UK) was purchased and dialysed into RPMI before addition into invasion assays.
Hoechst 33342 (Invitrogen) stained samples were excited with a 355 nm UV laser (20 mW) on a BD LSRII flow cytometer (BD Biosciences) and detected with a 450/50 filter. BD FACS Diva (BD Biosciences) was used to collect 100,000 events for each sample. FSC and SSC voltages of 423 and 198, respectively, and a threshold of 2,000 on FSC were applied to gate the erythrocyte population. The data collected were further analyzed with FlowJo (Tree Star). All experiments were carried out in triplicate. GraphPad Prism (GraphPad Software) was used to plot the generated parasitemia data.
|
10.1371/journal.pcbi.1007037 | Detecting interaction networks in the human microbiome with conditional Granger causality | Human microbiome research is rife with studies attempting to deduce microbial correlation networks from sequencing data. Standard correlation and/or network analyses may be misleading when taken as an indication of taxon interactions because “correlation is neither necessary nor sufficient to establish causation”; environmental filtering can lead to correlation between non-interacting taxa. Unfortunately, microbial ecologists have generally used correlation as a proxy for causality although there is a general consensus about what constitutes a causal relationship: causes both precede and predict effects. We apply one of the first causal models for detecting interactions in human microbiome samples. Specifically, we analyze a long duration, high resolution time series of the human microbiome to decipher the networks of correlation and causation of human-associated microbial genera. We show that correlation is not a good proxy for biological interaction; we observed a weak negative relationship between correlation and causality. Strong interspecific interactions are disproportionately positive, whereas almost all strong intraspecific interactions are negative. Interestingly, intraspecific interactions also appear to act at a short timescale causing vast majority of the effects within 1–3 days. We report how different taxa are involved in causal relationships with others, and show that strong interspecific interactions are rarely conserved across two body sites whereas strong intraspecific interactions are much more conserved, ranging from 33% between the gut and right-hand to 70% between the two hands. Therefore, in the absence of guiding assumptions about ecological interactions, Granger causality and related techniques may be particularly helpful for understanding the driving factors governing microbiome composition and structure.
| The human microbiome comprises thousands of microbial taxa, each of them potentially having interactions with many other taxa. Elucidating these interactions can be helpful for human health and beyond. The most conclusive approach for understanding microbial interactions would be myriads of controlled experiments in laboratory settings, each designed to test how taxa interact with one another. As this is a formidable challenge, an alternative strategy for constructing such an interaction network is a statistical framework employing formal models. Using conditional Granger causality, we demonstrate that correlation measures, which are very popular statistics in microbiome studies, are unreliable as proxies of microbial interactions. Overall, we found that strong interactions within a genus tended to be negative (e.g., competition), and they tended to occur on very short timescales (1–2 days). Many of such interactions were common across body sites. By contrast, strong interactions between genera tended to be positive (e.g., mutualism/facilitation) and were more evenly distributed over a range of timescales, up to the 20-day window that we considered. Very few of these interactions were conserved across body sites. We conclude that models of causality can be particularly useful in elucidating microbial interaction networks when laboratory investigations of interactions are impractical.
| Human microbiome research is rife with studies attempting to deduce microbial correlation networks from sequencing data [1–5]. In many cases, the goal is to identify candidate species interactions, often with an aim to explore implications for human health. Pairs of taxa that exhibit negative correlations, for example, could act as probiotics, particularly if negative correlations are with pathogens. Likewise, pairs of taxa that exhibit positive correlations could provide an understanding of microbial succession—processes that can impact the composition, and thus ‘ecosystem services’ provided by the human microbiome [6]. Unfortunately, “correlation is neither necessary nor sufficient to establish causation” [7]. Thus, many of the microbial interactions identified from standard correlation and/or network analyses may be misleading, at least when taken as an indication of taxon interactions. A twin study of the human gut microbiome, for example, showed that most co-occurrence patterns are driven by host genetics, rather than by microbial taxon-taxon interactions [8].
More generally, co-occurrence/correlation in cross-sectional data can emerge from two fundamental processes that are mutually non-exclusive: species interactions and habitat or environmental filtering. Among species interactions, competition can lead to mutual exclusion [9] or negative correlations, whereas mutualism, commensalism or parasitism can establish positive correlations. Environmental filtering [10,11], by contrast, describes a process where existence is only possible for species with suitable traits. In this case, a pair of microbial species may co-occur because they have similar nutritional requirements and/or environmental tolerances. Conversely, species may be mutually absent if they differ in environmental requirements or tolerances. Thus, both positive and negative taxon-taxon correlations can emerge from distinct underlying processes. This makes it impossible to tease apart the drivers of correlation.
Even when species do interact, correlation analyses can still be problematic. First, non-linear dynamics can weaken correlations among interacting species. Indeed, simulation shows that even in a deterministic and dynamically coupled two species system, zero correlation is possible [7]. When this occurs, correlation metrics erroneously imply lack of causation. Consequently, inferring causation from correlation is risky, particularly in biological systems where non-linear dynamics are ubiquitous [7]. Second, when a true species interaction occurs with a time lag (e.g., a metabolite secreted by a species takes some time to diffuse through the environment, get absorbed by the recipient, go through synthetic pathways and elicit an effect in the recipient), correlation will not capture the interaction because it depends on instantaneous covariance between two time series.
Because of the problems with a correlation network approach, at least for the purposes of inferring taxon interactions, correlation networks should only be used when plausible interaction mechanisms can be identified for strongly correlated taxon pairs [2], and even then with the realization that many interactions may be missed because of nonlinearity and/or time-lagged responses. Unfortunately, because of the diversity of the human microbiome, and the fact that many taxa are only recently identified and thus poorly studied, the map of ecological, or even metabolic interactions among taxa is highly incomplete [12,13]. Instead, microbial ecologists have generally used correlation as a proxy for causality. In a recent study, for example, a number of correlation metrics and linear models were applied to a human microbiome dataset to decipher co-occurrence and co-exclusion networks [2]. Although the study only used cross-sectional data of presence-absence states, the researchers drew inferences about competition and exclusion between microbial clades. Another study [14] concluded that the highly centralized correlation networks in diseased versus healthy human oral microbiomes make diseased microbiomes more prone to a community shift in response to environmental change. This would be the case if central nodes control the structure of community stability. To make such an argument, however, it is necessary to assume that observed correlations stem from taxon interactions. Many additional studies have similarly assessed ecological interactions using taxon-taxon correlation/co-occurrence data [5,15–20], while others still have computed correlation/co-occurrence metrics but left their implications open for interpretation [4,21–25].
The difficulties associated with co-occurrence studies extend beyond theoretical interpretation. Laboratory studies on a simple but highly controlled closed system of three interacting microbial species [26] indicate that there are only two ways to robustly detect species interactions: biological replicates of a system at a given time and observation of a system over time. Because biological replicates of entire microbiomes are far from possible, this leaves analysis of time-series data as the sole method for accurately identifying pairs of interacting taxa.
Despite the challenges associated with identifying causality in sequencing data, there is nonetheless a general consensus about what constitutes a causal relationship: causes both precede and predict effects [27]. In keeping with the conclusions from the Hekstra and Leibler study [26], this implies a time element that is missing from most correlation analyses. The statistical approach known as Granger causality [7,28], however, establishes causation by predicting the current state of a system using past states. Specifically, variable X is the “Granger cause” of variable Y if and only if X uniquely improves predictability of Y—i.e., if forecasting the future states of Y based on its own past states is improved when the past states of X are also included in the model [28]. Cross prediction of time-series (prediction improvement of Y by adding X in the predictor set that already included Y plus prediction improvement of X by adding Y in the predictor set that already included X) is a powerful and intuitive approach to establish causality. Furthermore, unlike correlation and co-occurrence analyses, which are non-directional, Granger causality is directional. This is important in understanding species interactions because correlation, even when it captures true ecological interactions between X and Y, cannot determine whether X impacts Y or vice-versa, or whether the impact goes in both directions. This deficiency of correlation studies is corrected by Granger causality, which allows researchers to tease apart the cause and effect in a species interaction. Though Granger causality analysis has been applied to a variety of datasets, including those showing beta oscillations in cortical networks [29], societal crises in response to climate change [30] and the relationship between daily Dow Jones stock returns and percentage changes in New York Stock Exchange trading volume [31], it is almost non-existent in the analysis of human microbiomes [32] (although some researchers have pointed to it as a potentially powerful approach in that context [33,34]). In the only study that we came across that applied the models of Granger causality to microbiome data, Gibbons et al [32] analyzed human gut microbiome (four time series, one of them being the data we analyzed) using three lags in the model, with the main goal being to compare different methods of analysis and to identify the drivers of temporal dynamics of microbial abundance. We analyzed high resolution time series of the human microbiome at four body sites [35] using 20 lags in the causality models, to decipher strong positive and negative interactions, to identify those interactions that are short-acting versus long-acting, and to determine if interaction are conserved across body sites. Hence, compared to Gibbons et al. [32], our study not only addresses a different scientific question, but also goes beyond one body site to four and beyond short-timescales to both short- and long-timescales. In this study, we specifically seek to: (1) evaluate the relationship between correlation and causation, (2) decipher the networks of strong positive and negative interactions in all taxa-pairs at different time scales, and (3) determine if causal networks generalize across body sites.
One of the challenges of using Granger causality to explore ecological data is the interpretation of results. In particular, the output from a Granger causality analysis is a series of coefficients, each representing a different timescale, with different taxon pairs having different numbers of significant coefficients. To deal with this issue, we take several approaches. First, we consider general metrics, incorporating all coefficients. Then, to account for the fact that timescales differing by only a few days may represent the same or similar processes, we group time-lags into general, qualitative timescales, considering trends for ‘short’ (1–5 days) and ‘long’ (15–20 days) interactions, independent of precise day values.
When all taxon-pairs with significant Pearson’s correlation and Granger causality across all body sites were considered, we observed a weak negative relationship between Granger’s causality and Pearson’s correlation (Fig 1). If correlation inferred causality, then we would expect a strong positive relationship between Pearson’s correlation and Granger causality. Such a weak negative relationship indicates that correlation is either a completely unreliable metric of causation or it actually suggests a taxon-interaction in the opposite direction as compared to that indicated by Granger causality.
We further examine causality-correlation relationship within long- and short-timescales of each body site. For this analysis, we only consider interspecific interactions. The only taxon pairs that are excluded are those that exhibit conflicting Granger causality signs of interaction within the short or long timescale. Tables in S1 Table through S8 Table show the numbers of interspecific taxon interactions that are positive, negative or insignificant for Granger causality and simultaneously positive, negative or insignificant for Pearson correlations for short and long timescale interactions. Applying chi-square tests of independence to these data shows that Pearson and Granger models are not independent for short timescales. In particular, having a negative Pearson coefficient makes a taxa-pair far less likely to have a negative Granger coefficient. This is consistent with the aggregated analysis of all body sites for the correlation between Pearson’s correlation and Granger causality in Fig 1. Interestingly, although this same trend appears marginally significant at long timescales in the gut, it does not apply to long timescale interactions at other body sites, where there appears no discernable relationship between Pearson and Granger causality results (See S1–S8 Tables).
When all the significant coefficients of Granger causality in each body site were examined collectively, vast majority of them are very small in magnitude (S1 Fig). Realizing that effect size can be important for ecological interpretation, we concentrate our results only to the top 5 percentile coefficients (“strong coefficients”, hereafter) for each of the positive and negative interactions in a body site. These strongest signals of causality were separately detected in each body site.
Fig 2 shows the number of strong coefficients for all pairs of taxa at each body site for interspecific (left column) and intraspecific (right column) interactions (for this analysis we take taxon A → taxon B and taxon B → taxon A as separate pairs). Of all the taxa in a body site, at least three-fourths had at least one intraspecific significant predictor (inset pie chart). Every taxon had at least one interspecific significant predictor (inset pie chart). When a taxon is predicted by another taxon, in vast majority of the cases, only one coefficient/lag (not necessarily the first lag) was significant for interspecific interactions whereas up to two coefficients made the vast majority of intraspecific interactions. In both cases, a few taxa-pairs are involved with up to four significant coefficients although the highest number observed was seven. The fact that intraspecific taxon pairs have more strongest coefficients than interspecific interactions suggests that a larger number of timescales are important for intraspecific relative to interspecific interactions.
Fig 3 shows the frequency with which strongest coefficients for different time-lags appear in the models for each taxon pair at each body site. Interspecific interactions appear disproportionately positive and relatively evenly distributed across time-lags (Fig 3, left column, see adjacent pie-charts). By contrast, almost all of intraspecific interactions are negative (Fig 3, right column), and the vast majority of interactions occur on a 1-day timescale, with a small fraction extending to 2 days and very few at 3 days or beyond. This suggests that there is short-term, intraspecific suppression for a wide range of different taxa at all body sites.
In Fig 4, we show how different taxa are involved in Granger causality relationships. Specifically, we show the number of positive/negative cause/effect interspecific interactions by genus. Meanwhile, Fig 5 shows the average time-lags associated with each taxon’s interspecific and intraspecific interactions respectively. As in Fig 3, Fig 5 reiterates the fact that interspecific interactions, even those that are relatively strong, occur over a range of timescales, whereas intraspecific interactions act primarily over short timescales. One notable exception to this trend is Pedobacter in right palm.
In S2 Fig (see S1 Text), we show a Venn diagram illustrating the number of shared taxa amongst different sets of body sites. The left- and right-hand are the most diverse, and also share the largest fraction of genera (73%). The gut is the least diverse and shares very few genera with the other three body sites (15%, 10% and 5% with the right-hand, left-hand and tongue respectively). This distribution of taxa constrains the number of interactions that can be conserved across body sites. However, even given these constraints, conservation is remarkably low. Amongst strong interspecific interactions, for example, there is only one that is conserved across two body sites, and that is a positive interaction between Micrococcus and Veillonella on the left- and right-hand at a timescale of 20 days. (See S1 Text for an analysis of conservation for all interactions, both weak and strong). Strong intraspecific interactions are much more conserved, ranging from 33% between the gut and right-hand to 70% between the two hands.
One of the problems with considering individual time-lags separately is that it makes observing conserved interactions across sites very difficult. This is because conservation must be precise. Countering such precision is the fact that sampling almost certainly did not occur at the same time each day (for the gut, for instance, sampling is restricted to the timing of bowel movements). Likewise, similar processes may occur at somewhat different timescales at different body sites, particularly if resource availability or other environmental factors cause organisms to grow at different rates. For this reason, we now turn our attention to a qualitative analysis, where we consider any time-lag between 1–5 days as a ‘short’ interaction, and any time-lag between 15–20 days as a ‘long’ interaction. We ignore ‘short’ and/or ‘long’ interactions for any taxon pair with multiple coefficients at the short or long timescale with opposite signs. Fig 6 shows how interactions break down for strong interactions with coefficients in the top 5% by magnitude (see S2 Text for an analysis of all coefficients). As in Fig 3, we see that strong interspecific interactions are predominantly positive, and this is particularly true for long timescales. By contrast, strong intraspecific interactions are almost uniformly short and negative, again consistent with Fig 3. Although, no strong interspecific coefficients are conserved across three body sites, a few are conserved across the two hands. These are shown in Table 1. All strong intraspecific coefficients are conserved.
Strong positive or negative correlations between taxa can emerge as a result of species interactions [9] or via environmental filtering [10,11]. With existing research, which primarily focuses on basic assessment of correlation patterns [1–5,36,37], it is difficult to determine the underlying causes of observed species distributions. Highlighting the challenges associated with teasing out species interactions from standard microbiome datasets, Berry and Wider [38] simulated multi-species microbial communities by generating interaction patterns with generalized Lotka-Volterra dynamics. They found that co-occurrence networks can be a proxy for interaction networks under certain conditions; however, with significant habitat filtering, the interpretation of co-occurrence becomes problematic. Unfortunately, few microbiome correlation studies explicitly discuss the pitfalls associated with using correlation/co-occurrence metrics to infer ecological interactions [12,39]. A prior human microbiome study explicitly showed that correlation is not a reliable metric of interaction [39]. The conclusion of that study, however, comes from a simulation experiment with a set of specified conditions and not from the analysis of real-world data. Here, by analyzing four of the longest and densest time series from the human microbiome, we show that correlation is not a reliable proxy of ecological interaction (measured with Granger causality) in human microbiome, and indeed the two measures are weakly negatively related across the dataset (Fig 1). As Granger causality is beginning to see its application in microbiome studies [32], many types of interactions can be elucidated with these causal models. Unfortunately, ground-truthing of the causal models is highly limited at present because of very few in-vitro studies (e.g., [40]).
Sometimes, knowledge of the biological system under investigation can help to inform models of community interactions; however, this is not always the case. Particularly for complex microbiomes, a deeper understanding of biology is often insufficient to resolve conflicting hypotheses. A genome-scale metabolic modeling of the human gut microbiome, for example, found that species that strongly compete with each other (i.e., species with highly similar nutritional profiles) tend to co-occur, whereas species pairs that co-occur least often have dissimilar nutritional profiles, suggesting that environmental filtering is the main driver of community structure [12]. In contrast, a subsequent community metabolic model assembled from models of species level metabolic exchanges analyzed >800 microbial communities and found that species interactions, in particular metabolic dependencies, are a “major driver of species co-occurrence” [41].
One solution for trying to identify species interactions in complex communities is to move beyond correlation (t = 0) to focus on causation (i.e., t > 0 correlation). Stated simply, this is the temporal dependence of one taxon on another. Although many studies identify a ‘core human microbiota’ which is stable over long timescales [35,42–44], there is still significant short timescale variation [35,42], making it possible to examine causal relationships within community dynamics. In the present study, we used the longest available time series of human microbiome dynamics [35] to elucidate causal networks among constituents of the human microbiome.
Interestingly, we find that strong interspecific interactions tend to be positive (see Fig 3). This is in opposition to the few experimental studies that exist. For example, Foster and Bell [45] analyzed overall respiration, and found that the great majority of interactions are net negative. Similarly, by culturing artificial microbial communities of 1–12 species for 60 generations and comparing community yield against the sum of species’ yields in monoculture, Fiegna et al. [46] demonstrated that interactions are, in general, negative, although they also showed that interactions become less negative over time.
Interestingly, our analysis also shows a tendency towards more positive interactions over longer timescales (see Fig 6), at least for strong interactions. One reason we may detect more positive interactions overall is that our analysis is based on data from communities in vivo, where cell-cell adhesion and formation of complex biofilms may be important for persistence. Physical requirements for persistence in environments like the gut or oral cavity may outweigh metabolic competition in terms of the net degree of mutualism/facilitation amongst community members.
Our finding that intraspecific interactions tend to occur on short timescales and tend to be strongly negative (see Fig 3) suggests that individual populations are kept in check not by other bacterial taxa, but rather by factors intrinsic to themselves. This might be resource limitation; however, for this to be the case, resource use would have to lead to overcompensatory dynamics, such that a large population on one day led to a crash the following day. Though not impossible, a more likely explanation is natural enemies, such as phages. In essence, then, the picture that emerges for maintenance of biodiversity in the human microbiome is one of a temporal Jansen-Connell effect. In particular, when any single population begins to dominate, ‘density-dependent’, host-specific pathogens attack the population, leading to collapse. An analogous way to view our findings is from a Lotka-Volterra competition framework. Specifically, we find stable coexistence among large numbers of microbes because each member of a pair of species inhibits its own population growth more strongly than it inhibits the population growth of others [47].
Comparing body sites, we find very few shared interactions, even when accounting for taxonomic differences in community composition among the gut, hands and tongue. Indeed, for our full quantitative analysis with all time-lags (see S1 Text, S9 Table), we find <1% of interspecific coefficients conserved across the tongue, and two hands, which share a total of 18 genera. Intraspecific models are more conserved, although even for these, only 13–20% of coefficients are shared across three or more sites (see S1 Text, S9 Table). When we use a qualitative analysis, results improve, although even here, very few interactions remain across multiple body sites (see Table 1 and S2 Text). Examining qualitative interactions that are conserved across the tongue, left- and right-hands, it is interesting to note that there are clusters of negative interactions that align with known spatial segregation of specific taxa, suggesting the potential for niche competition. For example, there is inhibition between Leptotrichia, Capnocytophaga and Fusobacterium (see S2 Text, S11 Table), all of which are predominantly found in a wide band just inside the periphery of ‘hedge-hog’ structures in plaque [48]. Meanwhile, we see another distinct cluster of negative interactions involving Rothia, Haemophilus, Neisseria and Veillonella (see S2 Text, S11 Table). Notably, Rothia, Veillonella and Haemophilus are members of ‘cauliflower’ structures within plaque [48]. Meanwhile, Rothia and Neisseria are both early colonizers of oral cavity surfaces, again suggesting niche overlap where competition might occur [49].
Contrasting Granger causality and more standard correlation analysis with Pearson Correlation coefficients, it is interesting to note that there does appear to be some relationship between the two for short timescale interactions. Contrary to what is expected, this relationship is, however, negative. That is, positive Pearson correlation coefficients are more likely to have negative Granger causality coefficients, at least at short timescales. Long timescale Granger causality results do not appear to be related to Pearson correlation except, perhaps, minimally in the gut. The explanation for these observations is unclear, but may point to a tendency to compete with other taxa that share a similar environmental niche.
Two previous studies [33,50] have taken a time series regression approach to the same microbiome dataset that we analyzed. However, neither qualify as a Granger causality analysis, differing from ours in two key ways. First, their methods did not integrate cross-prediction (where a given taxon is predicted by both its own lagged states and the suspected Granger cause, and vice versa). In our model, a causal relationship is established only when the cause significantly improves prediction of a model that already includes lagged states of the response variable. Second, their analysis included only the first lag. There is no reason to believe interactions completely disappear after one day, and indeed we found significant terms out to 20 days.
Biological interactions of microbial communities are so little known that mapping out the architecture of interactions is currently a formidable challenge [19,51]. Therefore, statistical analysis of correlation/co-occurrence networks, which are increasingly available because of high-throughput cross-sectional sequencing, could serve as a first approximation of biological interactions. However, making the leap from co-occurrence data to causality requires statistical tools that can actually elucidate causation, which is not possible with a simple correlation approach. Indeed, correlation networks may have no relationship to interaction networks. Therefore, in the absence of guiding assumptions about ecological interactions, Granger causality and related techniques may be particularly helpful for understanding the driving factors governing microbiome composition and structure.
Granger causality is, in essence, a method for determining taxon correlations through time. As a consequence, a minimum requirement is a well-resolved microbiome time series. For our work, we used data from Caporaso et al. [35], which is the longest human microbiome time series to date. In brief, this dataset comprises 16S sequences for all bacterial taxa over a period of 336–373 days from four body sites (Table 2). Although the original study considered two human subjects—one male and one female—we focused on the male data, since this subject was followed over a longer period of time. To simplify our analysis, we performed all calculations on genus level taxonomic assignments. Further, we only considered genera with a measurable abundance in at least 90% of the time series. However, even for these taxa, up to 10% of data points may be absences. Because absences can be problematic for Granger causality analyses, we replaced absences with randomly sampled small abundances (a random number generated between 10−5 and 10−3 of the mean of the time series). The assumption is that genera that are present on >90% of days were not actually missing on the other days, but were instead not detected because their abundances fell below the detection threshold [35]. For all of our analyses, we first calculated relative (i.e., normalized) abundances. However, because genera can differ in their relative abundances by several orders of magnitude, we performed our analyses on standard deviation changes in abundance. Specifically, we standardized the relative abundance of each genus by its own mean and standard deviation across the time series. Finally, we removed stochastic and deterministic trends by first-differencing. Analyses were performed separately for each of the four body sites.
The original Granger causality model included only two time series—the potential Granger cause and the response—as predictors [28]. Two types of processes can lead to spurious causation in such pair-wise analysis of causality. First, when taxon A interacts with taxa B and C at different time lags, but taxa B and C are causally independent, an analysis showing causal relationship between taxa B and C is incorrect. Second, when the flow of causation is from taxon A to B and then from taxon B to C, a detection of causation between taxa A and C is spurious. These two situations—the first one called as “differentially delayed driving” and the second one as “sequential driving”–have been tested in the traditional, pair-wise framework of Granger causation [52]. In a simulation experiment of 500 realizations with each of them being 100 timesteps long, Chen et al. [52] showed that the classical pair-wise Granger causality identifies spurious causation resulting from both aforementioned processes as the true causation. However, when Chen et al. performed a multivariate or conditional Granger causality analysis (including all three time series in the model), the spurious causations detected by the pair-wise Granger analysis were removed whereas all true causations were retained [52]. Realizing the power of multivariate methods, several multivariate extensions of Granger causality [53–56] have been developed recently, and we have implemented one of these multivariate approaches here.
The superiority of multivariate Granger causality over the traditional, pair-wise approach can be explained by statistical theory. In a model with two predictors, the coefficients reflect the effect of one predictor when the other predictor variable is also included in the model and is held constant [57]. This partial effect of a predictor is the unique effect of the variable in predicting the response variable [58]. As a consequence, the spurious causations detected using a pair-wise Granger causality are correctly eliminated by using a multivariate/conditional Granger causality analysis. Since both the statistical insights, as well as prior simulation experiments, show that multivariate or conditional Granger analysis eliminates the cases of spurious causations that plague the traditional pair-wise analysis of Granger causation, we applied a multivariate approach of Granger causality in this current study.
Another advantage of multivariate analysis is a dramatic reduction in the number of hypothesis tests and false positives. In a standard Granger causality analysis, the number of hypotheses tested would be 2*(n2) for n taxa. In our dataset, for example, this would yield 253 hypotheses for the gut microbial community and 1326 for the right palm. With the conventional approach of hypothesis testing with alpha of 0.05, we can potentially have up to 66 false positive cases of causation in the right palm alone (i.e., significant causation identified when no causation exists). By using a multivariate approach, however, the number of hypotheses tested is reduced to 23 and 52 for gut and right palm, respectively, and the number of false positives in the right palm is reduced to 3.
To take advantage of the multivariate approach, we implemented multivariate/conditional analysis of Granger causality in this study. To construct multivariate models, we assumed that the relative abundance of any particular genus could be dependent on all other genera in the community; we then considered 20 time-lags (for a maximum lag of approximately 20 days). We chose this time range because we found that lags beyond 15–17 days rarely improve the models. Because we wanted to include multiple time lags as predictors, we were faced with a large over-parameterization problem. For example, each response genus in the gut had 460 predictors, while each response genus on the right palm had 1040 predictors (Table 2). To reduce parameters and ensure predictive power, we applied what is known as a least absolute shrinkage and selection operator (LASSO) [59]. LASSO performs both regularization and variable selection by shrinking large coefficients and eliminating smaller ones. This results in a simpler model with better interpretability and stability. To perform LASSO, we tested a range of shrinkage parameters, selecting the best shrinkage parameter based on it, and yielding a model with minimum prediction error when tested against independent data (cross-validation).
Completion of model building was achieved in three steps that included rolling cross-validation [60,61]. Specifically, we built a model using only the first one-third of the time series. A range of penalty parameters were then tested by sequentially adding one observation at a time from the second third of the time series; the best penalty parameter was selected so as to minimize the mean-squared forecast error. Finally, independent validation of the model was performed using the final third of the time series. This method of rolling cross-validation assisted LASSO shrinkage of the model eliminated about three-quarters of the total parameters (Table 2). Whereas cross-validation is considered a gold standard of model evaluation, there is an additional reason to apply such methods to microbiome datasets: taxon associations in human gut microbiota have been shown to vary in strength over time [34]. This makes cross-validation of the model crucial for model evaluation and selection. Granger causality analysis was performed in R (version 3.4.1) using the “BigVAR” package [62]. For a visual presentation of the sequence of the overall methods in the study, we developed a flow chart (Fig 7).
Although prior simulation [52] and statistical insight [57,58] (discussed above) validate the multivariate/conditional Granger causality for its strengths of identifying true causation and eliminating spurious causation detected by pair-wise Granger analysis, we went a step further and determined the robustness of our model by reshuffling the data we analyzed and determining how many causations are detected in the randomized data. When the time series are randomized, the temporal relationship between lags should disappear and so, ideally speaking, no causations should be detected (some false positives are always allowed because of the non-zero Type I error [alpha] of the model). We calculated the number of strong coefficients detected from reshuffled time series as well as from real data. We applied the same threshold to determine strong coefficients in the actual and randomized data for a given body site (see “Full quantitative analysis” under the Results section for definition). The number of taxa-pairs with strong coefficients detected using the reshuffled data was a mere 7% of the number of strong coefficients detected in the actual data (S12 Table). This gives us high confidence that, despite the complexities in the model and data, the composite modeling approach we used, which goes beyond traditional Granger analysis, has detected true signal, as expected under a robust statistical model with reasonable tolerance of false positives.
Although we detected 41 interacting taxa-pairs with strong coefficients using the reshuffled data across all four body sites, only five of those taxa- pairs were found in the results of the actual data. We eliminated those five taxa-pairs from the results presented in this study.
Relative abundance data suffers from the problem of compositionality which can yield spurious correlation. There have been attempts to deal with this problem. Faust et al [2] proposed a method to generate an appropriate null distribution of correlation by permutation and renormalization, accounting for the compositional structure of the data. Although this approach yields a null appropriate for compositional data in a pairwise analysis, it does not alter the estimate itself, which makes it irrelevant to our multivariate analysis. Another study by Friedman and Alm [15] went a step further and proposed a new estimate of correlation. However, their mathematical formulation of this metric was developed for pairwise comparison and there is no straightforward way to include this in a multivariate regression framework. Although our analysis cannot rely on these recently developed approaches to minimize the impact of compositionality, we have employed a stringent LASSO shrinkage of the model that eliminated three-fourths of the coefficients, retaining only the very strongest and most highly significant relationships. On top of that, most of the results we have shown are for the strongest 5% of all the significant results. Additionally and importantly, our analysis does not suffer from one of the key problems of compositional data: singularity of the design matrix for the regression yielding no unique solution to the ordinary least square problem [39,63]. In our analysis, because on average only one fourth of the parameters are used, the regression model is well-defined.
After applying all the procedures, we tested the results against that of reshuffled time series of the actual data. This test shows that the false positives of our results is likely to be about 7% on average across body sites. Whereas we do not have a straightforward way of determining the extent compositionality affected our result, the reshuffled results give us high confidence in our results.
|
10.1371/journal.ppat.1007484 | EBV-miR-BART1-5P activates AMPK/mTOR/HIF1 pathway via a PTEN independent manner to promote glycolysis and angiogenesis in nasopharyngeal carcinoma | Abnormal metabolism and uncontrolled angiogenesis are two important characteristics of malignant tumors. The occurrence of both events involves many key molecular changes including miRNA. However, EBV encoded miRNAs are rarely mentioned as capable of regulating tumor metabolism and tumor angiogenesis. Here, we reported that one of the key miRNAs encoded by EBV, EBV-miR-Bart1-5P, can significantly promote nasopharyngeal carcinoma (NPC) cell glycolysis and induces angiogenesis in vitro and in vivo. Mechanistically, EBV-miR-Bart1-5P directly targets the α1 catalytic subunit of AMP-activated protein kinase (AMPKα1) and consequently regulates the AMPK/mTOR/HIF1 pathway which impelled NPC cell anomalous aerobic glycolysis and angiogenesis, ultimately leads to uncontrolled growth of NPC. Our findings provide new insights into metabolism and angiogenesis of NPC and new opportunities for the development of targeted NPC therapy in the future.
| The Epstein-Barr virus (EBV), the first reported human tumor virus found to encode miRNAs, which closely related to malignant progression of tumors. In our study, we have observed that EBV-miR-BART1-5P, an EBV-BARTs encoded miRNA, promotes glycolysis and induces angiogenesis in NPC. Interestingly, we showed that overexpression of EBV-miR -BART1-5P and restored PTEN at the same time, did not completely reverse the phenotypes of glycolysis, angiogenesis and proliferation, suggesting that EBV-miR-BART1-5P can mediate glycolysis and induction angiogenesis by a PTEN-independent manner. Further mechanism exploration demonstrated that EBV-miR-BART1-5P has important roles in cancer cell glucose metabolism and angiogenesis by inhibiting AMPKα1 and PTEN, which provides a molecular basis for the regulation of AMPK/mTOR/HIF1 and PTEN/FAK, Shc, AKT pathways, respectively.
| The development of malignant tumors is divided into several stages: malignant transformation of cells, clonal proliferation of transformed cells, local infiltration and distant metastasis. An abnormal supply of energy and sustained angiogenesis, two of the ten characteristics of tumor development [1], maintain the growth during the different stages of the cancer. They play an important role in cancer progression, including regulation of cancer growth, invasion and metastasis [2].
Cancer cells have a unique energy metabolism phenotype that consumes more glucose and converts pyruvate to lactate, even under normoxia, which is called the Warburg effect or aerobic glycolysis [3]. Aerobic glycolysis gives cancer cells a growth advantage not only by providing more glycolytic intermediates for various biosynthetic pathways, but also by minimizing the production of reactive oxygen species in the mitochondria [4,5]. Due to the rapid proliferation of cancer cells, hypoxia occurs to which cancer cells adapt by upregulating their glycolysis. This also leads to an increased acid production, which leads to a significant decrease in the local extracellular pH. The microenvironment acidification promotes cancer invasion and angiogenesis by disrupting adjacent normal cells and by acid-induced extracellular matrix (ECM) degradation [6–8]. However, in a variety of tumors, including nasopharyngeal carcinoma (NPC), the molecular mechanism leading to abnormal aerobic glycolysis remains obscure.
MiRNAs are highly conserved noncoding RNAs that regulate a variety of biological processes [9]. In recent studies, multiple cellular miRNAs, such as miR-199a-5p[10], miR-143 [11–15], miR-451 [16], miR-210 [17], miR-29b [18], miR-195-5p [19], miR-375 [20,21] have been reported to participate in the energy metabolism process by regulating the gene expression of metabolic-associated genes through posttranscriptional repression and mRNA degradation [22]. It is known that virus-encoded miRNAs can also regulate cell energy metabolism and angiogenesis[23–26], but the underlying mechanism is still largely unknown.
The Epstein-Barr virus (EBV), which is aetiologically linked to several cancers including Hodgkin lymphoma, Burkitt’s lymphoma, gastric cancer and NPC [27], is the first reported human tumor virus found to encode miRNAs [28]. Until now, EBV encodes 48 mature miRNAs that have been identified within two regions of the EBV genome. The BamHI fragment H rightward reading frame 1 (BHRF1) gene generating four mature miRNAs and the BamHI fragment A rightward transcript (BART) region producing 44 mature miRNAs. BART-miRNAs are highly expressed in epithelial malignancies including NPC and EBV-associated gastric cancers [29]. At present, EBV-miR-BARTs have become more and more important in the development of NPC where they have been reported to participate in a series of pathological processes such as proliferation, apoptosis, invasion and metastasis [30]. In a previous study, we have demonstrated that EBV-miR-BART1-5P and EBV-miR-BART1-3P directly target the PTEN-AKT signaling pathway to mediate NPC cells metastasis [31]. However, few articles have linked EBV-miR-BARTs to aerobic glycolysis and angiogenesis.
In this study, we have observed that EBV-miR-BART1-5P promotes glycolysis and induces angiogenesis in NPC. The underlying molecular mechanism revealed to directly target the α1 catalytic subunit of AMP-activated protein kinase (AMPKα1) and consequently regulates the AMPK/mTOR/HIF1 pathway which ultimately leads to growth of NPC cells.
In aerobic glycolysis, glucose is the starting material, while lactate is the final product. To validate the roles of BART1 in NPC glycometabolism, we detected the secretion of lactate and the consumption of glucose by transiently transfecting BART1-3P and BART1-5P mimics into 7 cell lines (2 EBV-negative NPC cell lines and 5 EBV-negative epithelial cell lines) respectively. Compared with BART1-3P, BART1-5P significantly increased the production of lactate and the consumption of glucose in all 7 cell lines (S1 Fig). This suggests that BART1-5P affects more genes related to glucose metabolism.
To further determine the role of BART1-5P on NPC glycometabolism we used lentiviral particles carrying BART1-5P precursor to generate two EBV-negative cell lines (One is NPC Cell line, another is an EBV-negative epithelial cell line.) stably expressing BART1-5P (HONE-BART1-5P and CNE1-BART1-5P). The expression levels of BART1-5P in these two cell lines were within a similar physiological range to pooled NPC tissue samples (S2 Fig). NPC cells with exogenous expression of BART1-5P secreted more lactate (Fig 1A), consumed more glucose (Fig 1A) and exhibited more cellular ATP levels (S3A Fig). Moreover, expression of BART1-5P significantly increased recipient cells' uptake of 2-NBDG, a fluorescent analogue of glucose which has been used to assess glucose transport in various cell types [32,33] (Fig 1A). On the contrary, silencing of BART1-5P reversed the changes in the secretion of lactate, consumption of glucose, uptake of glucose (Fig 1B) and ATP levels (S3B Fig). GLUT1, HK2 and LDHA are pivotal enzymes in glucose metabolism catalyzing many key steps in the glycolytic pathway. Hypoxia-inducible factor (HIF)-1α has been identified as a contributor to aerobic glycolysis by regulating of expression of various glycometabolism-associated genes [34]. We observed that BART1-5P significantly increased the expression of GLUT1, HK2, LDHA and HIF-1α (Fig 1C). Furthermore, downregulation of BART1-5P reduced the GLUT1, HK2, LDHA and HIF-1α levels (Fig 1D). Interestingly, overexpression of PTEN did not completely attenuate the effect of BART1-5P on glucose metabolism (Fig 1A–1D), suggesting that PTEN is not an independent factor affecting glucose metabolism.
The above results collectively indicate that BART1-5P enhances glycolysis in NPC cells.
In the majority of solid tumors, as the tumor grows, it becomes more and more difficult for the inner cancer cells to obtain sufficient oxygen from the blood. As a result, HIF-1α, a subunit of the heterodimeric transcription factor HIF-1, is overexpressed in the hypoxic microenvironment of most human cancers. HIF-1α regulates vascular endothelial growth factor (VEGF) [35], the principal mediator of angiogenesis in a number of cancers, under normal physiologic conditions. In the present study, we found that BART1-5P promotes the expression of HIF-1α. To directly detect whether BART1-5P promotes angiogenesis, the chorioallantoic membrane (CAM) assay was used. We stably expressed BART1-5P in HONE1, CNE1 and HK1 cells and applied the cell supernatant over chorioallantoic membranes via sponges [36]. BART1-5P, but not NC, significantly promoted neovascularization (Fig 2B and 2D and S4A Fig). In contrast, silencing BART1-5P significantly reduced angiogenesis (Fig 2B and 2D and S4B Fig). A similar result was seen in the endothelial tube formation assay. Compared with the NC, upregulation of BART1-5P significantly increased the tube formation potential. In agreement with the upregulation results, downregulation of BART1-5P reduced the tube formation potential (Fig 2A and 2C). Next, an in vivo matrigel plug assay was performed using conditioned media collected from BART1-5P or NC NPC cells. The results showed that BART1-5P overexpression led to a substantial increase in haemoglobin (Fig 2E), a surrogate marker for functional blood flow [37]. Earlier it was shown that PTEN could also inhibit angiogenesis by regulating the expression of VEGF expression through AKT activation [38]. At the protein level, BART1-5P significantly promoted the expression of VEGF, whereas restoring the expression of PTEN in BART1-5P/anti-miR treated NPC cells only partially influenced the expression of VEGF (Fig 2F). A similar phenomenon can be observed in the CAM (Fig 2B and 2D and S4 Fig), endothelial tube formation (Fig 2A and 2C).and matrigel plug assay (Fig 2E), suggesting that BART1-5P affects multiple genes including PTEN to promote angiogenesis in NPC.
Regarding energy metabolism, BART1-5P promotes potential NPC cell proliferation. Hence, we examined the effect of BART1-5P expression on the growth of NPC cells in vivo and in vitro. Using colony formation (Fig 3A and 3C), Edu incorporation assays (Fig 3B and 3D) and cell cycle analysis (S5A and S5B Fig), we observed that BART1-5P significantly promoted cell growth and G1/S transition in NPC cells. Conversely, silencing BART1-5P significantly inhibited cell proliferation and mediated G1/S transition in NPC cells (Fig 3A–3D). We also found that BART1-5P overexpression enhanced CCND1 expression but downregulated the expression of p27 and p21. BART1-5P inhibitors rescued these effects (Fig 3G and 3H).
Furthermore, HONE1-BART1-5P cells and HONE1-NC cells were subcutaneously injected into nude mice. Upregulation of BART1-5P significantly increased tumor growth when compared with NC (Fig 3E). To further confirm that BART1-5P can increase tumor growth, we inject nude mice with HONE-EBV-miR-ctrl and HONE-EBV-BART1-5P-antago-miR (Purchased from Shanghai GenePharma Co., Ltd) cells to evaluate the size of tumor decline. Experimental results showed that antago-mir-BART1-5P decreased tumor growth when compared with antagomir-ctrl (S7 Fig).
In 55 NPC tissues, we observed that the expression of BART1-5P was positively correlated with the T stage (Fig 3F) and clinical stage of NPC (Fig 3F), indicating a correlation between BART1-5P and the growth of NPC. Restoring PTEN expression in these experiments yielded similar results as in the glycolysis and angiogenesis experiments (Fig 3). These results suggest that BART1-5P promotes the growth of NPC cells.
To investigate glucose metabolism and angiogenesis-related BART1-5P-target genes we performed deep sequencing. Predicted EBV-miR-BART1-5P target genes were retrieved from TargetScan and RNAhybird. One of the potential candidates AMPKα1 is an important energy sensor that regulates the cellular metabolism to maintain energy homeostasis and affects angiogenesis by activating the mTOR pathway. Bioinformatics analysis showed that the 3'UTR of AMPKα1 was well matched with the seed sequence of EBV-miR-BART1-5P (Fig 4A).
To clarify whether BART1-5P could directly target AMPKα1, we performed a luciferase reporter assay. BART1-5P significantly reduced the luciferase activity of the wt AMPKα1 3'UTR reporter vector and this effect was abolished when the AMPKα1 3'UTR binding site was mutated. In addition, anti-miR increased the luciferase activity of the wt AMPKα1 3'UTR reporter vector rather than of the mt 3'UTR (Fig 4B).
We detected the effects of BART1-5P on AMPKα1 mRNA and protein expression in NPC cell lines, xenografted tumors and tissue samples. The expression of AMPKα1 was significantly lower in both HONE-BART1-5P and CNE1-BART1-5P cell lines and in xenografted tumors when compared with the relative controls as revealed by western blotting (Fig 4C) and immunohistochemical (IHC) staining (Fig 4D). We detected AMPKα1 expression in 55 NPC and 15 NP tissue samples. The results showed that the downregulation of AMPKα1 in clinical tumor samples was associated with the T stages and advanced clinical stage of NPC (Fig 4E). Moreover, the expression of AMPKα1 was negatively associated with BART1-5P expression (Fig 4F). we also collected 20 new nasopharyngeal carcinoma (NPC) samples and 10 chronic nasopharyngitis (NP) samples to be used for immunohistochemical staining. The result showed the downregulation of AMPKα1 in NPC samples compared with the NP samples (S6 Fig).
Collectively, these results indicate that AMPKα1 is a direct cellular target of EBV-miR-BART1-5P in NPC.
We examined alterations in the expression of key components of the AMPKα1 pathway in NPC [39,40]. We observed that the upregulation of BART1-5P significantly reduced the protein expression of AMPKα1 but increased the level of mTOR, p-mTOR and S6K1, whereas silencing of BART1-5P attenuated mTOR, p-mTOR and S6K1 levels when compared with the relative control (Fig 5A). Further, we found that the protein levels of mTOR, p-mTOR, S6K1, VEGF, HIF-1α, HK2, GLUT1, LDHA and CCND1 were significantly increased, while AMPKα1, p21 and p27 were reduced after AMPKα1 siRNA treatment (Fig 5B). Restoring PTEN expression in NPC cells could only partially restore the expression of mTOR and downstream proteins (Fig 5A). IHC staining of sections of tumor xenografts showed that upregulation of BART1-5P significantly reduced the expression of AMPKα1 but increased HIF-1α, GLUT1, LDHA and mTOR when compared with the relative control (Fig 5C). These data indicate that EBV-miR-BART1-5P activates the AMPK/mTOR/HIF1 signaling pathway.
We transfected HONE-BART1-5P cells with the AMPKα1 expression vector, the PTEN expression vector or both. Restoration of either AMPKα1 or PTEN expression significantly reduced the secretion of lactate, consumption of glucose (Fig 6A), angiogenesis (Fig 6B and 6E) and proliferation (Fig 6C and 6E) of NPC cells when compared with the negative vector control. Western blotting showed that the reconstitution of AMPKα1 and PTEN decreased the level of mTOR, p-mTOR, S6K1, VEGF, HIF-1α, HK2, GLUT1, LDHA and CCND1 but increased p21 and p27 when compared with NC (Fig 6D). As a result, restoration of AMPKα1 and PTEN reduced glycolysis, angiogenesis and proliferation of NPC cells.
Moreover, we detected related proteins in AMPK/mTOR/HIF pathway after treatment with aminoimidazole-4-carboxamide (AICAR, an activator of AMPK, purchased from Sigma-Aldrich, catalog number A9978, 100μM) and AMPK inhibitor Dorsomorphin. As expected, the experimental results of AICAR treatment exhibit AMPK/mTOR/HIF pathway activating effect as similarly to AMPKα1 rescue in stably expressing BART1-5P cell lines (S8 and S9 Figs). we also tested the key proteins in the AMPK/mTOR/HIF pathway after treatment with AMPK inhibitor Dorsomorphin (10μM, Sigma-Aldrich, catalog number P5499). Experimental results of AICAR treatment exhibit AMPK/mTOR/HIF pathway inhibiting effect as similar to si-AMPKα1 in HONE1-EBV cells (S10 and S11 Figs).
We quantified the expression of EBV-miR-BART1-5P in three EBV-positive NPC cell lines (C666-1, HONE1-EBV, and HK1-EBV) and compared it with NPC clinical samples by qRT-PCR (S13 Fig). We used the HONE1-EBV cell line as a representative to further validate the role of BART1-5P in NPC glucose metabolism, angiogenesis and to clarify the molecular mechanisms.
First, we detected the expression of HIF-1α, GLUT1, LDHA and the related proteins of AMPK/mTOR/HIF pathway within HONE1-EBV and HONE1 cell lines by western blotting assay. The result was displayed in the supplementary files (S12A Fig). Further, BART1-5P significantly increased the expression of HIF-1α, GLUT1, LDHA and the related proteins of AMPK/mTOR/HIF pathway in normal epithelium (NP460 cells) cells (S12B Fig).
HONE-EBV cells were transfected with anti-miR, anti-miR and si-AMPKα1, anti-miR and si-PTEN or with all three, respectively. The secretion of lactate, consumption of glucose (Fig 7A), angiogenesis (Fig 7B) and proliferation (Fig 7C) of NPC cells gradually increased when compared with NC. An additive effect was obtained when both si-AMPKα1 and si-PTEN were co-transfected. Similarly, western blotting demonstrated that downregulation of endogenous BART1-5P increased AMPKα1, P21 and P27 expression but reduced mTOR, p-mTOR, S6K1, VEGF, HIF-1α, HK2, GLUT1, LDHA and CCND1 expression when compared with the anti-control (Fig 7D). Thus, these data validate that BART1-5P promotes glycolysis, angiogenesis and proliferation of NPC cells by regulating the AMPK/mTOR/HIF1 signaling pathway.
Tumor cells need to adjust their energy metabolism for cell growth and division as rapid growth of solid tumors can lead to tissue hypoxia. By switching to glycolysis, tumor cells successfully adapted to local hypoxia. Furthermore, acidification of the microenvironment due to an increased lactate concentration can promote tumor proliferation [41], invasion and angiogenesis [42,43]. The key regulator of the glycolytic response is the transcription factor HIF-1α, which activates a number of pivotal enzyme in glucose metabolism and catalyzes the irreversible rate-limiting step and also activates VEGF to promote angiogenesis [4]. In fact, there is growing evidence that the ‘glycolytic switch’ occurs before the ‘angiogenic switch’ [44], and that increased glucose uptake is observed to coincide with the transition from pre-malignant lesions to invasive cancer [45,46]. This further suggests that the glycolytic phenotype plays a vital role in the development and growth of tumors.
NPC is typically characterized by rapid progress and high metastatic potential [47]. Investigating the key mechanisms in the development of NPC contributes to a thorough understanding of the molecular mechanisms underlying the development of tumors and to develop new clinical management strategies. Previously, we observed that EBV-miR-BART1 is highly expressed in NPC and is related to patients with advanced stages. Further, EBV-miR-BART1 promotes NPC cells invasion and metastasis by directly targeting PTEN [31]. At present, we found that EBV-miR-BART1-5P activates the AMPK/mTOR/HIF1 pathway by targeting AMPKα1 to upregulate the glycolysis of NPC cells, to induce angiogenesis, and ultimately to promote the growth of NPC cells.
EBV-miRNA-BARTs regulate both viral and cellular genes in NPC cells. Such as, EBV-miR-BART3* promotes the growth and transformation of NPC cells by incompletely matching with the Dice1 gene [48]. EBV-miR-BART9 promotes tumor cell invasion by targeting E-cadherin, while targeting PTEN promotes tumor cell proliferation [49]. EBV-miR-BART5 targets PUMA to protect NPC cells from apoptosis [50]. EBV-miR-BART7-3p promotes an EMT phenotype by targeting PTEN, eventually leading to NPC metastasis [51]. Previously, we observed that large numbers of metabolically related genes were abnormally expressed when EBV-miR-BART1 was overexpressed in NPC cells, suggesting that EBV-miR-BART1 may contribute to NPC energy metabolism [52]. Pre-experimental results suggested that EBV-miRNA-BART1-5P, but not EBV-miRNA-BART-3P, plays a critical role in glycolysis in NPC cells. To verify this hypothesis, we regulated the expression of EBV-miRNA-BART1-5P in both EBV-negative and EBV-positive NPC cells and found that EBV-miRNA-BART1-5P significantly promotes glycolysis and induces angiogenesis of NPC cells.
The target gene and related signaling pathway of EBV-miR-BART1-5P were found by RNA-deep sequencing, bioinformatics prediction, literature search and luciferase reporter assay. We demonstrate for the first time that metabolic sensor AMPKα1 is the critical cellular target of EBV-miR-BART1-5P in NPC. Exogenous EBV-miR-BART1-5P expression can attenuate the expression of endogenous AMPKα1, and AMPKα1 re-expression was able to reverse the EBV-miR-BART1-5P-mediated phenotypes.
AMPK is the key energy sensor for the body and the main regulator of cellular and organic energy stabilization. It coordinates a variety of metabolic pathways to balance supply and demand and ultimately regulate cell and organ growth [53]. Furthermore, AMPK has been involved in the regulation of tumorigenesis [54]. The major kinase LKB1, upstream of AMPK, is a defective gene in Peutz-Jeghers syndrome. Considering that Peutz-Jeghers syndrome is a rare hereditary disease that is prone to tumor formation, suggests that the LKB1-AMPK axis will be an important cancer suppressor pathway [55–58]. Another study confirmed that AMPK signal activator metformin and AICAR can inhibit cancer cell growth and tumorigenesis [59]. Some studies suggested that reduced expression of AMPKα2 has been linked to primary breast cancer, gastric cancer and ovarian cancer but is rarely involved in NPC [60,61]. We observed that the expression of AMPKα1 is lower in NPC and negatively correlated with the expression of EBV-miR-BART1-5P. Moreover, from a metabolic standpoint, inactivation of AMPKα1 in both transformed and non-transformed cells facilitates conversion to aerobic glycolysis and increases glucose distribution to lipids [62]. On the other hand, AMPK activation inhibits downstream AKT, mTOR, HIF1a expression and inhibits glycolysis [63]. In this study, we found that EBV-miR-BART1-5P downregulated AMPKα1 expression, which could increase the expression of mTOR and HIF1a in NPC cells. These observations support that EBV-miR-BART1-5P mediated glycolysis and induced angiogenesis occurs by targeting AMPKα1 to activate the AMPK / mTOR / HIF1 pathway.
In summary, our results show that EBV-miR-BART1-5P has important roles in cancer cell glucose metabolism and angiogenesis by inhibiting AMPKα1, which provides a molecular basis for the regulation of AMPK/mTOR/HIF1 pathway. Our findings provide new insights into glycolysis and angiogenesis of NPC and new opportunities for the development of targeted NPC therapy in the future.
The clinical processes for all the clinical tissues specimens were approved by the Ethics Committees of Zhongshan People’s Hospital. Informed written consent was obtained from all patients. And all patients were adult with independent morals or legal entitlements.
Animal experiments were approved by the Ethical Committee for Animal Research of the Southern Medical University (protocol number: 2011–020) and conducted based on the state guidelines from the Ministry of Science and Technology of China. White Leghorn chicken eggs with 9–10 days of embryonation were used for the chicken chorioallantoic membrane assay.
2 EBV-negative NPC cell lines (HONE1 and HK1, previously provided by Professor S. W. Tsao, HKU), 6 EBV-negative epithelial cell lines (CNE1, 5-8F, 6-10B, SUNE1, HNE1 and CNE2) and HEK293T cells were obtained from the Cancer Research Institute, Southern Medical University, Guangzhou, China. The STR profiling of HONE1 cell line was showed in the supplementary file (STR profiling for HONE1). According to the International Cell Line Authentication Committee (ICLAC) database, CNE1, 5-8F, 6-10B, SUNE1, HNE1 and CNE2 cells may be contaminated by Hela cells. However, unlike Hela cells, which are resistant to EBV infection, CNE1 and CNE2 cells are susceptible to EBV infection in vitro[64]. So, in our study, we mainly focused on the effect of EBV-encoded miRNA-BART1 on host cell. This effect may be appropriate for all EBV-associated tumors Three EBV-positive NPC cell lines (C666-1, HONE1-EBV and HK1-EBV) and NP460, an immortalized human nasopharyngeal epithelial cell line, were kindly provided by Professor S. W. Tsao, University of Hong Kong. The STR profiling of these cells were conducted by Professor S. W. Tsao. [65]. NPC cell lines were cultured in PRMI-1640 (Invitrogen) supplemented with 10% fetal bovine serum (FBS) (Hyclone, Invitrogen), 100 U/ml penicillin and 100 ug/ml streptomycin. NP460, an immortalized human nasopharyngeal epithelial cell line, was cultured in defined KSFM medium supplemented with epidermal growth factor (Invitrogen, Carlsbad, SA). All cells were maintained in a humidified chamber with 5% CO2 at 37°C.
55 primary NPC tissues (no treatment before biopsy) and 15 non-cancerous nasopharyngeal tissues were collected from patients at the Zhongshan People’s Hospital, Guangdong, China. All specimens were staged according to the TNM classification and used for qPCR and clinical analysis. Only those NPC samples that contained >80% of homogeneous cancer cells on frozen cross-sections visualized by haematoxylin-eosin staining were included in the study. The pathologic stage of all specimens was confirmed according to the 1992 Fuzhou NPC staging system of China.
Total RNA was extracted from cells by Trizol reagent (Invitrogen), complementary DNA (cDNA) was synthesized with the PrimeScript RT reagent Kit (TaKaRa Bio, Inc., Shiga, Japan). PCR analyses were performed with SYBR PremixTag (TaKaRa). The primers used are shown in S2 Table. Small nuclear RNA RNU6B (U6 snRNA) and GAPDH expression were used for normalizing the expression of miRNA and mRNA, respectively. The qRT-PCR reactions for each sample were repeated three times in three independent experiments. The fold changes were calculated by using the relative quantification method (2−ΔΔCt).
Lentivirus (GV209, H1-MCS-CMV-EGFP) particles carrying EBV-miR-BART1-5P precursor (BART1-5P) or its flanking negative control sequence (NC) were constructed by GeneChem (Shanghai, China) and transduced into NPC cells and CNE1(a contaminated EBV negative epithelial cell line) following the manufacturer’s instructions. The virus-infected cells, being GFP positive, were sorted by a BD FACS Aria cell sorter 72h after transduction.
The expression vector GV230 containing the whole coding sequence of PTEN, AMPKα1 and the control vector GV170 were purchased from GeneChem (Shanghai, China). Both HONE-BART1-5P and CNE1-BART1-5P cells were transfected with 200ng plasmid DNA using Lipofectamine 2000 reagent (Invitrogen). 48 hours post transfection, the cells were harvested for qRT-PCR and western blotting analyses. The EBV-miR-BART1-5P mimic (5’-UCUUAGUGGAAGUGACGUGCUGUG-3’), EBV-miR-BART1-3P mimic (5’-UAGCACCGCUAUCCACUAUGUC-3’), EBV-miR-BART1-5P inhibitor(anti-miR, 2′-O-methyl modification) (5’-CACAGCACGUCACUUCCACUAAGA-3’), EBV-miR-BART1-3P inhibitor (anti-miR, 2′-O-methyl modification) (5’-GACAUAGUGGAUAGCGGUGCUA-3’) and associated nonspecific mimic (5’-UUGUACUACACAAAAGUACUG-3’) or inhibitor (5’-CAGUACUUUUGUGUAGUACAA-3’) controls were synthesized by GenePharma, Shanghai,China.
All cells were maintained in a humidified atmosphere of 95% air and 5% CO2 at 37°C, and seeded 24h prior to transfection. EBV-miR-BART1-5P, BART1-3P mimic, anti-miR, and their mock control were transfected into cells at a final concentration of 50 nmol/l using Lipofectamine 2000 (Invitrogen) in serum-free conditions. Six hours later, the medium was changed to fresh RPMI-1640 (Invitrogen) with 10% fetal bovine serum (Hyclone, Invitrogen).
Total RNA from CNE1-BART1 cells or mock control cells was extracted with Trizol Reagent (Invitrogen) according to the manufacturer’s introduction. RNA-deep sequencing was performed and analyzed in BGI-Shenzhen of China as previously described[31].
Cell pellets were lysed in RIPA buffer containing protease (Sigma-Aldrich) and phosphatase inhibitors (Keygen, China), and the protein concentration was determined using the BCA assay (Beyotime, Beijing, China). Proteins were separated by a 10% SDS-PAGE gel, and blotted onto a polyvinylidene difluoride membrane (Milipore, Billerica, MA, USA). The membrane was probed with the first antibody listed in S1 Table and then with the peroxidase conjugated secondary antibody. GAPDH and β-actin were used as protein loading controls. Western blotting bands were visualized by the eECL Western Blot Kit (CWBIO Technology) and captured with a ChemiDoc CRSt Molecular Imager (Bio-Rad).
For the colony formation assay, NPC cells were seeded in duplicate in 6-well culture plates at a density of 100 cells/well. After incubation for 14 days at 37°C, colonies were washed twice with PBS and stained with hematoxylin solution. The colonies composed of more than 50 cells were counted under a microscope. All the experiments were repeated at least three times. For the EdU incorporation assay, proliferating NPC cells were examined using the Cell-Light EdU In Vitro Imaging Kit (RiboBio) according to the manufacturer’s protocol. FACS assays of NPC cells were performed after transfection with NC, anti-c, EBV-miR-BART1-5P mimics, inhibitor and/or PTEN plasmid, si-PTEN as previously described[66].
All nude mice (4–5 weeks old, female) were purchased from the Central Animal Facility of the Southern Medical University. To assess tumor growth, 100 ul of HONE1-BART1 cells or mock control cells (5x106) were subcutaneously injected into the left or right side of the back of each mouse (six mice per group). The tumor sizes were measured regularly and calculated using the formula 0.52 x L x W2 where L and W are the long and short diameter of the tumor, respectively.
HONE1 cells transfected EBV-miR-BART1-5P (50nM) alone or co-transfected EBV-miR-BART1-5P and PTEN plasmid. Cells were exposed to serum-free media for 48 h. We then collected the supernatants and centrifuged them to remove cells. The conditioned media were then mixed with phenol-red-free Matrigel (2:3 proportion, total 0.5 ml; BD Biosciences). The mixture was then injected into each mouse subcutaneously (n = 3 per group). The mice were killed on day 8 and matrigel plug was examined for haemoglobin content using the QuantiChrom hemoglobin assay kit as per the manufacturer’s protocol (BioAssay Systems).
An in vitro endothelial tube formation was done as described previously[67]. Briefly, matrigel was added (50 uL) to each well of a 96-well plate and allowed to polymerize. HUVECs were suspended in medium at a density of 3×105 cells/mL, and 0.1 mL of the cell suspension was added to each well coated with Matrigel. Cells were incubated for 12 hours at 37°C. The cells were then photographed, and branch points from 4 to 6 high-power fields (200x) were counted and averaged. The number of nodes (defined as when at least three cells formed a single point) per image was quantified.
For the chorioallantoic membrane (CAM) assay, white Leghorn chicken eggs (South China Agricultural University, Guangzhou, China) were incubated under routine conditions (constant humidity and 37°C) and a square window was opened in the egg shell at day 3 of incubation, to remove 3.5 mL of albumen and to detach the shell from the developing CAM. The window was sealed with a glass of the same size, and the eggs were returned to the incubator. Gelatin sponges were cut to a size of 1 mm3 and placed on the top of the CAM at day 8 under sterile conditions[68]. The sponges were then absorbed with 5 μL of low molecular weight heparin and cancer cells were implanted on the CAM surface to be tested. Sponges containing PBS were used as negative controls. CAMs were examined daily and photographed in ovo at day 12. The areas occupied by the vessel plexus were quantified using an IPP 5.0 image analysis program. The blood vessel density was expressed as the percentage of area occupied by the blood vessels of control over the whole area under the microscopic field[69].
Paraffin sections prepared from in vivo experiments were applied to IHC staining for the detection of protein expression levels of mTOR, S6K1, VEGF, HIF-1α, HK2, GLUT1, LDHA and VEGF. The indirect streptavidin-peroxidase method was used. All antibodies used for IHC are listed in S1 Table. The stained results were reviewed and scored by two pathologists independently. The intensity of immunostaining was scored as negative (0), weak (1), medium (2) and strong (3). The extent of staining, defined as the percent of positive staining cells, was scored as 1 (≤10%), 2 (11–50%), 3 (51–75%) and 4 (>75%). An overall expression score, ranging from 0 to 12, was obtained by multiplying the score of intensity and that of extent. The final staining score was presented as negative (overall score of (0), 1+ (overall score of 1–3), 2+ (overall score of 4) or 3+ (overall score of ≥5).
HEK293T cells (1 × 104) were cultured in 24-well plates and co-transfected with 20 nM EBV-miR-BART1-5P mimic or NC, 5 ng of pRL-CMV Renilla luciferase reporter and 30 ng of luciferase reporter that contained the wild-type or mutant 3’UTR of AMPKα1. For antagonism experiments, cells were also co-transfected with 20 nM anti-miR or anti-C(anti-control). Transfections were performed in duplicate and repeated in three independent experiments. Forty-eight hours after transfection, the luciferase activities were analyzed with a Dual-Luciferase Reporter Assay System (Promega, Madison, WI, USA).
Cells were cultured in DMEM without phenol red for 15 h, and the culture media was then harvested for measurement of lactate or glucose concentrations. Lactate levels were quantified using the Lactate Assay kit (BioVision, Mountain View, USA), glucose levels were determined by using a glucose assay kit (Sigma-Aldrich). All values were normalized to total protein levels (BCA Protein Assay Kit, Thermo Scientific, Waltham, USA). Recipient cells were labelled with 100uM 2-[N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl) amino]-2-deoxy-D-glucose (2-NBDG) (Sigma-Aldrich) diluted in glucose-free media and incubated for 40 min at 37°C. 2-NBDG levels were determined for measurement of fluorescence intensity by a confocal microscope (Olympus FV1000, Tokyo, Japan).
All experiments were performed in triplicate. Data shown are mean ± s.e.m. (unless otherwise specified) from at least three independent experiments. SPSS 19.0 software was used for statistical analyses. Differences were considered to be statistically significant at values of P<0.05 by Student’s t-test for two groups, one-way ANOVA (analysis of variance) analysis for multiple groups and parametric generalized linear model with random effects for tumor growth. Correlation was analyzed with two-tailed Spearman’s correlation analysis. Single, double and triple asterisks indicate a statistical significance of *P<0.05, **P<0.01 and ***P<0.001 respectively.
|
10.1371/journal.ppat.1006656 | Vpma phase variation is important for survival and persistence of Mycoplasma agalactiae in the immunocompetent host | Despite very small genomes, mycoplasmas retain large multigene families encoding variable antigens whose exact role in pathogenesis needs to be proven. To understand their in vivo significance, we used Mycoplasma agalactiae as a model exhibiting high-frequency variations of a family of immunodominant Vpma lipoproteins via Xer1-mediated site-specific recombinations. Phase-Locked Mutants (PLMs) expressing single stable Vpma products served as first breakthrough tools in mycoplasmology to study the role of such sophisticated antigenic variation systems. Comparing the general clinical features of sheep infected with a mixture of phase-invariable PLMs (PLMU and PLMY) and the wild type strain, it was earlier concluded that Vpma phase variation is not necessary for infection. Conversely, the current study demonstrates the in vivo indispensability of Vpma switching as inferred from the Vpma phenotypic and genotypic analyses of reisolates obtained during sheep infection and necropsy. PLMY and PLMU stably expressing VpmaY and VpmaU, respectively, for numerous in vitro generations, switched to new Vpma phenotypes inside the sheep. Molecular genetic analysis of selected ‘switchover’ clones confirmed xer1 disruption and revealed complex new rearrangements like chimeras, deletions and duplications in the vpma loci that were previously unknown in type strain PG2. Another novel finding is the differential infection potential of Vpma variants, as local infection sites demonstrated an almost complete dominance of PLMY over PLMU especially during early stages of both conjunctival and intramammary co-challenge infections, indicating a comparatively better in vivo fitness of VpmaY expressors. The data suggest that Vpma antigenic variation is imperative for survival and persistence inside the immunocompetent host, and although Xer1 is necessary for causing Vpma variation in vitro, it is not a virulence factor because alternative Xer1-independent mechanisms operate in vivo, likely under the selection pressure of the host-induced immune response. This singular study highlights exciting new aspects of mycoplasma antigenic variation systems, including the regulation of expression by host factors.
| Though implicated to play important roles in mycoplasma pathogenicity, the biological significance of large multigene families causing phase variation of immunodominant surface antigens has never been directly proven. Using M. agalactiae and its Xer1-mediated high-frequency variation system of Vpma surface lipoproteins as a model, we investigated the in vivo significance of this variable system by comparing the infection characteristics of two major expression variants, namely VpmaY and VpmaU. ‘Phase-Locked Mutants’ (PLMs) of these expression variants (PLMU and PLMY), served as ideal tools as they steadily express a single Vpma product without further switching. Interestingly, the PLMs altered their Vpma profiles during conjunctival and intramammary co-challenge experiments in sheep despite xer1 disruption using novel complex switches involving vpma chimeras, duplications and deletions. This illustrates that although the Xer1 recombinase is not a virulence factor for M. agalactiae, Vpma phase variation is important for its survival and persistence in the immunocompetent host. Furthermore, PLMU was outcompeted by PLMY, almost completely at local infection sites during the early stages of infection, indicating a better in vivo fitness and survival of VpmaY expressors. This demonstrated the differential infection potential of mycoplasma phase-variable lipoproteins using PLMs in infection studies of the natural host.
| Mycoplasmas are not only the smallest but also belong to the most successful bacterial pathogens that cause persistent and often difficult-to eradicate infections in humans and animals [1]. Although a number of mycoplasma diseases have a huge socio-economic significance, proper control strategies are missing, mainly due to lack of knowledge about their pathogenicity mechanisms. They lack typical pathogenicity factors found in other bacteria, and although genomes of many important mycoplasma pathogens have been sequenced, questions pertaining to their virulence and survival remain [1–3]. Having evolved 2.5 billion years ago from a Gram-positive ancestor, mycoplasma evolution was marked by severe genomic reduction, so that contemporary mycoplasmas have lost several metabolic pathways in addition to the genes dedicated to cell wall synthesis. Hence, most mycoplasma species display small genomes while retaining the functions necessary for survival in their respective hosts on which they depend for most of their nutrition [4]. However, many of them devote a sizeable part of their genomes to large multigene families encoding phase- and/or size-variable surface antigens [1, 5]. Although this highlights the biological significance of these multigene-encoded variable antigens in these minimalist organisms, the exact in vivo functions are rarely understood in the context of disease progression [6].
Although reversible high-frequency surface variations of bacteria are commonly believed to play a major role in host immune evasion [7, 8], experimental proof of such a role remains circumstantial for mycoplasmas [6]. To our knowledge there is only one in vivo study on M. pulmonis [9], and a couple of in vitro or indirect experiments that support the hypothesis that antigenic variation is critical for long-term survival of mycoplasmas, but they do not establish a causal link [5, 6]. In a recent in vivo analysis, Pflaum et al [10] have demonstrated similar phase variation events in M. gallisepticum to be nonstochastic and independent of host adaptive immunity. For most mycoplasma pathogens, the precise role of such multigene-encoded surface lipoproteins and their antigenic switches is yet to be understood, although for M. pulmonis the Vsa size variations are also shown to modulate adhesion, biofilm formation and protection against complement and phagocytosis [11]. Such studies are hindered due to the cumulative effect of many factors, such as the very high frequency of switching involving large gene families encoding several different clonal variants that are difficult to isolate or distinguish from each other, lack of appropriate animal models, and despite significant advances the continued recalcitrance of mycoplasmas to targeted gene disruptions [12]. Overcoming these challenges, we had earlier constructed M. agalactiae Phase-Locked Mutants (PLMs) by disrupting the xer1 gene that encodes the site-specific recombinase causing vpma gene inversions responsible for high-frequency switching of the Vpma surface lipoproteins [13, 14]. These PLMs served as breakthrough tools for evaluating the role of such multigene-encoded phase-variable mycoplasma proteins in pathogenesis, as they exhibited stable expression of one defined protein (the ‘locked’ Vpma phenotype) for several in vitro generations without further switching to alternate Vpma phenotypes [13]. Besides, M. agalactiae serves as an excellent model for studying mycoplasma pathogenesis because, unlike many other mycoplasma pathogens, (i) it grows well in the lab, (ii) the pathology associated with disease has been well-documented, (iii) many molecular tools, genomic and proteomic data are available, (iv) different sheep infection models are well-established, and most importantly, (v) it is phylogenetically very close to M. bovis [15], a worldwide serious pathogen that causes huge economic losses and induces similar clinical signs of mastitis, arthritis and pneumoniae in cattle [3, 16]. Until 1976, M. agalactiae and M. bovis were regarded as one species [17], and their genomes show synteny and inversions with many homologous genetic loci [18]. Especially striking is the similarity between the M. bovis phase-variable Vsp system and the M. agalactiae Vpma system, as both multigene families have the same conserved 5’ UTR and signal peptides, they contain the same lipoprotein cleavage motif (AAKC), bear repeated sequences, encode similar short cytadherence epitopes, and both undergo site-specific recombination for switching in expression [13, 19, 20]. Hence, Vpmas and Vsps might play similar roles during infection and disease progression, and any leads in our understanding of the role of Vpmas and their phase variation in pathogenesis could be potentially extrapolated to the Vsp system of M. bovis, and in general to all related mycoplasma phase-variable antigen systems.
Although implicated in host immune evasion, the biological significance of Vpma phase variation is yet to be proven. In a previous study we tried to address this by testing the wild-type M. agalactiae strain in comparison with the xer1-disrupted phase-invariable PLMs in two separate sheep infection models. A mixture of PLMY and PLMU, constitutively expressing VpmaY and VpmaU respectively, and a clonal population of the phase-variable type strain PG2 expressing all six Vpmas were compared for their infection traits in intramammary and conjunctival sheep infections for 28 and 20 days post infection (p.i.), respectively. The study was largely based on clinical data and quantitative mycoplasma loads at various host sites without analyzing their Vpma profiles. The data had demonstrated that although Xer1 recombinase is not a virulence factor of M. agalactiae, Vpma phase variation might critically influence its persistence during natural infections as suggested by the better dissemination and systemic responses of the phase-variable PG2 infection group [21].
The initial goal of the present study was to check if there were any differences in the colonization potential, in vivo fitness, tissue tropism and Vpma-specific humoral responses of Vpma proteins. For this, two major Vpma expression variants of type strain PG2, namely VpmaY and VpmaU, were selected for qualitative analysis of mycoplasma reisolates obtained after PLMY and PLMU co-challenge experiments in the natural sheep host, carried out consecutively via the conjunctival and intramammary infection routes, respectively. PLMY showed a marked predominance over PLMU in both infection models implicating a differential pathogenicity potential of Vpma variants. Furthermore, Vpma phenotypic and genotypic analysis of selected reisolates yielded very interesting results, most importantly in two aspects: (i) PLMs which were stable for several generations in vitro had ‘switched’ vpmas in vivo, and were now expressing new Vpma phenotypes via novel Xer1-independent mechanisms as xer1 was still disrupted, and yet (ii) these clones showed several complex gene rearrangements, including chimeras, deletions and duplications in the vpma locus that were never before observed in the PG2 type strain or any clones or PLMs derived from it. Overall, the results highlight the significance of Vpma phase variation for the pathogen’s survival in the immunocompetent host, so much so that although Xer1 is an essential factor responsible for Vpma switching in vitro, alternative molecular mechanisms operate in its absence in vivo, likely under the selection pressure of the host’s immune response. These are indeed interesting new features of surface antigenic variation systems in pathogenic mycoplasmas.
Absence of VpmaY and VpmaU phenotypes (corresponding to the PLMY and PLMU input population) in various milk samples obtained at Days 22–26 p.i. (Table 1) was an interesting result, and initially we thought that it could have two main implications: (i) either the PLMs had stopped their Vpma expression altogether, or, (ii) contrary to their so far known in vitro stability and phase-invariable phenotypes, they had now switched to express other Vpmas in vivo, even in the absence of Xer1 recombinase, the molecular switch that was earlier shown to be necessary for Vpma phase variations [13]. To check this hypothesis, milk and other necropsied tissue samples that showed neither VpmaU nor VpmaY expression (S2 Fig) were tested for Vpma phenotypes other than VpmaY and VpmaU. Results confirmed that the PLMs had indeed changed their Vpma profiles in vivo to express one of the other four Vpmas that were initially not expected, namely VpmaZ, VpmaW, VpmaV and VpmaX (Fig 2, Tables 1 and 2, S1 and S3 Figs). For instance, S3 Fig is a representative figure that shows the expression of the latter three Vpmas by such ‘switchover’ clones in the right udder of MS 8. Similarly, S1 Fig shows the ‘switchover’ expression of VpmaW in the mesenterial LN of MS 6. However, when picked and filter cloned, the majority of the ‘switchover’ clones did not undergo further switching like the phase-variable WT strain and behaved as new PLMs with alternate Vpma phenotypes as shown in Fig 4. They seemed to be ‘locked’ for the expression of only one Vpma and gave positive colonies (without sectoring) only with the corresponding Vpma-specific pAb, and were negative with all the other five Vpma-specific pAbs (Fig 4). Hence, PLMY and PLMU, which were stable for several in vitro generations encountered selection pressure in the immunocompetent host to switch to different Vpma phenotypes allowing us to select new PLMs, namely PLMW (from mesenterial LN of sheep MS 6), PLMZ (from right parotideal LN of sheep S 7), PLMV (from right udder of sheep MS 10) and PLMX (from right udder of sheep MS 9) (Figs 2 and 4, Tables 1 and 2, S1 and S3 Figs).
Two mechanisms were considered to explain the unexpected altered Vpma phenotypes in many of the mycoplasma reisolates from the PLM-inoculated sheep: reversion of the xer1 mutation or occurrence of vpma gene rearrangements in absence of Xer1. Disruption of the xer1 gene was confirmed in the ‘switchover’ clones by PCR (Fig 5A) and Southern hybridization (Fig 5B) demonstrating the presence of the pR3/tetM disruption plasmid within xer1 gene as described earlier for PLM construction [13]. Briefly, unlike the WT PG2 strain that displays a native 13 kb fragment in Southern analysis (Fig 5B, lane 1), pR3 integration at the chromosomal xer1 locus results in duplication of the partial xer1 sequence that segregates onto two ClaI fragments, one carrying the plasmid oriC region and part of the C-terminal coding region of xer1, and the other bearing the bla and tetM plasmid sequences together with the N-terminal region of xer1, followed by the vpma genes (Fig 5B, lanes 3–8).
As VpmaW and VpmaX were the most predominant new alternative Vpma phenotypes selected in vivo via the generation of PLMW and PLMX clones, we decided to clone and sequence the whole vpma locus of two selected representative clones, namely PLM16 (PLMW) and PLM18 (PLMX) that constitutively express VpmaW and VpmaX, respectively (Fig 4). PLM16 was picked and filter cloned from the mesenterial LN of sheep MS 6 reisolates showing 100% VpmaW expression and the complete absence of the other five Vpma phenotypes (S1 Fig). PLM18, on the other hand, was picked from the predominant VpmaX expressors found in the right udder of MS 9.
The vpma configuration of PLM18 was found to be similar to PLMY (Fig 6A) except that two hybrid genes, vpmaYX’ and vpmaXY’ are present instead of vpmaX and vpmaY genes of the type strain PG2. Located downstream to the single identified promoter, vpmaXY’ constitutes the expressed gene in PLM18 (Fig 6B). Comparison of the coding sequence of the hybrid vpmaXY’ gene with the original vpma sequences of the PG2 clonal variant 55–5 [24] and PLMY [13] (Fig 6A) indicates that an intergenic recombination event has occurred at a 38-bp sequence (Fig 6C, bold letters) that is common to both vpmaX and vpmaY gene sequences. In the chimeric gene vpmaXY’, the sequence upstream of this 38-bp homologous region was found to be 100% identical with the vpmaY gene, whereas the downstream region showed complete identity with the sequence vpmaX (Fig 6C). Recombination at this 38-bp homologous sequence not only resulted in the generation of two chimeric genes, but also led to the inversion of the single promoter resulting in the alteration of the observed Vpma phenotype depending on the specific Vpma antibody epitopes present in the downstream region (Fig 6B).
In contrast to PLM18, examination of the PLM16 vpma locus revealed a much more complex recombination scenario. Phenotypic expression of VpmaW in this clone correlates well with the sequence analysis that revealed the presence of vpmaW gene located downstream of the promoter (Fig 7). Sequencing data also revealed the duplication of two vpma genes, namely vpmaX and vpmaW, as also supported by Southern blot analysis whereby two bands were observed with both vpmaW and vpmaX probes when the genomic DNA was digested with restriction enzymes that cut outside these genes, respectively (Fig 8A and 8C). Furthermore, the sequence of PLM16 vpma locus also showed the complete absence of vpmaU, vpmaY and vpmaZ genes (Fig 7). In accordance, no hybridization signal was observed during Southern blot analysis with probes specific for vpmaU, vpmaY and vpmaZ genes (Fig 8B) clearly verifying the absence of these three genes. In comparison, PLM18 demonstrated the expected bands with the specific vpma gene probes, for instance, 2.8 kb HindIII or HindIII/XbaI fragments corresponding to vpmaU, vpmaW and vpmaZ genes, respectively (Fig 8B and 8C). A 1.7 kb HindIII fragment corresponds to the vpmaX-derived sequence of the hybrid vpmaYX’ gene in PLM18, whereas the presence of two HindIII/XbaI fragments (0.8 kb and 1.9 kb) indicates duplication of vpmaX in PLM16 (Fig 8C). Similarly, PstI-digested genomic DNA probed with a vpmaY-specific probe detects 0.6 kb and a 1.7 kb fragments corresponding to the vpmaY-derived sequence of the hybrid vpmaYX’ gene in PLM18 and both these fragments are absent in PLM16 (Fig 8B).
Although, it is difficult to precisely reconstruct the hierarchy of genomic rearrangements that occurred in PLM16, the final vpma configuration of this clone (Fig 7) is likely a result of recombination events that occurred in PLMY. Potential duplication of vpmaX and vpmaW placed the vpmaW gene downstream of the promoter, whose position is unaltered compared to PLMY, and simultaneously deleted the genomic fragment carrying vpmaU, vpmaY and vpmaZ (Figs 7 and 8).
Pathogenicity is a complex summation of many biological processes that allow the pathogen to survive, multiply and persist in an immunocompetent host. This is especially true for mycoplasmas that usually cause highly persistent infections without being recognized by host immune factors and are thus considered useful models to investigate transitions from a parasitic to endosymbiotic life style [25]. While their pathogenicity mechanisms are largely unknown, persistence is often attributed to the presence of large multigene families causing high-frequency phase variation of surface lipoproteins [1, 4, 5]. This study provides the first experimental in-host demonstration of their significance and uncovers new facets of mycoplasma antigenic variation systems.
The biological processes relevant to pathogenicity are often regulated by a complex spectrum of different host and microbial molecules, stressing the need for more relevant in vivo studies, without which, the in vitro models would be largely incomplete, and sometimes even misleading [26]. This further highlights the relevance of our study as it was carried out in the natural host using two different natural routes of infection.
M. agalactiae PLMs constitutively and stably express the same defined Vpma protein in vitro and are unable to undergo further switching as their xer1 gene responsible for Vpma phase variation is disrupted [13]. Yet, when tested in intramammary and conjunctival sheep infection models, PLMY and PLMU, initially expressing just VpmaY and VpmaU, respectively, were able to express new Vpmas after in-host passage. (Tables 1 and 2; Fig 2; S1 and S3 Figs). The in-host selection of these ‘switchovers’, previously undetected in vitro, indicate that these display a phenotypic advantage. Furthermore, these ‘switchovers’ behaved like new PLMs and showed no further phase variations for the rest of the experiment in vitro (Fig 4). This finding demonstrated the occurrence of an alternative Xer1-independent mechanism of Vpma switching and strongly emphasizes the significance of Vpma antigenic variation for M. agalactiae. Similar alternative in vivo switching of surface layer proteins (SLP) has also been reported for a recA-disrupted PLM of Campylobacter fetus during experimental sheep infections [27]. However, unlike our study, in this case the two PLMs used for infection predominantly did not switch and just one of the ovine–passaged isolates switched to the other mutant SLP phenotype without creating new SLP variants [27, 28].
In-host change of Vpma expression by PLMs was a result of complex vpma gene rearrangements that differed from conventional Xer1-mediated site-specific vpma recombinations. In one of these switchover clones, three of the six vpma genes were completely missing and two others were duplicated (Figs 7 and 8). In another clone (Fig 6), intergenic vpma recombination at a 38-bp homologous region generated new chimeric vpma genes that resulted in changed Vpma phenotype of the PLM. This is remarkable in many aspects. Firstly, the likelihood of a recombination occurring at this short homologous sequence is rare and emphasizes the strong in vivo selection pressure faced by the PLMs to change their Vpma phenotype. This contrasts the in vitro scenario where PLMs never changed Vpma expression [13], and generally speaking, targeted gene manipulations via homologous recombination are still a challenge for mycoplasmologists [12]. The presence of several intergenic and intragenic repeat sequences in the vpma locus offers an enormous potential for generating many new hybrid Vpma antigens via recombination in vivo. Repeated regions are also found in the phase variable loci of other mycoplasmas and the generation of hybrid genes by homologous recombination likely represents a common mechanism for additional antigenic diversification for immune evasion in these species. For instance, similar intrachromosomal recombination events in the vsp gene locus of M. bovis led to the generation of a new chimeric lipoprotein [29]. Vpma hybrids have been reported in strain 5632 of M. agalactiae, which incidentally contains an extended repertoire of 23 vpma genes distributed on two separate vpma loci. However, the occurrence of a second vpma cluster at a separate locus is a rare event, so far observed only in strain 5632 and found absent in all the 92 different M. agalactiae strains tested [30]. Yet our study provides the first demonstration that such variations are a result of in-host selective pressure, in the absence of which the pathogen maintains its basic vpma genotypic switching via site specific recombination [13, 14, 20]. A possible explanation for this could be the better fitness of these classical “simple variants” in the absence of immune responses as demonstrated for Msp2 antigenic variants of Anaplasma marginale [8].
Among various strategies used by pathogens to persist in the immunocompetent hosts, evasion of host defence reactions by antigenic variation is quite significant [8, 31, 32]. However, in mycoplasmas, which are known for their sophisticated phase-variable surface protein families, the function of these proteins in immune evasion in the host has not been unequivocally demonstrated [5, 6].
For M. agalactiae, this issue was unexplored until now although Vpma immunogenicity and in vivo phase variations have been demonstrated [23]. Since the Vpma switching rates are very high [13] it was not possible to reliably assess the role of Vpma phase variation in host colonization and immune evasion during in vivo studies with the wild type PG2 strain. The PLMs offered this opportunity, and when used in the sheep infection model they exhibited Xer1-independent complex ‘switches’ that were observed only in the immunocompetent host and never during several in vitro passages. From this we envisaged that these Vpma switches are likely involved in host immune evasion. In Anaplasma marginale, “complex variants” of Msp2 are favoured only under selective pressure of adaptive immune response as they generally have a reduced fitness compared to the “simple variants” [8]. Similar competing selection pressures for immune evasion and variant fitness might also apply to the Vpma variants obtained by conventional Xer1-mediated and complex Xer1-independent recombination. Furthermore, it has been shown in Campylobacter fetus that host immune responses against the antigenic SLPs are delayed in sheep infected with strains capable of varying their SLPs, as compared to those infected with SLP phase-invariable mutants [22]. Although mycoplasma lipoproteins are well-known preferential targets of the humoral immune response, yet the influence of host antibodies on mycoplasma phase-variable lipoproteins, including Vpmas, has never been directly investigated in vivo [5].
In the current study, VpmaU- and VpmaY-specific antibody responses in milk and sera of the PLM-infected sheep (Fig 3) correlate strongly with the presence of specific Vpma expressors at different phases of infection, and with the appearance of ‘switchover’ clones in PLMs (Tables 1 and 2). This suggests that Vpma switching may be a key factor of M. agalactiae persistence.
VpmaY expressors (PLMY) clearly exhibited a better fitness in vivo compared to VpmaU expressors (PLMU) (Tables 1, 2, 3 and 4; Fig 3). This is a very interesting result which could have a great impact on our understanding of Vpma-host interactions. Growth in vitro is very different than growth in vivo, and the latter is a primary requisite for pathogenesis [33]. As PLMU did not show any in vitro growth retardation, this is an exclusive in vivo characteristic likely dependent on one or more host factor interactions. Lipoproteins of bacterial pathogens are known to be involved in host adaptation and immune evasion, and can act as cytadhesins, invasins, potent immunogens and immune modulators by interacting with the innate and adaptive host immune responses [6, 34]. Although unknown at this point, VpmaY and VpmaU lipoproteins could interact and likely differ with each other at any of these levels. It is possible that VpmaY being longer in size has a better ‘shielding’ capacity against phagocytic cells as observed for Vsa lipoproteins of M. pulmonis [11], or perhaps it enhances adherence of the organism or even the retrieval of nutrients for growth and multiplication resulting in a better survival or in vivo growth rate.
M. agalactiae infections initiate a dynamic innate cellular response, followed by a chronic adaptive response [35, 36]. Mycoplasma lipoproteins being potent macrophage stimulants are often implicated in the initiation of a characteristic immunopathology [6]. Hence, considering the clearance of PLMU right at Day 1 p.i., it is likely that the VpmaU lipoprotein either hindered initial colonization, perhaps due to lower cytadhesion ability compared to VpmaY, or it made the mycoplasma cells more vulnerable to early immune responses. However, all this needs further investigations, which are currently underway.
Furthermore, the in vivo selection and enrichment of VpmaW and VpmaX expressors is also noteworthy as these were the only Vpma phenotypes detected exclusively in some tissues (Fig 2, Table 2, S1 Fig), which might be indicative of some preference or tendency for tissue tropism though not detected in all sheep and would require further evaluations. A similar in vivo enrichment of specific VlhA lipoprotein isotypes of M. gallisepticum was observed in multiple independent hosts during experimental chicken infections [10]. This in-host expression and enrichment of specific Vpma phenotypes is indicative of their importance in host-pathogen interactions. Interestingly, VpmaX and VpmaW were also observed to be most immunogenic amongst the Vpma proteins when we tested the serum from a convalescent naturally infected ewe (PAL 97) [37] against the corresponding six MBP-Vpma fusion proteins [13] (S4 Fig).
Requirements for virulence factors are known to change with different infection models and also as the infection proceeds. It has been reported that in some pathogens a particular phase variant, or even phase variation per se, is required only at a particular infection stage and not throughout the entire infection process [38, 39]. This may as well apply to Vpmas and their phase variation during sheep infections. Although PLMU was cleared in the beginning, PLMY could successfully establish local infection even in the absence of Vpma phase variation. But, at later stages of infection, likely to cope with host defence and to persist, or to become systemic, it was necessary for the organism to vary the Vpma phase, and hence the selection pressure that induced Xer1-independent alternate ‘switches’ in PLMs. This is supported by an earlier study where PG2 type strain induced better systemic responses and also disseminated better in infected sheep compared to PLMs [21], likely by delaying the host anti-Vpma antibody response by varying the Vpma antigens as observed in case of SLP antigenic variation of Campylobacter fetus [22]. Also, considering that Vpmas have adherence epitopes [20], their differential adherence rates might also govern the selection of different Vpma variants at specific stages of local or systemic infections as observed in other pathogens [40].
Our results strongly emphasize the in-host significance of Vpma rearrangements for M. agalactiae. Although Xer1 recombinase is the sole factor responsible for Vpma switching in vitro in the absence of any discriminatory constraints, alternative molecular switches become prominent in its absence under selective pressure as observed inside the immunocompetent sheep host mounting Vpma-specific humoral responses. Furthermore, these host responses are instrumental in selecting complex novel antigenic variants, a causal relationship though known in other pathogens [8] was so far not shown for M. agalactiae. This is the first report where mycoplasma PLMs have been used for infection studies in the natural host pointing towards the differential pathogenicity of such phase variable lipoproteins and providing new insights into the mechanisms of their antigenic variation inside an immunocompetent host during disease progression. This will likely impact the understanding of the in vivo role of surface antigenic variation systems and their regulation in other pathogenic mycoplasmas. Enhancing our understanding of the fitness costs of specific expression variants might also make it feasible to develop new vaccines that prevent the period of high infectivity and disease by targeting highly fit clonal variants [32].
Amongst the six Vpma lipoproteins expressed by the M. agalactiae type strain PG2, VpmaY and VpmaU are not only the most abundant variants observed in vitro, but also, based on N-terminal and other repeat sequences, belong to the two separate homology groups identified by Glew et al [20]. Hence, PLMY and PLMU, constitutively expressing VpmaY and VpmaU respectively [13] were selected as representatives for sheep infection trials to check their comparative in vivo behavior. M. agalactiae pathogenic type strain PG2 [41] was used for inoculating the positive control group. The type strain and PLMs were grown in Aluotto broth [42] and processed for inoculum preparation as described earlier in detail [21]. Unlike the phase-invariable PLMU and PLMY, the PG2 population expressed all six Vpmas with high-frequency phase variations [13]. The Vpma phenotypes were reconfirmed by plating and re-analysing the residual pooled inocula using Vpma-specific antibodies, whereby the PLM inoculum was confirmed to contain only VpmaY and VpmaU expressors and the other four Vpma phenotypes were not detected.
Details of the experimental set-up of the sheep infection trials have been reported earlier [21] and the main points enlisted under Table 5. Briefly, for both the intramammary and the conjunctival infection routes, 5 sheep were used for each of the three infection groups (lambs, denoted by the initial ‘S’, for conjunctival infections, and lactating ewes, denoted by ‘MS’, for intramammray infections): (i) the experimental PLM group infected with an equal concentration mixture of PLMU and PLMY (5 x 108 cfu, each), (ii) a positive control group infected with 109 cfu of the PG2 strain, and (iii) a negative control group inoculated with pyrogen-free saline (see Table 5). Animal samples, such as blood, milk and ocular and nasal swabs were collected regularly as described earlier [21], always proceeding from the negative control group to the PLM-infected sheep, followed by the positive control group at the end. The samples were processed further for storage at -80°C or for immediate analysis. Sheep were euthanized and necropsied at Day 20 p.i. (conjunctival infection) or Day 28 p.i. (intramammary infection). Samples from organs (spleen, lungs, kidneys and udders) and LNs (such as mandibular, mediastinal, mesenterial, medial and lateral retropharyngeal, parotideal, iliac, supramammary, etc) were cut-up into small pieces and flash-frozen in liquid nitrogen before storing at -80°C.
The sheep conjunctival and intramammary route infections were performed with the approval of the Austrian Federal Ministry for Education, Science and Culture (BMBWK-68.205/0145-BrGT/2006) and the Austrian Federal Ministry for Science and Research (GZ/BMWF-68.205/0092-C/GT/2007), respectively. All procedures related to the sheep experiment were carried out according to the Animal Experiments Act (TVG, BGBI.Nr. 501/1989, last modified by BGBI. I Nr. 162/2005). The animals were housed in the stables at the University of Veterinary Medicine Vienna and experiments executed after approval by the Ethics and Animal Welfare Commission of the University of Veterinary Medicine Vienna. The sheep were anesthetized by Thiopental before euthanizing them via intravenous injection of T61 as recommended and approved for sheep according to the drug directory of the Austria Codex.
Original or diluted frozen stocks corresponding to the samples found positive for M. agalactiae in a previous report [21] were subjected to analysis for Vpma phenotyping. Undiluted or appropriate serial dilutions of the stocks were directly plated on Aluotto or SP4 agar plates and incubated at 37°C for a minimum of 7 days. The frozen tissues were also re-analyzed by cutting-off small pieces, finely chopping and resuspending in 3–5 ml Aluotto broth, which was incubated at 37°C for 3 h before plating. Agar plates with an appropriate number of well-isolated colonies were selected for Vpma phenotyping via colony immunoblot analysis.
Colony immunoblotting was performed essentially the same way as described earlier [13]. Briefly, the colonies were lifted on Protran BA 83 nitrocellulose membranes (Schleicher & Schuell), allowed to dry at room temperature (RT) and rinsed 3 times in TBS buffer (10 mM Tris, 154 mM NaCl, pH 7.4) before incubating overnight at 4°C in 1: 2000 dilution of VpmaY-specific α-Y or 1: 600 dilution of VpmaU-specific α-U rabbit polyclonal antisera [13]. Membranes were washed three times with TBS buffer containing 0.05% Tween 20 (Roth) for 10 minutes each with shaking. Subsequent incubation in swine anti-rabbit IgG conjugated to horseradish peroxidase (Dako) was carried out at least for 1 h at RT at 1: 2000 dilution. The membranes were then washed three times for 10 min each in TBS buffer before developing them in 4-chloro-1-naphthol (Bio-Rad) and hydrogen peroxide for 15–30 min. Finally, the blots were rinsed in water, dried at RT and viewed under a Nikon SMZ-U stereomicroscope to count the number of positive colonies among the total colonies present on the blot. If needed, the negative colonies were counterstained in pink color using reversible Ponceau S (Roth) staining to calculate the percentage of PLMU or PLMY clones in the total mycoplasma load of the specific sample.
After the initial analysis with α-Y or α-U Abs, the expression of the other four Vpmas, namely VpmaW, VpmaX, VpmaZ and VpmaV was also checked in the samples via similar colony immunoblot analysis using α-W, α-X, α-Z and α-V rabbit polyclonal Abs [13], respectively.
Two different -80°C stocks containing the PLMU and PLMY mixture for sheep inoculations were thawed. The average mycoplasma count had been earlier calculated (by thawing and plating of two other independent aliquots) to be 3.4 x 1010 cfu/ml. The thawed stocks were gently vortexed before removing 70 μl aliquots to inoculate 25 ml Aluotto broth containing penicillin and phenol red as described earlier [43]. Dilutions (10−4 to 10−6) of the duplicate cultures were plated on SP4 and Aluotto agar at time T0 of the growth assay before incubation at 37°C. Further 100 μl samples were removed after 3, 5, 23, 31, 48 and 70 h of incubation to prepare duplicate serial dilutions (10−4 to 10−7) for plating 100–200 μl on Aluotto and SP4 agar plates in duplicate. The plates were incubated at 37°C for 5–7 days. Colonies were counted under the Nikon SMZ-U stereomicroscope to calculate the final cfu/ml, and for each time point, plates with an appropriate number of well-isolated colonies were used for colony immunoblot analysis. Each of the duplicate colony blots was cut into two equal halves, one half was immunoblotted with an α-U Ab specific for VpmaU and the other half with α-Y antiserum specific for VpmaY. In each case, positive colonies were counted under the stereomicroscope before counterstaining the negative colonies with Ponceau S to calculate the proportion of PLMU and PLMY in the total population. Absence of the other four Vpmas in this PLMU and PLMY mixed culture was confirmed by colony immunoblot analyses with α-W, α-Z, α-V and α-X antisera, all of which gave negative results.
A Mini-PROTEAN II multiscreen apparatus (Bio-Rad) was used to screen the individual sheep sera and milk samples on Western blots. For antigen preparation, strain PG2, PLMY and PLMU were grown in 50 ml Aluotto broth for 3–4 days at 37°C before centrifugation at 10,000 g for 15 minutes. The cell pellets were resuspended in 1–2 ml PBS, and the protein concentrations were determined using the Pierce BCA Protein assay kit (Thermo Fisher Scientific). About 300–400 μl of each of the cell suspensions was separately loaded onto 12% polyacrylamide gels containing 3% (w/v) urea for running standard SDS-PAGE mini gels which were blotted onto Protran nitrocellulose membranes as described earlier [13]. The blots were blocked with 3% (w/v) skimmed milk and rinsed in TBS before applying to the multiscreen apparatus. Milk and sera samples were diluted 1:100 and 600 μl of each was used per channel of the multiscreen apparatus. After overnight incubation at 4°C, the blots were processed using the same protocol as described for colony immunoblotting except that polyclonal rabbit anti-sheep immunoglobulins (Dako) were used as secondary antibodies at a dilution of 1: 2000.
Many of the biological samples obtained from PLM (PLMU and PLMY)-infected sheep revealed colonies expressing Vpma phenotypes other than VpmaU and VpmaY. Colony immunoblots with α-V, α-X, α-Z and α-W antibodies revealing predominantly positive colonies were used to pick the corresponding clones (expressing VpmaV, VpmaX, VpmaZ and VpmaW, respectively) from the agar plates under the microscope, transferred into 1 ml SP4 broth containing tetracycline (2 μg/ml) and grown for 5–7 days at 37°C. A small aliquot was then used for making serial dilutions to be plated for colony immunoblot analysis with Vpma-specific antibodies, and the rest was used to extract crude DNA,which was tested for xer1 disruption by PCR using primer RecEndET28, corresponding to the chromosomal xer1 gene, and primer T3ISLrev, corresponding to the plasmid backbone of the disruptant plasmid as described earlier [13]. Once the xer1 disruption was confirmed with the presence of a 2 kb band, additional colony immunoblots were analysed with the six Vpma-specific antibodies. As the picked colonies were positive only with a single Vpma-specific antibody (and negative for all the other five Vpmas) without any sectoring phenotype, we concluded that these ‘switchover’ Vpma expressors were new PLMs, now expressing Vpma phenotypes other than VpmaU and VpmaY. For each of these new PLMs, namely PLMV, PLMX, PLMZ and PLMW, filter cloning was performed using three rounds of plating, colony immunoblotting and isolating single colonies. Briefly, cells from a colony (corresponding to a positive colony immunoblot phenotype as visualized by a specific Vpma Ab) were picked using a sterile micropipette, transferred into 1 ml SP4 medium and vortexed. The cell suspension was then sucked into a syringe with a 0.9 mm needle and filtered successively through 0.45 μm and 0.2 μm disposable membrane filters. Dilutions (10−1 to 10−3) were made from both filtrates, and 200 μl of each dilution was plated on SP4 agar plates containing tetracycline (2 μg/ml). The plated mycoplasma cells were incubated at 37°C until colonies were visible, and another round of colony immunoblotting was made to again pick single well-isolated positive colonies from corresponding agar plates. This procedure was repeated once again, and at the end, the new PLMs were thoroughly characterized with all six Vpma-specific Abs and also checked for the xer1 gene disruption as described above. Southern blot analyses for verification of xer1 disruption and vpma gene configuration are described ahead under separate section headings.
Mycoplasma genomic DNA was isolated by QIAamp DNA Mini Kit (Qiagen) and digested with ClaI (New England Biolabs) at 37°C for at least 6–7 h before subjecting to electrophoresis on a 1% agarose gel. This was followed by standard Southern blotting procedures and hybridization using Digoxigenin (DIG)-labelling system (Roche) as described previously [43]. A DIG-labelled probe corresponding to a 513 bp partial xer1 fragment was prepared by PCR amplification of genomic DNA of type strain PG2 using primers XerR and XerS in presence of 2.5 mM MgCl2 using 30 cycles of 95°C for 43 s, 56°C for 43 s and 72°C for 43 s. After purification with the QIAquick PCR Purification Kit (Qiagen), hybridization and subsequent chemiluminescent detection of the DIG-labelled nucleic acids using Anti-DIG -AP was carried out according to the manufacturer’s instructions (Roche). As described earlier [13], xer1 disruption corresponded with the presence of two disruption bands of about 3.4 kb and 18.9 kb, whereas the wild type PG2 strain yielded a 13 kb band under these conditions.
In order to define the vpma configuration of selected ‘switchover’ clones, such as PLM16 and PLM18, their ClaI-digested genomic DNA was self-ligated and transformed into E. coli DH10B. Tetracycline- and ampicillin-resistant transformants were selected and recombinant plasmids, namely pPLM16 and pPLM18, were isolated from them using the EZNA Plasmid Miniprep Kit DNA (Peqlab). The obtained plasmid DNA was used for sequencing the vpma gene loci present in these clones using the primer walking approach. Sequencing and synthesis of all oligonucleotides used in this study (see S1 Table) was carried out at LGC Genomics GmbH, Berlin, Germany. Sequences were analyzed by advanced BLASTX searches made at the website for the National Centre for Biotechnology Information: https://blast.ncbi.nlm.nih.gov/Blast.cgi.
DIG-labelling (Roche) of vpma-specific gene probes by PCR was carried out according to manufacturer’s instructions using PG2 genomic DNA as template and a set of following primer pairs (oligonucleotide sequences are enlisted in S1 Table) for individual vpma-specific gene probes: U2F and Urev1 (vpmaU), WDIGfw and WDIGrv (vpmaW), X1F and X1R (vpmaX), Y3F and Y3R (vpmaY), Z1F and Z2R (vpmaZ). PCR cycling conditions were as follows: 1 cycle of initial denaturation for 3 min at 94°C, 30 cycles of 95°C for 1 min, 57°C (vpmaU, vpmaX, vpmaY, vpmaZ) or 65°C (vpmaW) for 1 min, followed by 30s (vpmaU, vpmaX, vpmaY, vpmaZ) or 1 min (vpmaW) at 72°C and a final extension step for 5 min at 72°C.
Mycoplasma genomic DNA was isolated as described previously [43]. For vpma-specific Southern blots, genomic DNA was digested with appropriate restriction endonucleases: HindIII for hybridization with vpmaW-specific probe, HindIII and XbaI for vpmaU-, vpmaX- and vpmaZ-specific probes and PstI for vpmaY-specific probe. Digested DNA was subjected to agarose gel electrophoresis and DNA fragments were transferred to nylon membranes (Roth) using standard procedures described earlier [13]. Hybridization with DIG-labelled probes and washing under stringent conditions followed by non-radioactive detection was carried out according to the manufacturer’s recommendations (Roche).
|
10.1371/journal.pntd.0002897 | Scabies Mites Alter the Skin Microbiome and Promote Growth of Opportunistic Pathogens in a Porcine Model | The resident skin microbiota plays an important role in restricting pathogenic bacteria, thereby protecting the host. Scabies mites (Sarcoptes scabiei) are thought to promote bacterial infections by breaching the skin barrier and excreting molecules that inhibit host innate immune responses. Epidemiological studies in humans confirm increased incidence of impetigo, generally caused by Staphylococcus aureus and Streptococcus pyogenes, secondary to the epidermal infestation with the parasitic mite. It is therefore possible that mite infestation could alter the healthy skin microbiota making way for the opportunistic pathogens. A longitudinal study to test this hypothesis in humans is near impossible due to ethical reasons. In a porcine model we generated scabies infestations closely resembling the disease manifestation in humans and investigated the scabies associated changes in the skin microbiota over the course of a mite infestation.
In a 21 week trial, skin scrapings were collected from pigs infected with S. scabies var. suis and scabies-free control animals. A total of 96 skin scrapings were collected before, during infection and after acaricide treatment, and analyzed by bacterial 16S rDNA tag-encoded FLX-titanium amplicon pyrosequencing. We found significant changes in the epidermal microbiota, in particular a dramatic increase in Staphylococcus correlating with the onset of mite infestation in animals challenged with scabies mites. This increase persisted beyond treatment from mite infection and healing of skin. Furthermore, the staphylococci population shifted from the commensal S. hominis on the healthy skin prior to scabies mite challenge to S. chromogenes, which is increasingly recognized as being pathogenic, coinciding with scabies infection in pigs. In contrast, all animals in the scabies-free cohort remained relatively free of Staphylococcus throughout the trial.
This is the first experimental in vivo evidence supporting previous assumptions that establishment of pathogens follow scabies infection. Our findings provide an explanation for a biologically important aspect of the disease pathogenesis. The methods developed from this pig trial will serve as a guide to analyze human clinical samples. Studies building on this will offer implications for development of novel intervention strategies against the mites and the secondary infections.
| Scabies is a neglected, contagious skin disease caused by a parasitic mite Sarcoptes scabiei. It is highly prevalent world-wide, and now recognized as a possible underlying factor for secondary bacterial infections with potential serious downstream complications. There is currently few experimental data demonstrating directly that mite infestation promotes bacterial infections. Due to remarkable similarities in terms of immunology, physiology and skin anatomy between pigs and humans, we developed a sustainable porcine model enabling in vivo studies of scabies mite infestations. Here, we investigated the impact of the scabies mite infection on the normal pig skin microbiota in the inner ear pinnae in young piglets. Samples obtained prior to, during infection and after acaricide treatment were analyzed by sequencing of bacterial 16S rDNA. We report that scabies infestation has an impact on the host's skin microbiota. Staphylococcus abundance increased with the onset of infection and remained beyond treatment and healing. A shift from commensal to pathogenic Staphylococci was observed. This study supports the link between scabies and Staphylococcus infections, as seen in humans. It is the first in vivo demonstration of a mite induced shift in the skin microbiota, providing a basis for a similar study in humans.
| Scabies is a skin disease caused by the parasitic mite Sarcoptes scabiei variety hominis in humans. It is common worldwide, predominantly affecting overcrowded, socio-economically disadvantaged populations [1]. Scabies is ubiquitous and a significant public health burden in the developing world with prevalence of up to 70% in rural India and between 18 and 42% in the South Pacific, and erratic reports from Africa and South America [2], [3]. Recently scabies outbreaks are also reported regularly in economically rich regions, particularly from institutional settings such as health care facilities, elderly homes, prisons and child care centers [4], [5], where it is a well recognized and serious problem and control is notoriously difficult. Some studies suggest that the general population in economically stable societies has recently become more affected [6]. Moreover, scabies is a significant public health burden in the Indigenous population of tropical northern Australia [7]–[9]. Crusted scabies, a highly contagious manifestation of the disease presenting with extreme parasite numbers, can be seen predominantly in elderly individuals or in immunocompromised patients, especially those with infections due to HIV and human T lymphotropic virus 1, or drug-induced immunosuppression [10].
Epidemiological studies indicated a link between the epidermal infestation with S. scabiei and cutaneous bacterial infections (pyoderma, impetigo), particularly in tropical settings [9], [11]–[14]. Secondary bacterial skin infections commonly associated with scabies infestations are primarily caused by two clinically important pathogens, i.e. Streptococcus pyogenes and Staphylococcus aureus, including methicillin-resistant and methicillin-sensitive strains [8], [15]. These bacterial pathogens potentially cause life-threatening invasive infections, if left untreated. One obvious factor for the close association between mite infestation and bacterial disease is the breach of the physical barrier, i.e. the epidermal layers of the intact skin, through mechanical infringement by the burrowing scabies mites. This creates a suitable niche for pathogens to establish in the mite burrows. Our recent studies showed that scabies mites interfere locally with human complement mediated protection [16], [17], thereby promoting growth of S. pyogenes [18] and S. aureus (Swe et al., manuscript in preparation). However, the tripartite interactions between host, mites and bacteria are largely unexplored. In particular it is unknown whether scabies mites facilitate the transmission of pathogens or rely on obligatory endosymbionts for survival. Also, scabies-induced changes in the skin microbiome could provide a surrogate diagnostic biomarker for ordinary scabies infections, which are notoriously difficult to diagnose by conventional methods. Investigations into the identification of all bacteria associated with scabies mite infestation are important in developing novel control strategies as well as in improving current management and prevention policies.
Obtaining clinical samples from human scabies patients is generally a logistical challenge, as scabies is not a notifiable disease, it is difficult to diagnose and outbreaks are sporadic. Ordinary scabies patients harbor only few mites, which complicates targeted sampling. In humans, scabies manifests at multiple skin sites providing multiple, highly diverse habitats for bacteria [19]. Additional elements such as climate, hygienic procedures, age and genetics potentially increase the variation of mite associated microbiota further. Finally, a study in human patients over the course of an infection, in line with ethical limitations, would be very difficult. Therefore a longitudinal pilot study of a defined infection site in a controlled porcine animal model was undertaken. Pigs are natural hosts for S. scabiei var. suis, developing clinical manifestations closely resembling human scabies [20]. The integumentary system, the innate immunity and many biochemical parameters are incredibly similar between pigs and humans [21]–[26] making the pig a well-recognized model to study human infectious diseases [27].
Focusing on the natural site of primary mite infestations in young pigs, i.e. the inner surface of the ear pinnae, we investigated the impact of a S. scabiei var. suis infection on the skin microbiota. In a longitudinal study conducted over 21 weeks we compared the 16S rRNA gene sequences reflecting the development of the normal skin microbial community structure in healthy control animals with the microbiota present prior to, during and after S. scabiei var. suis infection.
Animal care and handling procedures used in this study followed the Animal Care and Protection Act, in compliance with the Australian code of practice for the care and use of animals for scientific purposes, outlined by the Australian National Health and Medical Research Council. The study was approved by the Centre for Advanced Animal Science (CAAS) and the QIMR Berghofer Medical Research Institute Animal Ethics Committees (DEEDI-AEC SA2012/02/381, QIMR A0306-621M).
Pigs were housed at the CAAS, Gatton, QLD. A total of 15 female, 3 weeks old Sus scrofa domesticus “Large White” breed siblings from the same pig breeding facility were adjusted to standard stable and feed conditions for 3 weeks prior to the start of the trial. After one week pigs were allocated randomly to 3 experimental groups (n = 5) and housed in identical but separate standardized, climate-controlled rooms, set at an average temperature of 27°C–30°C. All rooms were run on a continuous flow basis and concrete floors were cleaned twice daily. To assist monitoring and sampling, the formation of skin lesions and crusts associated with mite infestation was recorded on a weekly basis using an established scoring system [28].
The tractable experimental porcine scabies model was developed previously [28] where high mite numbers are observed in the inner surface of the ear pinnae of young piglets. The development of crusted scabies is generally seen within 10 to 13 weeks after mite infestation, but only in a subset of infected animals. The majority of individuals, analogously to humans, develops ordinary scabies and then, depending on the mite challenge, overcomes the infection without developing severe infestation and crust formation. Therefore, a previously developed strategy [28] was adopted for one cohort of the trial, where extreme parasite infestation was achieved by treatment with the synthetic gluco-corticoids immune-suppressant Dexamethasone (Provet, Brisbane). Synthetic gluco-corticoids are commonly used to promote infection in animal models [29], and the crusted scabies following corticosteroid therapy has been observed in humans [30]. The schedule of 1 week as adjustment period to stable conditions and of subsequent 2 weeks as the period to achieve in all individuals the Dexamethasone levels required for fast mite infestation was previously established [28].
The three experimental cohorts of pigs served as a Mite infected group (M), a Mite infected and Dexamethasone treated group (MD), and an uninfected, untreated Control group (C) (Figure S1, supplemental information). The MD group was administered a daily oral dose of 0.2 mg Dexamethasone per kg of body weight, starting 2 weeks prior to scabies mite challenge and sustained continuously throughout the trial (week -2). At the start of the trial (week 0), prior to infection of the cohorts M and MD, a scraping of approximately 1×1 cm2 of the epidermal layer of the skin was sampled with a sterile curette (Ø = 7 mm, Stiefel Laboratories Pty Ltd). This baseline sample was taken from the same site on the inside of the right ear pinna of each pig across all cohorts. Both MD and M cohorts were then challenged with comparable dosages of the scabies mite S. scabiei var. suis, as described previously [28]. In brief, crusts, sourced from a single site of a heavily infested individual, were dissected into approximately 0.5 cm2 pieces containing a few hundred mites and inserted into the vertical ear canal of both ears of the piglets. The animals were temporarily restrained, which prevented dislodgement of the crusts by agitation and ensured successful infestation. Skin scrapings were taken every two weeks unless otherwise stated, from all cohorts to monitor the development of the infection from healthy to moderate and subsequently to severe status of disease. Scrapings were performed in the same manner for every pig, alternating ears and following a predetermined map to ensure maximal conformity and to avoid repeated sampling of the same site. In the case of severe infestation, crusts were first lifted and collected for mite isolation. Subsequently the exposed skin area was sampled. After sample collection at week 16, pigs in all cohorts including the mite-naive cohort C were treated with the acaricide Doramectin (1% ivermectin solution, Pfizer Animal Health) by intramuscular injection at the recommended dosage of 300 mg per kg of body weight. The skin was allowed to heal completely for 5 weeks after Doramectin treatment. Final skin scrapes were taken at week 21. All skin samples were collected in 2 ml reinforced centrifuge tubes (Precellys, Bertin Technologies) containing 200 µl of enzymatic lysis buffer (20 mM Tris, pH 8.0, 2 mM EDTA, 1.2% (v/v) Triton-X100) and stored at −80°C until further processing. Samples representing base line (week 0), mild infestation (week 7), severe infestation (week 10) and after healing (week 21) were taken at pigs' ages of 6, 13, 16, 27 weeks, respectively (Figure S1, supplemental information).
The skin crusts collected from severe scabies infections were placed in lidded glass petri dishes and incubated over a moderately warm light source, which allowed mites to leave the substrate. Between 50 and 200 mites were collected into 2 ml reinforced centrifuge tubes (Precellys, Bertin Technologies). Mites were washed twice for 7 minutes with shaking in 4% paraformaldehyde (PFD) followed by one wash in PBS to remove external bacteria. Mites were centrifuged for 2 min at 10,000 rpm, wash solutions were removed and the mite pellet was stored at −80°C.
The skin samples were first incubated in lysozyme (20 mg/ml) for 30 min in a 37°C water bath. Isolated mite samples were not treated with lysozyme. To facilitate homogenization, six 2.8 mm stainless steel beads (Precellys, Bertin Technologies) were added to both skin and mite samples. The samples were processed in a tissue homogenizer (Precellys24, Precellys, Bertin Technologies) at 6,800 rpm for 30 s. Beads were removed and DNA was extracted from the samples using the DNeasy Blood and Tissue Kit (Qiagen) according to the manufacturer's instructions with the minor modification as follows. Samples were first incubated with 200 µl of Buffer AL, 40 µl Proteinase K in a 56°C water bath overnight and the standard protocol was followed for all subsequent steps. Purified genomic DNA was eluted in 100 µl buffer AE and the purity and concentration of the samples were analyzed by a spectrophotometer, NanoDrop 2000 (Thermo Scientific). DNA samples were stored at −20°C until required.
The bacterial 16S rDNA tag-encoded FLX-titanium amplicon pyrosequencing (bTEFAP) was contracted to Mr. DNA Molecular Research LP, Texas [31], [32]. Briefly, 16S rDNA was amplified from purified genomic DNA using the primer pair 27F (5' AGRGTTTGATCMTGGCTCAG 3')-519R (5' GTNTTACNGCGGCKGCTG 3') spanning V1-V3 region (∼500 bp product). The concentrations of all DNA samples were adjusted to a nominal 20 ng/µl and a 1 µl aliquot of each sample was used per 50 µl PCR reaction. A single-step fusion 30 cycle PCR using HotStarTaq Plus Master Mix Kit (Qiagen, Calencia, CA) was performed under the following conditions: 94°C for 3 min, followed by 28 cycles of 94°C for 30 s, 53°C for 40 s, 72°C for 1 min, and a final elongation step at 72°C for 5 min. Amplicon products from different samples were mixed in equal concentrations and purified using Agencourt Ampure beads (Agencourt Bioscience Corporation, MA, USA). Samples were then sequenced utilising Roche 454 FLX-titanium instruments and reagents following manufacturer's guidelines.
The 16S rDNA sequences were processed using the software packages QIIME 1.5[33]. Barcode sequences were removed by searching for exact matches and primer sequences were trimmed allowing 1 mismatch. Chimeras were removed using ChimeraSlayer [34]. Taxonomic assignments were retrieved by the RDP Classifer v2.2 [35] with a confidence threshold of 0.6. Operational taxonomic units (OTUs) were generated using the USEARCH package v5.2.32 [36] with an identity threshold of 97%. A representative sequence was selected for each OTU and taxonomically assigned with the RDP Classifier with a confidence cut-off of 0.6. Subsequently, statistical analysis was performed using R and the Calypso software (bioinfo.qimr.edu.au/calypso). The statistical analysis was done using one relative genus and OTU abundances, i.e. the number of reads assigned to each OTU or genus divided by the total number of reads obtained for each sample. Significant changes in the abundance of genera at different time points were detected by paired t-tests. Shannon index was used to estimate microbial community diversity (OTU level).
Representative sequences of OTUs assigned to Staphylococcus or Streptococcus by the RDP Classifier were used for phylogenetic analysis. Only OTUs with a relative abundance of at least 0.2% and 0.1% were included for the Staphylococcus and Streptococcus trees, respectively. Multiple alignments of reference 16S sequences of the corresponding genera were retrieved from the RDP database [37] and used as a reference to align the representative OTU sequences using HMMER3 [38], [39]. Phylogenetic trees were reconstructed using FastTree [40] with a generalized time-reversible (GTR) model.
All control animals (C group) remained clear of mites and healthy throughout the entire trial. At week 7 all scabies treated animals (M and MD groups) showed symptoms of successful mite infestation, such as a typical rash over large parts of the body leading to scratching behavior. No other symptoms of compromised health were detected in the M and MD animals. Severe infestations with crust formation in the inner part of the ear pinnae were seen between weeks 10 and 16, allowing isolation of mites. Crusts were formed in one member of the M group and in all animals treated with Dexamethasone (MD group). The antiparasitic drug Doramectin® was administered by intramuscular injection at the start of week 16, at a recommended dosage to kill off the mites within a week [41]. Infected skin healed within 5 weeks before the final scrapings were taken at week 21. A total of 140 skin samples were collected over the 21 week trial period. The sampling site was restricted to the inner sebaceous surface of the ear pinnae because this is the primary site of early scabies infestations in pigs. Ninety six samples were subjected for pyrosequencing bacterial 16S rDNA, yielding 1,366,477 high quality 16S sequences with an average length of 407 base pairs. Sequence data has been deposited at http://www.ncbi.nlm.nih.gov/sra with the accession number SRX392076. Samples taken from week 2 were excluded due to redundancy. A remaining subset of 57 samples collected in weeks 0, 7, 10, 13, 16 and 21 yielding 744,225 16S sequences were subjected to further analysis.
Twenty samples from the inner sebaceous surface of the ear pinnae were obtained from the Control cohort C containing on average 8,053 sequences per sample (range: 5,036–30,698). In these twenty skin samples we identified 204 different bacterial genera with at least 5 assigned sequences. The major genera are listed in Table 1. We observed significant changes in the skin microbiota of healthy animals during the 21 weeks period of the trial (Figure 1). At week 0, Streptococcus was the most abundant genus of the skin microbiota of healthy pigs (23% of 16S sequences) followed by Lactobacillus (13%) (Table 1, Figure 1). While Streptococcus remained relatively constant throughout the trial, Lactobacillus transiently dropped in abundance to 2% in weeks 7 and 10 (p = 0.01), but then became the most abundant genus with 45% of 16S sequences at week 21 (p = 0.001). We also observed a considerable reduction in microbial diversity: Lactobacillus and Streptococcus together represented about 30% of sequences at week 0 and had risen to about 70% at week 21, thereby having largely replaced the next 10 abundant genera. This change in community composition was reflected in a reduction of the community diversity measured by Shannon index. The Shannon index at week 21 was significantly lower than in weeks 0 (p = 0.009) and 10 (p = 0.008) (Figure 2bi) and the evenness had dropped from 0.75 to 0.66 (Figure S2, supplemental information). During the first 10 weeks the microbial diversity fluctuated only moderately, as indicated by similar Shannon indices (Figure 2bi).
In summary, the skin microbiota became gradually less diverse as the healthy piglets matured, reducing from a complex and diverse assemblage during the juvenile stage to two dominating genera, i.e. Lactobacillus and Streptococcus at week 21. While there is no comparable data from human infants, Grice et al. have previously reported that bacterial communities in adult human sebaceous microenvironments were less diverse than in other skin sites [42].
In total, 57 samples with a median of 8,798 sequences (range: 2,467–55990) were included to study changes of the skin microbiota associated with scabies. Prior to S. scabies infection in week 0 the control (C) and scabies treated (M) cohorts showed similar skin microbial community profiles, whereas the skin microbiota of the scabies+Dexamethasone treated cohort (MD) was significantly different. Streptococcus and Lactobacillus were the most dominant genera in cohorts C and M (Figure 2a) but Lactobacillus (24.9%) and Aerococcus were the most abundant genera within the cohort MD. Also, the microbial diversity was significantly lower in the MD cohort compared to the M and C cohorts (Figure 2b). Dexamethasone treatment has previously been reported to improve neutrophil-mediated killing of streptococci in a rat animal model [43], which could explain the observed differences in community composition. Notably, Dexamethasone treatment did not impact on the main changes in the skin microbiota seen in mite infested animals, as outlined below.
At week 0, before scabies mites were introduced, only a small percentage (≤0.5%) of Staphylococcus was present in all three cohorts (Figure 2a). During moderate scabies infection at week 7, a major increase in Staphylococcus was observed. At week 7, Staphylococcus abundance ranged from 6% to 76% in cohort M, and from 1% to 20% in cohort MD (Figure 2a). In contrast, Staphylococcus was nearly absent in the mite-free control cohort C, ranging from 0% to 0.2%. During severe scabies infestation (week 10), similar abundance of Staphylococcus remained in M and MD cohorts as in week 7, with one animal of the MD cohort reaching to 49%, compared with low numbers (0.2% to 1%) in the mite-free control cohort C (Figure 2a).
In week 10, when crust formation had occurred in one individual of the M cohort (M2) and all animals of the MD cohort, a dramatic reduction in community diversity was observed in samples taken from crusted sites in these animals (Figures 2bii and 2biii). The microbial community profiles from these skin samples, which were highly infested with mites, comprised up to 80% of Corynebacterium at the expense of other genera (Figure 2a).
In week 16, pigs were treated with the antiparasitic drug Doramectin. At week 21, after the mite infection was cleared in cohorts M and MD, the skin microbiota showed a similar diversity in all three cohorts (Figure 2b). However, the relative abundance of individual genera was markedly different between the control cohort C and the previously mite infected groups M and MD (Figure 2a). Staphylococcus remained abundant in most samples of the previously mite infected animals (reaching 27.4% in cohort M and 35.5% in cohort MD) compared to a minuscule presence of 0.2 to 1% in the control animals (Figure 2a). At the same time Lactobacillus was present at low levels in cohorts M (14%, Table 1) and MD (12%) while it was the dominant genus in the control cohort C (45%). Streptococcus was more dominant in the previously mite infected groups M (21%) and MD (29%) compared to the control cohort C (18.6%). At week 21, when all pigs were free of mites, Corynebacterium was equal to or below 6% in all cohorts.
The significant apparent increase in the Staphylococcus abundance in the scabies infected pigs indicated a correlation between S. scabiei infestations and Staphylococcus growth possibly combined with selective removal of other genera. To further identify the tentative species of Staphylococcus present, we constructed a phylogenetic tree of pig skin OTUs and known Staphylococcus species (Figure 3a). Prior to trial commencement at week 0, an OTU closely related to S. hominis (OTU3) was present at a very low abundance across all cohorts (Figure 3b). Staphylococci remained low throughout the trial in the scabies free control C. In contrast, from week 7 onwards Staphylococcus abundance significantly increased in cohorts M and MD and the community shifted from OTU3 to OTUs 1, 6, 8, 9 and 14, which are closely related to S. chromogenes (Figures 3a and 3b). After the scabies mites had been killed by drug treatment at week 16, staphylococci persisted in week 21 in cohorts M and MD. Two major taxonomic units, OTU1 and OTU2 (closely related to S. chromogenes and S. auricularis respectively) were in high abundance, followed by a lower abundance in OTU3, OTU4 (closely related to S. hominis and S. pasteuri) and OTU7 (closely related to S. felis) (Figure 3b).
Staphylococci are part of the healthy skin microflora but are also common causative agents of pyoderma in pigs and other animals [44]–[46], including humans [47], with different species predominating in pig and human skin [42], [48], [49]. S. hominis is a normal skin commensal of human and animal skin, whereas S. chromogenes is recognized as the causative agent of exudative epidermitis in pigs [45]. Further, S. chromogenes was described to play a role in skin lesions, dermatitis and otitis media in sheep, associated with infestation by the sheep scab mite Psoroptes ovis, and has been identified in the skin microbiota associated with P. ovis [50]. S. auricularis, S. pasteruri and S. felis are coagulase negative, skin residents of human and animals, and were reported to occasionally cause diseases [39], [51]–[53]. S. epidermidis prevails on healthy human skin while S. aureus is considered an important primary pathogen. In contrast, S. aureus was identified as the predominant species on healthy adult pig skin while S. epidermidis was less frequent [48]. A global increase in transmission of pathogenic methicillin resistant S. aureus strains has been reported between humans and animals [54]. In humans the link between scabies and S. aureus infections is well documented, however exclusively in epidemiological studies [55]–[60]. On human skin S. epidermidis usually has a benign relationship with its host [61] and was proposed to have a protective role in preventing colonisation with pathogenic bacteria, such as S. aureus [62]. Since S. aureus and S. epidermidis had been isolated from pigs before [48] we expected to detect these species on healthy skin and/or an increase in S. aureus during scabies infection; however neither was the case. Their absence in this experimental setting may be due to the sampling site being the sebaceous pinnae of the ears and not the back of the pigs, where they were detected previously [48]. Moreover, S. aureus likely becomes more abundant on the porcine skin after environmental exposure and human contact when housed under normal farming conditions [48]. The piglets in this trial were housed in a controlled facility without contact to other animals or humans, except for handlers who wore gloves and freshly washed overalls for every procedure.
Notably, in the trial presented here, staphylococci were abundant only in the pigs that had encountered scabies mites (cohorts M and MD), but barely measurable in the microbiome of mite free pigs in the control cohort C (Figure 3b). The presence of Staphylococcus in large abundance unveils an obvious risk factor for future recurrent bacterial infections in scabies infected animals. While the overall microbial diversity of the mite free skin in control cohort C was similar to that of scabies infected skin in cohorts M and MD, the significant increase of the genus Staphylococcus in the skin of scabies infected pigs strongly suggests that the mite infection selectively favored the establishment of staphylococci. Complement appears to be a major primary defense mechanism of the vertebrate host targeted at mites as well as bacteria. Scabies mites release proteins that inhibit complement [16], [17] and by reducing complement defense in their vicinity the mites may provide a microenvironment that fosters the survival of pathogenic bacteria [18]. S. aureus also displays an impressive arsenal of complement interference mechanisms [63]. Intriguingly, during human scabies infestations a substantial increase of S. aureus pyoderma is observed [64], implying that the combined presence of mites and bacteria may further amplify the inflammation response. Similarly, the dominance of the pathogenic S. chromogenes over the commensal S. hominis in the mite infested pig skin may be driven by the production of a range of virulence factors produced by S. chromogenes and the mites.
S. pyogenes is another important human pathogen commonly isolated from scabies associated pyoderma in human patients, a species that has not been reported in pigs. However, the genus Streptococcus was the most stable genus and relatively prominent in almost all skin samples throughout the duration of this trial (Figure 2a). Importantly, this Streptococcus population was not severely affected by S. scabiei infection, presenting at a constant abundance prior to, during and after the scabies mite infection. A phylogenetic tree of Streptococcus was constructed to study changes on OTU level (Figure 4a). OTUs closely related to S. alactolyticus dominated the streptococcal population (OTUs 1, 2, 4, 8 and 9) (Figure 4b). S. alactolyticus is considered to be part of the normal gut microbiota of pigs and has been isolated from gastrointestinal tracts of newly weaned piglets and feces [65], [66]. Our data showed that S. alactolyticus is also likely part of normal skin microbiota in juvenile pig ears. Interestingly OTU5 and 6 most similar to an opportunistic pathogen Streptococcus suis was present in a small proportion in the majority of the samples analyzed, but did not increase at any time point. S. suis is primarily an opportunistic pathogen of pigs but also an emerging human pathogen in the tropics [67]. Although S. suis is known to cause diseases in humans, particularly among pig handlers [68], our results suggest that handling scabatic pigs may not pose additional risk in this regard.
In week 10, when crust formation had occurred in individuals M2 and MD1–5, a dramatic reduction in community diversity was observed (Figures 2bii and 2biii). Skin samples taken from crusted sites comprised up to >70% of Corynebacterium at the expense of other genera (Figure 5). While in an ordinary scabies skin scraping sample even a single mite is very rarely seen, the immediate skin layer below a crust generally contains high numbers of mites, which together with their internal bacteria are part of the skin microbiota. The samples taken from crusted areas in week 10 were heavily infested with mites. Coincidently, a high proportion of the same Corynebacterium sequences were also seen in samples generated from isolated washed mites, indicating that Corynebacteria are part of the mite internal microbiota. Consequently, the high abundance of Corynebacterium sequences detected in the skin scrapings taken from the severely infected sites is likely due to the mite internal microbiota from mites present in the scraping. At week 21 the previously mite infected but now mite free cohorts M and MD showed low levels of Corynebacterium (Figure 2a, Table 1), reinforcing the hypothesis that this genus could be enriched predominantly within the gut of the mites themselves. Members of this genus are facultative anaerobes and hence well suited to the locally lowered O2 within the mite gut beneath dense skin crusts. Symbiotic Corynebacteria have been isolated from the alimentary systems of a variety of arthropods feeding on skin, such as Triatoma infestans [69] and the tick species Ixodes ricinus, Dermacentor reticulatus and Haemaphysalis concinna [52], providing potential novel strategies to control the transmission of diseases [70]. Thus, a subsequent study will focus on a thorough characterization of the scabies mite-internal microbiome and potential symbionts which could be targeted for scabies control.
We demonstrated that scabies infestation has an impact on the host's skin microbiota. In an experimental porcine skin model, abundance of potentially pathogenic Staphylococcus species increased with the onset of infection, and remained beyond treatment and healing. At the end of the trial previously scabies infested animals showed a much reduced presence of Lactobacillus compared to the control animals. The local Streptococcus population remained stable in all cohorts throughout the trial and seemed almost unaffected by scabies. Corynebacterium abundance in heavily mite infested skin samples was possibly related to the mite-internal microflora.
This study provides the first in vivo demonstration of a mite induced shift in the healthy skin microbiota, supporting direct evidence of the previously alleged link between scabies and pyoderma due to Staphylococcus infections, as seen in humans [9], [11]–[13]. The study focused on the primary site of mite infestations in young piglets and the experimental setup allowed monitoring of the site over time, i.e. prior to, during and after scabies mite infestation. It provides a basis for future investigation in human patients. The study highlights that scabies mite infestation is not a simple ‘itch’ but should be viewed as a complex disease involving a change in the status of the skin microbiota, which gives rise to serious secondary infections.
|
10.1371/journal.pgen.1006002 | Sc65-Null Mice Provide Evidence for a Novel Endoplasmic Reticulum Complex Regulating Collagen Lysyl Hydroxylation | Collagen is a major component of the extracellular matrix and its integrity is essential for connective tissue and organ function. The importance of proteins involved in intracellular collagen post-translational modification, folding and transport was recently highlighted from studies on recessive forms of osteogenesis imperfecta (OI). Here we describe the critical role of SC65 (Synaptonemal Complex 65, P3H4), a leprecan-family member, as part of an endoplasmic reticulum (ER) complex with prolyl 3-hydroxylase 3. This complex affects the activity of lysyl-hydroxylase 1 potentially through interactions with the enzyme and/or cyclophilin B. Loss of Sc65 in the mouse results in instability of this complex, altered collagen lysine hydroxylation and cross-linking leading to connective tissue defects that include low bone mass and skin fragility. This is the first indication of a prolyl-hydroxylase complex in the ER controlling lysyl-hydroxylase activity during collagen synthesis.
| Fibrillar collagens are major components of connective tissue extracellular matrix (ECM). Among them, type I collagen is the most abundant protein in the human body and a large constituent of bone, dermis, tendon and ligament ECMs; type I collagen is also present in the stroma of other organs including heart, lung and kidney where, when dysregulated, it significantly contributes to pathological fibrosis. Type I and other collagen molecules have triple-helical folding requirements and undergo numerous intracellular post-translational modifications in the endoplasmic reticulum (ER) and Golgi apparatus. We and others have shown that alterations/loss of specific collagen modifications can lead to severe congenital disease such as osteogenesis imperfecta (OI). Here, using a multidisciplinary approach, we describe functional studies of the SC65 protein (Synaptonemal Complex 65 or P3H4), a poorly characterized member of the Leprecan gene family of proteins. We provide evidence that SC65 is a critical component of an ER complex with prolyl 3-hydroxylase 3 (P3H3), lysyl-hydroxylase 1 (LH1), and potentially cyclophilin B (CYPB). Loss of Sc65 in the mouse results in instability of this complex, site-specific reduction in collagen lysine hydroxylation and connective tissue defects including osteopenia and skin fragility.
| Fibrillar collagens are abundant components of connective tissue extracellular matrix (ECM) [1]. They are formed by three polypeptides (termed α chains) each characterized by the presence of a long uninterrupted Gly-X-Y sequence repeat which folds into a characteristic triple-helical structure [1,2]. Among them, type I collagen (α1(I)2α2(I)) is the most abundant protein in the human body and the major constituent of bone, dermis, tendon and ligament ECMs. It is also expressed in the stroma of other organs including heart, lung and kidney where, when dysregulated, it can significantly contribute to pathological fibrosis [3]. Type I and other collagens undergo many intracellular post-translational modifications in the endoplasmic reticulum (ER) and Golgi apparatus [4]. For this reason, type I collagen has been considered a ‘prototypical model’ for mechanistic studies of enzymes that modify specific residues and chaperone proteins that facilitate folding of newly synthesized polypeptides in the secretory pathway. Intracellular modifications of collagens are critical for the structural integrity of the ECM (e.g. specific lysine residues in α1(I) and α2(I) can be hydroxylated and glycosylated and go on to form intermolecular cross-links in the ECM [5]). The extent of these modifications can also be tissue-specific and thus provide different biochemical properties to the same primary amino acid sequence of collagen [6]. With the recent identification of mutations in multiple genes encoding proteins involved in collagen post-translational modification and folding [7,8,9,10,11,12], the importance of these processes in human disease has become evident.
In the ER, specific proline and lysine residues of newly translated procollagen chains are modified by prolyl- and lysyl-hydroxylases, respectively. These enzymes share a highly conserved catalytic 2-oxoglutarate, ascorbate- and Fe(II)-dependent dioxygenase domain [13]. Collagen prolyl 4-hydroxylases (P4Hs) modify proline residues in the Y position of the Gly-X-Y repeat into 4-hydroxyproline (4Hyp). This frequent modification confers thermal stability to the folded triple-helical structure [14]. Interestingly, P4Hs form a tetrameric complex (α2β2) where the alpha (α) subunit is one of three prolyl 4-hydroxylase genetic isoforms and the beta (β) subunit is protein disulfide isomerase (PDI) [15]. Modification of proline residues into 3-hydroxyproline (3Hyp) by collagen prolyl 3-hydroxylases occurs far less frequently and always on proline residues in the X position of a Gly-X-4Hyp repeat (most completely on Pro986 of α1(I)) [16]. We previously demonstrated that CRTAP, a non-enzymic member of the Leprecan family of proteins which includes Synaptonemal Complex 65 (SC65) and the prolyl 3-hydroxylases (P3H1, P3H2 and P3H3), is an essential third subunit of a complex with prolyl 3-hydroxylase 1(aka LEPRECAN) and cyclophilin B (CYPB) in the ER, forming the so-called collagen prolyl 3-hydroxylation complex [17]. Loss of CRTAP in mice resulted in loss of the complex, lack of Pro986 hydroxylation, defective collagen fibrillogenesis and severe osteopenia. Indeed, human mutations in CRTAP cause recessive osteogenesis imperfecta (OI) [17]. This discovery then led to the identification of mutations in additional genes encoding ER-resident proteins that included P3H1, CYPB, FKBP65, HSP47 and OASIS, all causing recessive forms of OI [7,8,9,10,11,12].
Collagen lysyl-hydroxylases (LH1, LH2 and LH3) usually modify the side chain of lysine residues in the Y position of the Gly-X-Y repeat into hydroxylysine (Hyl) [5]. Certain triple-helical domain hydroxylysines can be o-glycosylated in the Golgi. Lysine and Hyl in collagen telopeptide domains are substrates for lysyl oxidase and the resulting aldehydes form cross-links in fibrils with helical domain lysine, Hyl or glycosylated Hyl side-chains [5,18]. Importantly, collagen lysyl-hydroxylases have not yet been formally demonstrated in a specific protein complex. In this study we describe the functional characterization of SC65, a protein that by primary sequence is closely related to CRTAP with as yet no known function. Using a multi-disciplinary approach combining mouse genetics, cell biology, microscopy, proteomics, and biochemistry, we demonstrate that SC65 is a critical component of an ER complex with prolyl 3-hydroxylase 3 (P3H3), lysyl-hydroxylase 1 (LH1), and potentially CYPB. The loss of SC65 in a mouse genetic model results in instability of this ER complex, leading to severely depleted lysine hydroxylation at helical domain cross-linking sites, disordered fibrils and significant defects in connective tissues that include osteopenia and skin fragility.
We have shown that SC65, similar to other Leprecan members, is an ER-associated protein that is expressed in multiple tissues and highly enriched in bone, cartilage, skin and kidney [19]. Using an Sc65 loss-of-function mouse model a low bone mass phenotype was observed [19]. However, no effect on prolyl 3-hydroxylation was detected and the molecular function of SC65 remained unknown. The original murine model (involving a gene-trap insertion) contained a neomycin-resistance cassette driven by its own promoter which, in theory, could have impacted the expression of nearby genes. For instance the Fkbp10 gene lies less than 900bp from Sc65 exon 1, is transcribed in the opposite direction and likely shares common regulatory elements. Importantly mutations in FKBP10 have been associated with recessive OI [10]. Therefore to eliminate this possibility and confirm the bone-related function for SC65, we report here the generation of a new Sc65-null allele. Rather than targeting the 5’ end of the gene, with the potential risk of interfering with Fkbp10 expression, LoxP sites were introduced flanking the last two exons of Sc65 (Fig 1A).
Heterozygous mice carrying the new floxed allele were mated first with mice expressing Flp-recombinase to eliminate the neomycin selection cassette, and then with mice expressing Cre-recombinase driven by an early ubiquitous promoter to generate a global KO (Fig 1A). Mice carrying the Sc65 Cre-excised allele (herein referred to as Sc65KO) were bred to homozygosity and western blot analysis of extracts from 3 day-old mouse tissues confirmed complete loss of SC65 protein (Fig 1B), indicating that the Cre-excised allele is a null-allele. This result was further supported by immunohistochemistry using an SC65 polyclonal antibody which did not stain osteoblasts on sections of Sc65KO compared to WT long bone (Fig 1C). Moreover, real-time PCR analyses confirmed rapid degradation of the SC65 truncated transcripts by NMD and additional western blot showed lack of potential truncated SC65 protein products in Sc65-KO primary skin fibroblasts (S1 Fig). These ex vivo data confirmed the expression of SC65 in bone forming osteoblasts and the validity of the new mouse model. Sc65KO mice were born at the expected Mendelian ratio and did not show any macroscopic differences compared to their WT littermates. To confirm the previously described osteopenic phenotype [19], micro-CT analysis of both femur and tibia of 6 month-old male mice (n = 9) was performed. A statistically significant decrease in trabecular bone volume/tissue volume (BV/TV) was detected in Sc65KO compared to WT long bones (p<0.05) (Fig 1D). Consistent with this, the connectivity density (Conn.D), which quantifies the connected structures within the trabecular bone network, was also significantly reduced in Sc65KO long bones. Moreover, the bone cortical thickness at the diaphysis differed significantly between the two genotypes (Fig 1D). Additional bone micro-CT measurements were acquired and were consistent with those reported for the previous mouse model of Sc65 inactivation [19], confirming a specific role for SC65 in the accrual/maintenance of normal bone mass.
To begin to investigate the function of SC65, we adopted a co-immunoprecipitation approach to identify SC65 candidate protein interactors. Murine 714 cells, a transformed mouse fibroblast cell line derived from BALB/3T3 [20], were transiently transfected with a SC65-Flag expression construct or an empty vector control and the lysates immune-precipitated (IP) using a Flag antibody and separated by gel electrophoresis (S2A Fig). IP fractions from both samples were analyzed and quantified by mass-spectrometric analysis. A total of 253 proteins, enriched at least 1.5 fold in the IP fraction of cells transfected with the SC65-Flag construct compared to the empty control vector, were identified as candidate SC65 interacting partners (S2B Fig). Among these proteins were several alpha-chains from different types of fibrillar collagen, including type V, I and XI (Table 1) strongly implicating SC65 in collagen processing and/or trafficking.
A role for SC65 in collagen post-translational modification was first indicated experimentally by SDS-PAGE analysis of extracts of skin collagen which consistently showed reproducible differences between Sc65KO and WT mice (Fig 2). The visual differences between null and WT SDS-PAGE banding patterns were reproducible across multiple extracts and gels.
Densitometry to quantify gel bands showed a significant decrease in the β12 dimer and an increase in γ112 (Fig 2B). The ratio of α1(I) to α2(I) also differed with α2(I) being more prominent relative to α1(I) in mutant skin extracts. In addition, a modest but consistent increased mobility was evident for both α1(I) and α2(I) chains from the Sc65KO dermis based on multiple SDS-PAGE gels using tissue from 1 month- and 6 month-old mice. The latter observation suggested decreased lysine hydroxylation and/or glycosylation in the triple-helical domain. Heat denatured extracts of decalcified bone collagen (required to see sufficient soluble collagen on SDS-PAGE) showed no marked electrophoretic differences in cross-linked chain properties and less effect on mobility of α-chains from Sc65KO versus WT bone.
Collagen α1(I) and α2(I) chains excised from SDS-PAGE gels (Fig 2) were subjected to in-gel trypsin digestion and tandem mass spectral analysis to quantify collagen post-translational modifications. Given the role of Crtap in prolyl 3-hydroxylation, known sites of 3-hydroxyproline (3Hyp) were initially targeted. As reported for the original gene-trap Sc65KO mice [19] no known prolyl 3-hydroxylation sites showed any significant suppression in level of 3Hyp in the Sc65KO mice (–see S1 Table for results for type I collagen from skin and bone).
Based on the increased collagen α1(I) and α2(I) chain mobility and disturbed cross-linked chain ratios, tandem mass spectral analyses were also directed at sites of lysine hydroxylation, particularly those at cross-linking sites. In-gel trypsin digested collagen type I chains initially revealed marked under-hydroxylation at helical cross-linking sites K87 and K930/933 for Sc65KO compared to WT skin. These residues are preferred substrates for LH1 activity [21]. To confirm this more quantitatively for whole tissue collagen and also target collagen telopeptide lysines, bacterial collagenase digests of decalcified bone and skin were analyzed by tandem mass spectrometry. The results are summarized in Fig 3 and confirm a marked under hydroxylation of K87 and K930 in α1(I) of bone (a) and skin (b).
At K87, which normally is also glycosylated (mostly glcgalHyl in skin, galHyl in bone), the under-modification is more complete in skin than bone perhaps explaining the more obvious α-chain increased mobility for skin collagen (Fig 2). The α2(I) chain also showed under-hydroxylation at its equivalent two crosslinking sites K87 and K933. Neither N-telopeptide nor C-telopeptide hydroxylysine content, which requires LH2 activity [22], was found to differ between Sc65KO and WT bone or skin collagen α1(I) or α2(I) chains.
Based on the observed under hydroxylation of cross-linking lysines in Sc65KO mouse bone and skin collagen, bone collagen was acid hydrolyzed and analyzed by HPLC to quantify pyridinoline cross-links. The results showed a reversed ratio of hydroxylysylpyridinoline (HP) to lysylpyridinoline (LP) of 0.3/1 from Sc65KO bone compared with 5/1 from WT bone. This is consistent with the mass spectral data on linear peptides showing severe under-hydroxylation of the helical-domain lysines that control whether HP or LP cross-links are formed. The total content of HP+LP pyridinoline cross-links was in the normal range for mouse bone from both genotypes, consistent with the mass spectral findings that linear N- and C-telopeptide lysine hydroxylation was unaffected.
The severe reduction of collagen triple-helical lysine hydroxylation identified by mass-spectrometry indicated that LH1 function is dependent on the presence of SC65 protein. Importantly, in our previous mass-spectrometry screening of potential SC65 interactors, LH1 was 4.9 fold enriched in the IP fraction derived from cells transfected with the SC65-Flag construct compared to controls, suggesting that SC65 and LH1 may cooperate in a protein complex.
To confirm this specific interaction 714 cell lines stably expressing SC65-Flag or an empty vector (EV) control were developed and transiently transfected with an HA-tagged LH1 expression plasmid. Direct co-immunoprecipitation assays utilizing either a Flag or HA antibody were then performed, followed by Western blotting with the HA or the Flag antibody or both. These assays confirmed a reciprocal and direct interaction between SC65 and LH1 (Fig 4A).
Moreover, Western blots of primary calvarial osteoblast and skin fibroblast lysates from Sc65KO mice consistently demonstrated a significant decrease in LH1 protein compared to WT controls (Fig 4B). This important result indicated that loss of SC65 results in partial loss of LH1 protein.
Considering the close homology between CRTAP and SC65 and the knowledge that CRTAP is required to stabilize the P3H1/CRTAP/CYPB protein complex, we hypothesized that LH1 is associated with a similar complex. Significantly, mice with loss of function of Leprel2 (encoding P3H3) have the same loss of tissue type I collagen lysine-hydroxylation as that observed in the Sc65KO mice when analyzed by tandem mass spectrometry (Hudson et al, in preparation). This suggested that P3H3 was also part of a SC65/LH1 complex. A direct interaction between SC65 and P3H3 was confirmed by a similar co-IP procedure to that described above, using 714 cells stably expressing SC65-Flag or an EV control but using primary antibodies against P3H3 or the Flag tag for the pull down (Fig 5A).
Also in this case, Western blot analysis showed that lack of SC65 in mouse primary calvarial osteoblasts and skin fibroblasts results in severe loss of P3H3 protein compared to WT control cells (Fig 5B). We then transiently transfected 714 cells with HA-tagged LH1 and successfully co-precipitated P3H3 or LH1, using either a HA-tag or P3H3 antibody, respectively (Fig 6A).
This strongly suggested that SC65/LH1/P3H3 are interlinked within a protein complex in the ER. To assess whether CYPB could also be part of such complex, as in the P3H1/CRTAP/CYPB complex, we transfected 714 cells with HA-tagged CYPB and successfully co-precipitated SC65-Flag using the HA antibody (Fig 6B). Notably, initial evidence for a potential interaction between LH1 and Cyclophilin B was recently provided [23,24]. To further pursue the nature of the suspected SC65/P3H3/LH1/CYPB complex we performed preparative, high resolution size exclusion chromatography. 714 cells stably transfected with SC65-Flag were used in multiple large IP experiments using the monoclonal antibody against Flag. To maintain its integrity the protein complex pulled down in the IP was eluted from the magnetic beads by competitive binding with an excess of Flag peptide and the eluted suspension was then loaded and run on a Superdex 200 column. Collected fractions were concentrated, run on SDS-PAGE and blotted with antibodies against members of the proposed complex (S3 Fig). SC65 (detected by the Flag antibody) and P3H3 protein co-eluted in fractions 23–26 (equivalent to an estimated MW of about 250kDa). LH1 protein peaked in fractions 28–30 with trace amounts in fractions 25–27. Taken together these findings demonstrate the existence of an ER protein complex that includes SC65/P3H3 and LH1 but no evidence so far that CYPB is included. We have also observed that lack of SC65 significantly reduces the content of LH1 and P3H3 in primary osteoblasts and fibroblasts, but not of CYPB (Fig 6C).
Because collagen cross-links are essential for the integrity of the dermis as well as the tensile properties of skin [25,26], we harvested biopsies of dorsal skin from Sc65KO and WT mice for light microscopy, visualization of collagen fibrils under polarized light and electron microscopy and biomechanical testing. Light microscopy of skin sections stained with hematoxylin & eosin showed decreased density of collagen fibers in the dermis, thinning of the muscle layer below the hypodermis and frequent presence of tears in the Sc65KO compared to WT sections (Fig 7A).
Serial skin sections were also labeled with Sirius Red, a collagen specific stain, and visualized under polarized light. The birefringent pattern of collagen in skin from Sc65KO mice appeared different compared to WT controls, and suggested a decreased size of collagen fibers that tended to be more green than orange [27] (Fig 7B).
Immunohistochemical analysis using the anti-SC65 antibody showed SC65 expression in WT dermal fibroblasts and lack of specific signal in skin sections from Sc65KO mice (S3 Fig). Ultrastructural analysis of dermal collagen from skin of 7 month-old Sc65KO mice revealed several differences compared to WT controls. Mutant dermal collagen fibrils were packed less orderly than age-matched wild type and their diameters were more uniform than WT fibrils (Fig 7C and 7D). Thus, 91.7% of mutant fibrils ranged from 40 to 120 nm in diameter compared with 78.7% of WT fibrils. The cross-sectional profile of collagen fibrils from Sc65KO skin was more disordered than WT with occasional thick and abnormally shaped fibrils with irregular contours, resembling ‘cauliflower-shaped’ fibrils [28] (Fig 7C inset). These fibrils are considered a hallmark of disturbed collagen fibrillogenesis, and are frequently observed in dermis from Ehlers-Danlos syndrome patients [29]. Furthermore, the quantified space between fibrils in Sc65KO skin was greater than in WT controls (Fig 7E). These findings suggest defects in the maturation and the lateral fusional growth of fibrils.
Finally, it is generally accepted that skin tensile strength correlates directly with the overall organization, content, and physical properties of the collagen fibril network [30]. Therefore, to obtain an objective measurement of skin fragility, we determined the tensile strength in portions of dorsal skin from wild-type and mutant mice. In load-to-failure biomechanical testing, skin from Sc65KO animals ruptured prematurely at a peak load of approximately 12N compared to WT skin that ruptured at approximately 23N (p<0.01)(Fig 7F and 7H). The tensile failure loads were reduced equally in both mutant male and female skin samples. The skin stiffness was also significantly reduced in Sc65KO compared to WT mice (Fig 7G). The finding of decreased tensile strength and stiffness directly support the histological observation of skin fragility and decreased collagen density.
The results presented demonstrate that SC65 is required for the normal post-translational modifications of collagen chains in the ER. This function is clearly distinct from, but related to, that of CRTAP, a highly homologous protein (~55% identical in amino acid sequence) that we have shown is essential for prolyl 3-hydroxylation17. Similar to CRTAP, SC65 appears necessary in another protein complex involved in the post-translational modifications of fibrillar collagen. While CRTAP binds to and forms a stable complex with P3H1 [17], SC65 forms a stable complex with P3H3. The formation of this complex is strongly supported by size-exclusion chromatography where SC65 and P3H3 eluted in the same fractions. However, what is unexpected is the specific effect of SC65 on collagen lysyl hydroxylation rather than on prolyl 3-hydroxylation. Our findings indicate that the SC65/P3H3 complex also involves LH1 and that the loss of SC65 results in at least partial loss of LH1 protein and activity. Interestingly, while some degree of overlap exists in chromatography fractions containing SC65, P3H3 and LH1, suggesting they can exist within a single complex, the later elution of LH1 also suggests a less stable interaction with the SC65/P3H3 core and dissociation from the complex during elution with Flag peptide and/or in less than ideal conditions such as those present in the column. It is also important to note that LH1 is not present in fractions that contain only SC65 or only P3H3, therefore it is unlikely that LH1 forms a complex with either of these proteins alone but rather with both of them together. While the precise stoichiometry of the complex is not yet defined, a putative trimeric complex containing SC65/P3H3/LH1 would have an estimated MW>200 kDa. These findings provide the first description of a prolyl-hydroxylase complex in the ER that affects lysyl-hydroxylation of collagen. However, while loss of SC65 causes under-hydroxylation of helical lysine residues, we have not been able to detect loss of prolyl 3-hydroxylation at any known site in collagen I from bone or skin from either Sc65KO (Suppl. Table 1) or P3h3-null mice (Hudson et. al., in preparation). Either the enzyme activity of P3H3 is redundant (e.g. compensated by P3H2) or lost during evolution but complex formation and collagen chaperone function are retained.
Although we show that SC65 is capable of interacting with Cyclophilin B in vitro (Fig 6), as might be expected given its homology to CRTAP, the results of gel filtration chromatography (S3 Fig) imply that in the ER this is likely a weak or transient interaction. It is also notable that mutations in PPIB (encodes CYPB) depending on their location can cause not only recessive OI [8,9] but also a severe form of skin fragility and blistering in American quarter horses, known as HERDA [23,31]. Therefore, mutations affecting different regions of CYPB can have different tissue effects, perhaps as a result of which ER complex function is compromised.
Given the current evidence for a distinct complex between SC65 and P3H3 that involves LH1, along with our previous characterization of the prolyl 3-hydroxylation complex (CRTAP/P3H1/CYPB) [17], it is becoming clear that multiple high molecular weight complexes involved in regulating collagen post-translational modification have evolved in the ER (Fig 8). It is interesting to note that among the known and suspected proteins involved in these 2 complexes, only P3H1 and P3H3 have ER retention signal peptides (KDEL or REEL, respectively). Complex formation therefore may have the added function of helping retain the other proteins in the ER.
It remains to be seen, given that three prolyl 3-hydroxylases and three lysyl hydroxylases are expressed differentially in vertebrate tissues [32,33,34], whether other combinations of protein subunits are possible. In terms of clinical significance, loss of CRTAP causes a pronounced skeletal phenotype in mice and severe osteogenesis imperfecta in humans [17,35]. Loss of SC65 in mice causes no obvious phenotype during embryonic development or growth; while adult Sc65KO mice showed significant osteopenia [19], the most pronounced defect was in skin, consistent with the underlying pathobiochemistry revealed in skin collagen. The under-hydroxylation of lysines at triple-helical domain cross-linking sites, and resulting inversed ratio of HP/LP cross-links in bone collagen closely resembles the collagen pathology in Ehlers-Danlos Syndrome type VI (EDS VI) caused by PLOD1 (encodes LH1) mutations (for a review see [36]). In the latter disorder, telopeptide domain lysine hydroxylation is not affected, only triple-helical domain [21,37]. The clinical phenotype includes soft skin, lax joints and kyphoscoliosis [38]. It is possible, therefore, that mutations in SC65 (and LEPREL2) may be responsible for mild EDS variants sharing features of EDS VI but without a PLOD1 mutation.
A possible mechanism that might operate to affect collagen cross-linking and hence tissue material properties selectively in skin would be through the lack of glycosylation of the K87 lysines in α1(I) and α2(I) chains. We interpret the increase in γ112 trimer and decrease in β12 dimer (Fig 2) as a consequence of altered cross-link placement in fibrils. Normally, skin type I collagen has a fully glycosylated α1(I) K87 which forms an aldimine cross-link with an α1(I)C-telopeptide allysine in fibrils after lysyl oxidase action. In Sc65KO skin, α1(I) and α2(I) K87 have no sugars attached and are only partially hydroxylated (Fig 3). This potentially could enable two α1(I)C-telopeptides in the same molecule to interact and form an intramolecular aldol (but not when α1(I)K87 of neighboring molecules are glycosylated). Such a shift in bond placement would result in the enrichment of native molecules extracted by dilute acetic acid that have intramolecular aldols at both ends which would run as a γ112 trimer on SDS-PAGE.
In summary, our studies of SC65 have revealed its importance in fibrillar collagen processing. There is clearly still much to learn about the complexity of protein interactions that have evolved to regulate the diversity in collagen post-translational quality and function evident between different tissue types.
Due to the close proximity of the Fkbp10 gene to the transcription start site of Sc65, the Sc65 conditional targeting construct (using the vector p-flrt-neo-2XloxP) was designed to introduce loxP sites flanking the last 2 exons (7 and 8) of Sc65. This strategy eliminates the poly-A signal and creates an unstable transcript upon Cre-recombinase activity, and was successfully used at the University of Connecticut Health Center Gene Targeting and Transgenic Facility (Dr. Siu-Pok Yee, Director) where the construct was electroporated into mouse ES cells derived from F1(C57B6/jx129sv) embryos. Drug-resistant clones were screened for homologous recombination by long-range PCR using primer sets on both the 5' and 3' end of the gene. Confirmed positive clones were expanded and used to generate chimeric animals by ES-morula (obtained from CD1) aggregation. Chimeric males derived from two independent clones (1E6 and 2G4) were bred with Rosa26-Flpe females (Jax stock no. 003946 that had been backcrossed with C57BL6/j for over 20 generations) for germline transmission and removal of the Neo cassette to generate the final Sc65 floxed allele. To generate a global Sc65 inactivation, males carrying the floxed allele were then mated with Hprt-Cre females (Jax stock no: 004302, which has been backcrossed for over 20 generations with C57BL6/j). Pups from these females were heterozygous for the KO allele, and were bred as needed to generate heterozygote, homozygote mutant or littermate control wild type mice. The excision into the Sc65 locus was verified by PCR analysis using long-range PCR. PCR genotyping was done using the GoTaq PCR kit and reagents (Promega, Madison, WI, USA) and a Master cycler PCR machine to run the samples (Eppendorf AG, Hamburg, Germany). The use of laboratory mice was approved by the University of Arkansas for Medical Sciences (UAMS) IACUC committee. Mice were housed in a pathogen free facility with 12h light/dark cycle with unlimited access to water and standard chow diet.
714 mouse embryonic fibroblasts [20] and primary murine skin fibroblasts cells were grown in DMEM; primary cells from murine calvariae were grown in α‐MEM. All media contained 4500mg/L glucose and 110mg/mL sodium pyruvate (HyClone, Thermo Scientific, Pittsburgh, PA, USA) and were supplemented with 10% fetal bovine serum (FBS), L‐glutamine (2mM), 100units/mL penicillin, and 100mg/mL streptomycin. Primary osteoblasts from calvaria from 2–5 day-old pups were obtained using standard procedures as previously described [17]. Primary murine fibroblasts were isolated from 2–5 day old pups from ear snips that were minced in 0.25% trypsin and allowed to digest for 1 hour at 37°C, plated in 6 well dishes and allowed to expand. Primary fibroblasts were also treated (where indicated) with cycloheximide at a concentration of 100ug/ml and cells were collected 12hr after treatment. RNA was extracted using TriPure Isolation reagent according to the manufacture’s (Sigma Aldrich) protocol. For Real Time PCR, cDNA was synthesized from 1ug of RNA using the First Strand cDNA Synthesis Kit (Roche). cDNA samples were diluted 1:5 and quantitative PCR was performed with the LightCycler version 1.5 instrument using Roche Applied Science reagents according to manufacturer recommendations. Primer sequences were as follows: for SC65—FWD: 5’-ATGCAGCAGAACCTGGTATATT-3’ and RVS: 5’-GTCTGGTTGTGGTAGAGCATA-3’; for GAPDH—FWD: 5’-GCAAGAGAGGCCCTATCCCAA-3’ and RVS: 5’-CTCCCTAGGCCCCTCCTGTTATT-3’. The pCMV-SC65-DDK-C plasmid (murine, SC65-FLAG) was purchased from TransOMIC Technologies, Huntsville, AL, USA). Full-length LH1 and CypB cDNAs and exons 1–6 of Sc65 were amplified from 714 mouse embryonic fibroblasts and cloned into the pCMV-HA-C plasmid (TransOMIC Technologies, Huntsville, AL, USA) using the In-Fusion HD Cloning kit (Clontech, Mountain View, CA, USA) according to the manufacturer’s protocol and sequence verified.
To confirm candidate interactors of SC65, 714 cell lines stably expressing SC65-Flag were generated as follows. The SC65-DDK-C sequence (from the pCMV-Sc65-DDK plasmid) was sub-cloned into the pLEN expression vector [44] (carrying G418 resistance) utilizing the In-fusion HD Cloning kit as above. The construct was linearized by digestion with Nde1 (New England Biolabs, Ipswich, MA, USA), transfected into 714 cells with Lipofectamine 3000 (Life Technologies) according to the manufacturer’s protocol and maintained in G418 selection (Sigma Aldrich, St. Louis, MO, USA). A negative control cell line was generated by transfecting an empty pLen vector (EV) into 714 cells. For direct co-immunoprecipitation of LH1-HA, five 10cm dishes of both cell lines were transfected with 20ug of LH1-HA with Lipofectamine 3000 (Life Technologies) according to the manufacturer’s protocol. Forty-eight hours after transfection, immune-precipitation of SC65-FLAG or LH1-HA was performed using the Thermo Scientific Magnetic Crosslinking IP Kit with modifications. Cells were lysed in IP Lysis/Wash buffer to a volume of 2ml, centrifuged for 10min @ 10,000 RPM at 4°C. The supernatant was incubated with 10-15ug of anti-DDDK (Bethyl Laboratories, Montgomery, TX, USA) or anti-HA (Santa Cruz Biotechnologies, Dallas, TX, USA) antibody overnight at 4°C. A/G magnetic beads were added to the lysate/antibody mixture and allowed to conjugate for 4 hours at 4°C. Beads were washed twice with IP/Lysis wash buffer and ultrapure water. Beads were incubated in an elution buffer (pH 2) for 10min and elutions were run on 10% SDS-PAGE gels and immunoblotted for target proteins. Co-immunoprecipitation of P3H3 and CYPB-HA was carried out the same as LH1-HA except endogenous P3H3 was immunoprecipitated utilizing a polyclonal P3H3 antibody (2.5ug, ProteinTech, Chicago, IL, USA). To determine if LH1 and P3H3 interact, 714-pLen- EV cells were transfected with LH1-HA as above in the presence of ascorbic acid and lysed after 48hours. Immunoprecipitation was performed as above using anti-HA and P3H3 antibodies. Controls for these experiments included untransfected 714-pLen-EV cells incubated with HA antibody to determine non-specific binding proteins and LH1-HA transfected cells that were incubated with beads alone to determine any proteins non-specifically binding to the beads.
To further pursue the nature of the putative SC65/P3H3/LH1/CYPB complex we performed a preparative, high resolution size exclusion chromatography. For this experiment, five to ten 100mm cell culture dishes of confluent 714 cells stably transfected with SC65-Flag and treated for 72 hours with ascorbic acid (100μg/ml) were used in multiple large IP experiments using a monoclonal antibody against Flag (see “Co-immunoprecipitation assays” in Methods), (Bethyl). To maintain its integrity, the protein complex pulled down in the IP was eluted from the magnetic beads by competitive binding with an excess of Flag peptide (500ug/ml, Sigma Aldrich). Size exclusion chromatography was performed using a Superdex 200 Increase 10/300 GL column (approximate bed volume of 24ml) with the AKTA purifier FPLC (GE Healthcare). All experiments (standards and samples) were performed at room temperature following the manufacture’s protocol. The column was first equilibrated with 2 column volumes (CV) of room tempered water followed by equilibration with 2 CV of eluent utilized in each experiment (50mM phosphate buffer containing 0.15M NaCl pH7.4). In order to standardize the column, a Gel Filtration Calibration Kit (GE Healthcare) containing Thyroglobulin, Ferritin, Aldolase, Conalbumin and Ovalbumin was used. The void volume (Vo) at ~8.5 mL was used with the elution volume (Ve) to generate a protein standard curve to calculate the size of eluted proteins (y = -0.3048x+0.9212, R2 = 0.9937). The eluted suspension from the immunoprecipitation was then loaded onto the column and ran at the recommended flow rate of 500ul/min. Fractions of 500ul each were eluted from the column, further concentrated by Trichloroacetic Acid (TCA, Sigma Aldrich) precipitation, re-suspended in PBS, run on a SDS-Page and blotted with relevant antibodies (see above “Western blotting”).
Tissues and primary cell cultures were lysed into RIPA buffer (50mM Tris‐HCl pH 7.5, 150mM NaCl, 0.1% SDS, 1mM EDTA, 0.5% sodium deoxycholate, and 1% Triton X‐100) containing a cocktail of protease inhibitors including EDTA (cAMRESCO, Solon, OH, USA). Lysates were centrifuged (15,000g) and supernatants collected and quantified using the Bio‐Rad protein assay dye reagent (Bio‐Rad, Hercules, CA, USA). Proteins were separated by 10% SDS-PAGE gels according to standard techniques, transferred to a nitrocellulose membrane and blocked for 30min in 5% milk. Primary antibodies were Sc65 (cat# 15288-1-AP, ProteinTech), β-actin (cat# A00702, GenScript, Piscataway, NJ, USA), LH1 (cat#NBP2-38770, Novus Biologicals, Littleton, CO, USA), P3H3 (cat# 16023-1-AP, ProteinTech), anti-DDDK (cat# A190-102A, Bethyl Laboratories), anti-HA (cat# sc-7392, Santa Cruz Biotechnologies). Secondary antibodies were goat anti‐rabbit or anti‐mouse IRDye 680LT (LICOR Biosciences, Lincoln, NE, USA). Membranes were scanned using a LICOR Odyssey instrument.
Femur, tibiae and skin were harvested and fixed in 10% buffered formaldehyde. After demineralization in 30% EDTA or non-decalcified femur or tibiae, the specimens were dehydrated, cleared and embedded in either paraffin or in Methyl methacrylate according to standard procedures [45]. Paraffin embedded sections were cut at 5 microns on a Leitz1512 microtome and specimens embedded in Methyl Methacrylate were cut on a Leica RM2255 automatic heavy duty retractable microtome using a D-profile tungsten carbide knife. Sections were mounted on Silane+ slides (Newcomer’s Supply, Middleton, WI) using Haupt’s media to allow the sections to adhere better to the slides. Skin was stained with hematoxylin and eosin and both bone and skin sections were stained with Sirius red. A 0.1% solution of Sirius red (Direct Red 80, Sigma-Aldrich) was prepared in saturated aqueous solution of picric acid. After staining, sections were rinsed in acidified water (87.5mM acetic acid) and dehydrated in absolute alcohol, cleared and mounted in synthetic resin (DPX Mountant for histology, Sigma-Aldrich). Sirius red stained sections were analyzed under polarized light and images were taken using a Zeiss AxioImager equipped with DIC filter, polarizer, Zeiss AxioCam MRc5 digital camera and Zeiss Axiovision software.
For immunohistochemistry staining, bone and skin sections mounted on slides were placed in Citrate Buffer pH 6.0 which had been heated to boiling in a microwave oven and allowed to cool for 30 minutes at room temperature. Endogenous peroxidase was blocked using the Peroxidase Solution supplied in the CTS008 Anti-Goat HRP-DAB Cell and Tissue Staining Kit (R&D Systems). After 5’ the slides were washed with PBS pH7.4, blocked with the blocking buffer supplied in the kit and incubated overnight with the antibody. The next morning the sections were washed in PBS pH7.4, secondary antibody applied incubated for 30 minutes, washed and DAB applied. After development of DAB the sections were washed with distilled water, counterstained with Mayer’s Hematoxylin, dehydrated, cleared and mounted with Permaslip. Pictures were taken using a Zeiss AxioImager equipped with Zeiss AxioCam MRc5 digital camera and Zeiss Axiovision software.
For immunofluorescence staining, skin sections were de-paraffinized and hydrated in a series of xylene and ethanol incubations according to standard protocols. Sections were heated in 10mM citrate buffer (pH 6.0) until boiling and let incubate for 30 minutes until cooled. Sections were then blocked in normal serum blocking buffer (3% goat serum, 1mg/ml BSA, 0.1% Triton-X) for 1 hour. Sections were then incubated with SC65 primary antibody (ProteinTech) overnight. After washing, sections were incubated with AlexaFluor 488 goat anti-rabbit secondary antibody (Life Technologies) for 30 minutes followed by washing and mounting in DAPI Fluoromount (SouthernBiotech, Birmingham, AL, USA). Images were acquired on a Zeiss AxioImager scope.
All μCT analyses were consistent with current guidelines for the assessment of bone microstructure in rodents using micro-computed tomography [46]. Formalin-fixed tibiae and femurs from WT and Sc65KO mice 6 months old (n = 9–10) were imaged using a Micro-CT 40 (Scanco Medical AG, Bassersdorf, Switzerland) using a 12μm isotropic voxel size in all dimensions. The region of interest selected for analysis comprised 240 transverse CT slices representing the entire medullary volume extending 1.24mm distal to the end of the primary spongiosa with a border lying 100μm from the cortex. Three-dimensional reconstructions were created by stacking the regions of interest from each two-dimensional slice and then applying a gray-scale threshold and Gaussian noise filter (sigma 0.8, support 1, threshold 245) as described [45] using a consistent and pre-determined threshold with all data acquired at 55kVp, 114mA, and 200ms integration time. Fractional bone volume (bone volume/tissue volume; BV/TV) and architectural properties of trabecular bone such as trabecular thickness (Tb.Th, μm), trabecular number (Tb.N, mm-1), and connectivity density (Conn. D, 1/mm3) were calculated using previously published methods [45]. Femoral and tibial cortical geometry was assessed in a 1mm-long region centered at the femoral midshaft. The outer contour of the bone was found automatically using the built-in manufacturer’s contouring tool. Total area was calculated by counting all voxels within the contour, bone area by counting all voxels that were segmented as bone, and marrow area was calculated as total area minus bone area. This calculation was performed on all 25 slices (1 slice = 12.5μm), using the average for the final calculation. The outer and inner perimeter of the cortical midshaft was determined by a three-dimensional triangulation of the bone surface (BS) of the 25 slices, and cortical thickness and other cortical parameters were determined as described.
Dorsal skin samples were collected from 7 months old WT and Sc65KO animals (n = 4) and fixed overnight at 4°C in 2.5% glutaraldehyde (Electron Microscopy Sciences (EMS), Hatfield, PA, USA), 0.05% malachite green (Sigma Aldrich) in 0.1M sodium cacodylate buffer, pH 7.2 (EMS). After washing with 0.1M sodium cacodylate buffer, the samples were post-fixed for 2 hrs with 1% osmium tetroxide (EMS), 0.8% potassium hexaferrocyanide (Sigma Aldrich) for 2 hours and 1% tannic acid (EMS) for 20 min. The samples were rinsed with molecular grade water and stained with 0.5% uranyl acetate (EMS) for 1 hour then dehydrated with a graded alcohol series and propylene oxide before embedding in Araldite/Embed 812 (EMS). Sections (50nm) were cut on a Leica UC7 ultra-microtome and collected on formvar carbon coated slot grids and post-stained with uranyl acetate and lead citrate. Imaging was done using a Technai F20 (FEI, Netherlands) at 80kV. Analyses of collagen fibril diameter (>200 fibrils/mouse) and of collagen inter-fibril space were performed with the Leica Application Suite v 3.0 image analysis software (Leica Microsystems, Milan, Italy) on 5 sections for each mouse (n = 3) at the magnification 19000X.
Samples were prepared from the dorsal skin of adult (7 month-old) males and females Sc65KO and WT mice (n = 7 or more, see Fig 7). The skin was harvested and cut into about 2cm wide by 4cm long pieces. The long axis of the sample coincided with the cranio-caudal axis of the mouse. The samples were clamped between two custom built aluminum fixtures at the superior and inferior ends (see Fig 7G). Tension tests were performed on a MTS 858 Bionix Test System (Eden Prairie, MN) servo-hydraulic material test machine. The samples were preloaded to 0.2Newtons and then loaded to failure in tension at a constant rate of 10mm/min. Peak load and stiffness (calculated as the slope of the load/deflection curve) were recorded using TestWorks 4 software (Eden Prairie, MN).
All parameters measured are presented as mean ± Standard Deviation and were analyzed with the Student’s t-test using a two-tailed distribution and two-sample equal variance as appropriate. All calculations were performed utilizing Microsoft Excel. P values <0.05 were considered statistically significant and reported as such.
All animal work (i.e. on rodents) performed in this study was conducted in accordance to local, State and US Federal regulations. The UAMS IACUC has approved the animal protocol (AUP#3349 entitled "Role of the Leprecan Genes in Skeletal Formation") describing all the procedures performed in this study. Mice were euthanized to harvest relevant tissues according to the recommendations of the Guide for Care and Use of Laboratory Animals (8th Edition).
|
10.1371/journal.pcbi.1002573 | A Model-Based Bayesian Estimation of the Rate of Evolution of VNTR Loci in Mycobacterium tuberculosis | Variable numbers of tandem repeats (VNTR) typing is widely used for studying the bacterial cause of tuberculosis. Knowledge of the rate of mutation of VNTR loci facilitates the study of the evolution and epidemiology of Mycobacterium tuberculosis. Previous studies have applied population genetic models to estimate the mutation rate, leading to estimates varying widely from around to per locus per year. Resolving this issue using more detailed models and statistical methods would lead to improved inference in the molecular epidemiology of tuberculosis. Here, we use a model-based approach that incorporates two alternative forms of a stepwise mutation process for VNTR evolution within an epidemiological model of disease transmission. Using this model in a Bayesian framework we estimate the mutation rate of VNTR in M. tuberculosis from four published data sets of VNTR profiles from Albania, Iran, Morocco and Venezuela. In the first variant, the mutation rate increases linearly with respect to repeat numbers (linear model); in the second, the mutation rate is constant across repeat numbers (constant model). We find that under the constant model, the mean mutation rate per locus is (95% CI: ,)and under the linear model, the mean mutation rate per locus per repeat unit is (95% CI: ,). These new estimates represent a high rate of mutation at VNTR loci compared to previous estimates. To compare the two models we use posterior predictive checks to ascertain which of the two models is better able to reproduce the observed data. From this procedure we find that the linear model performs better than the constant model. The general framework we use allows the possibility of extending the analysis to more complex models in the future.
| Genetically typing the bacterium responsible for tuberculosis is useful for understanding the evolutionary and epidemiological characteristics of the disease. Typing methods based on variable number tandem repeat (VNTR) loci are increasingly being used. These loci, which are composed of repeated units, mutate by increasing or decreasing in the number of these repeats. Knowledge of the mutation rate of molecular markers facilitates the epidemiological interpretation of the observed genetic variation in a sample of bacterial isolates. Few studies have examined the rate of mutation at these markers and estimates to date have varied considerably. To address this problem we develop a stochastic model of evolution of these markers and then estimate their mutation rate using approximate Bayesian computation. We examine two alternative forms of the mutation process. The observed data are from four published data sets of tuberculosis bacterial isolates sampled in Albania, Iran, Morocco and Venezuela. We find that these markers have fairly high rates of mutation compared with estimates from previous studies.
| Mycobacterium tuberculosis, the bacterial pathogen that causes tuberculosis, latently infects one third of the world's population and is responsible for the highest mortality rate of any single bacterial pathogen [1]. Recent advances in genotyping techniques have increased our ability to discriminate among M. tuberculosis isolates, helping to shed light on the genetic diversity, demographics and evolution of this pathogen [2], [3]. For instance, Pepperell et al. [4], [5] suggested that the restricted diversity in this bacterial species is likely the result of population bottlenecks and founder effects. Genotyping or fingerprinting also refines our understanding of the epidemiological characteristics of the disease in a population, for example by revealing the extent of local transmission and factors associated with this transmission (e.g., [6]).
Frequently used methods for genetic fingerprinting of M. tuberculosis include restriction fragment length polymorphism typing based on mobility of the insertion sequence IS 6110 [7] and spoligotyping which exploits variation at the Direct Repeat or CRISPR locus [8]. More recently, a multilocus typing method based on variable numbers of tandem repeats (VNTR) has been developed for M. tuberculosis [9]–[11]. These loci are minisatellites, and are also known as mycobacterial interspersed repetitive units (MIRUs). We will refer to these as “VNTR loci”.
VNTR-based methods are increasing in importance and efforts are being made to standardise the loci used [9]. The larger the number of loci used, the greater the discrimination among isolates resulting in a large number of smaller clusters of identical profiles in a sample. The early standard of 5 locus VNTR typing lacked the discriminatory power of IS6110-typing but comparative studies have shown that using at least 12 loci can have comparable or better discrimination relative to IS6110 [12]–[14]. An advantage of using VNTR is that if the mutation rate is low there is the possibility of adding more loci to increase discriminatory power [10].
Inferences about transmission are sensitive to the degree of genetic clustering, which is a function of the mutation rate of the marker [15]. It is therefore important to have accurate estimates of the mutation rate of VNTR loci. Knowledge of the mutation rate of VNTR also allows calibration of the molecular clock to make inferences about the evolutionary history of M. tuberculosis, for instance, the time until the most recent common ancestor of a clade [3].
A standard model for the evolution of VNTR loci is the stepwise mutation model [16], [17], which has successfully been used to describe microsatellite evolution in eukaryotes (e.g. [18]). The stepwise mutation model has also been applied to VNTR evolution in M. tuberculosis [19], leading to estimates of the rate of mutation. Such estimates in the literature vary widely from per locus per year [19] to per locus per year [3] to – [20]. This wide variation in estimates has led to debate in the literature [21]–[24]. Taking a model-based approach can help to resolve this question. It allows our understanding of biological mechanisms underlying VNTR evolution to be incorporated into the analysis, while providing a natural framework for model validation and criticism. Similarly, examination of multiple data sets under the same models and methods could provide support or otherwise for resulting estimates.
In this study we estimate the mutation rate of VNTR markers by developing a stochastic stepwise mutation process of the evolution of genotypes through gains and losses of repeat numbers [16], [19] embedded in a model of disease transmission [25]. We consider and evaluate two alternative formulations of the stepwise mutation model under a Bayesian statistical framework, applying our methods to four geographically distinct data sets. Our study provides a posterior estimate of the VNTR mutation rate under an explicit model of evolution placed within an epidemiological context.
In the model of disease transmission we use, tracks the number of individuals who are susceptible to infection and tracks infectious individuals, where is time measured in years. For simplicity, we assume a population of fixed size . Let be the rate of transmission and be the rate of death or recovery. First consider a deterministic model where the dynamics are given by(1)We start the process with a single infected individual (). Define to be the basic reproductive ratio, that is, the number of cases resulting from a single infectious case in a wholly susceptible population. For this model, . The analytical solution of Equation (1) can be written as(2)The steady state of the infectious population is
We use this deterministic model as the basis for a continuous-time stochastic model that incorporates mutation at VNTR loci. The transition rates of this model, summarised in Table 1, are as follows: the rate of new infections is and the rate out of the infectious class from death or recovery is . An infection event increases by 1 while a death-or-recovery event decreases by 1. Each infection is associated with a bacterial genotype by which we mean the set of repeat states across all loci considered in a VNTR typing technique, determined for a particular isolate. Let be the number of individuals infected with bacterial genotype so thatwhere is the number of distinct genotypes in the population at time .
We apply the stepwise mutation model to describe VNTR mutation [16], [17], [19] in which an event results in a unit increase or decrease in the number of repeats at a locus. We define to be the mutation rate per infectious case for genotype so that the transition rate for mutation of genotype is . A mutation event results in either a new genotype, or a pre-existing genotype in the population (i.e., homoplasy). In the event of mutation to a new genotype, the number of individuals from the mutating genotype decreases by 1 and the number of individuals in the new class becomes 1. In the case of homoplasy, the number of individuals in the mutating genotype decreases by 1 while the number of individuals in the existing class increases by 1. In either case the total number of infected cases, , does not change.
We consider two alternative ways to specify VNTR mutation. In the first model, the mutation rate at a locus is proportional to the number of repeats at that locus. In this linear model, the per-locus mutation rate increases linearly with the number of repeats at the locus. In the second constant model, the mutation rate the per-locus mutation rate is constant and thus not dependent on repeat number. Defining to be the number of loci, to be the number of repeats at locus for genotype , and to be the rate of mutation at a locus with a single repeat, under the linear modelUnder the constant modelwhere is the per locus mutation rate and where the indicator function if is true and 0 otherwise. In both models the boundary condition is an absorbing state in that a locus with zero repeats cannot gain or lose repeats.
The process starts at time with a single infected individual and the population evolves until time . The initial individual has genotype given by , which we call the founding genotype. At time a sample of size is taken from the population. We simulate this process using the Gillespie exact algorithm [26] so that the time between events is distributed exponentially, with parameter , whereGiven an event, the probability of a specific outcome is proportional to the rate of that outcome, so thatGiven a mutation event, the probability of mutation in an individual with genotype isand given a mutation event in genotype , the probability that it occurs at locus under the linear model isand under the constant model isWe assume that given a mutation event at locus in genotype , the probability of repeat gain is equal to the probability of repeat loss, following [3], [19].
We implement a standard Bayesian analysis of model parameters using approximate Bayesian computation (ABC) [27]–[29]. ABC methods permit approximate Bayesian inference when numerical evaluation of the posterior distribution is either computationally prohibitive or not available, and have been successfully applied to problems in molecular epidemiology [30]–[34].
Intuitively, given a candidate parameter vector, , prior distribution and model likelihood with observed data , ABC methods proceed by generating an artificial dataset from the model and then reducing the dataset to a low dimensional vector of summary statistics, . If is similar to the same vector of statistics obtained from the observed data, , then could have credibly reproduced the observed data under the model. As such, the parameter vector is then retained as part of the approximate posterior, otherwise it is discarded. More precisely, the posterior obtained under ABC methods is given by(3)where is a standard smoothing kernel with scale parameter . As becomes small, the approximation (3) becomes increasingly accurate, although computational overheads increase. If the vector of summary statistics are informative for the model parameters, then this posterior distribution approximates the true posterior distribution so that . See e.g. [30], [31], [35], [36] for further description of ABC methods.
The parameter vector for the constant model above is where is the repeat structure of the founding genotype in the simulation. For the linear model we have . Except where this may cause confusion, we will refer to a non-model-specific parameter vector as .
Conditional on the parameter vector , and following simulation under the model, a sample of size individuals is drawn from the resulting population. Summary statistics, , are then computed, determined as quantities expected to be highly informative regarding the model parameters. Using lower case letters (e.g. ) to denote sample-based values of the population-level counterparts (e.g. ), the summary statistics include the number of distinct genotypes in the sample, , and the set of sample means of repeats at each locusfor , which is expected to contain information about the initial repeat numbers for some time after the founding case. Here, denotes the number of individuals in the sample with genotype , and denotes the within-sample number of repeats at locus for genotype . The final two statistics are based on the ANOVA decomposition given bywhere , from which and can be computed. These two statistics are expected to be informative about the mutation rate between and within loci. The complete vector of summary statistics is then given byTo complete the model specification, we set the parameter to , following [32], [37]. This death/recovery rate is the sum of the death rate due to tuberculosis, the death rate due to other causes, and the recovery rate from tuberculosis. We chose an informative prior distribution for based on the study of the basic reproductive value of tuberculosis by Blower et al. [38]. We use a distribution approximating the histogram in Figure 3a in reference [38] which has a mean of 5.16 and a standard deviation of 2.82, and in particular define the prior of to be a gamma distribution with a shape parameter of and a scale parameter of . The priors for , , and are uniform with wide ranges as shown in Table 2.
We examine the effectiveness of the ABC inference procedure by evaluating its ability to recover accurate estimates of the mutation rate based on data generated under the constant and linear models We simulated a population of individuals with loci, , , and considered a range of mutation rates under each model varying across orders of magnitude and . The number of repeats of the founding genotype were initialised as (determined as random draws from ), where denotes loci with repeat number . Based on a sample of size we generated data under each mutation rate value, and obtained weighted samples from the ABC posterior approximations (c.f. 3) using a population-based ABC algorithm, following [32], [39], [40]. The technical algorithmic details are given in Text S1.
The estimated posterior distributions of and using the simulated data are shown in Figure 1. These results indicate that mutation rates can generally be recovered accurately, with the true parameter values lying in regions of high posterior density close to the posterior mode, and with a clear location shift in the density with varying mutation rate. Higher precision can be attained by using a larger sample size, although already represents a sample larger than the real datasets used for this study (c.f. Table 3). In the ABC setting, posterior precision can also be improved by reducing the kernel scale parameter in (3) or by the inclusion of more summary statistics [30], [31], [35], [36], although each of these can substantially increase computational overheads. Improving the precision of posterior parameter estimates for given summary statistics is currently an area of active ABC research [41].
We selected recently published VNTR loci data sets from studies undertaken in four countries: Albania [42], Iran [43], Morocco [11] and Venezuela [44]. We chose data sets with a high number of isolates largely from the same clade, a high number of VNTR loci in the typing method, and relatively short periods of isolate collection. The data from Albania and Venezuela are based on 24-locus typing, and the data from Iran and Morocco are based on 15 and 12 loci respectively. A summary of these data are provided in Table 3, along with the incidence of tuberculosis for each country.
As an initial exploratory examination of these data, we computed gene diversity [45] (also known as virtual heterozygosity), for each locus in each data set. This statistic is given by where is the number of isolates with repeat size at locus . Figure 2 (left plots) shows the empirical cumulative distribution function of gene diversity across loci for each of the data sets. There is no obvious bimodality in these distributions. This feature is consistent with a common process generating diversity, compared to, for example, the potential bi- or multi-modality in the empirical cumulative distribution function arising from a multi-modal distribution of mutation rates. Similarly, plotting the proportion of VNTR states per locus per repeat (right plots of Figure 2) reveals that while some loci are more variable than others, there is no obvious separation between loci exhibiting high and low variation.
Figure 3 shows the marginal posterior distribution of the mutation rate of VNTR loci for each of the four data sets analysed. In the case of the linear model we also show (middle panel of Figure 3) the posterior of , the per-locus mutation rate at repeat size 1 scaled by the average repeat number of each dataset to provide estimates of the mean per-locus mutation rate in a population with the same distribution of repeats as found in each sample. The posterior means of the mutation rate under the two models, along with 95% central credibility intervals are given in Table 4. The mean per-locus mutation rate at a locus with a single repeat from the four data sets under the linear model is , and under the constant model the mean per-locus rate is . Note that the prior distributions of the mutation parameters are uniform on a logarithmic (base 10) scale, and so Figure 3 displays the posterior distributions on this scale.
To evaluate the suitability of the constant and linear models to describe the observed data, we follow [36], [46], [47] and implement posterior predictive model checks. This approach examines the predictive distribution of specified validation statistics (based on data-generation under the fitted models) expected to be informative about various model aspects. Comparing the predictive distribution of these statistics with the same statistics derived from the observed data, enables some degree of discrimination between models. To avoid confusing model fitting with model assessment, these statistics should be different from those used in the ABC model fitting process.
Unlike the constant model, the mutation rate increases with repeat number under the linear model, and so we expect variation in repeat numbers to increase with repeat numbers. Our model assessment statistics aim to capture these differences from the data. Specifically, we focus on measures of the spread of repeats over the loci. Definingwhere , andwhere , and indexes loci as before, we consider the maximum (over loci) range (), the difference between maximum and minimum range (), maximum variance () and the difference between maximum and minimum variance ().
Under the linear model, the distributions of these statistics are expected to be shifted to higher values compared to the constant model. We also fit a simple linear regression to each data set with the standard deviation of repeat number at a locus as the response variable and the mean repeat number at a locus as the predictor variable. Based on this fit, we considerwhere is the fitted standard deviation in repeats at a locus with a mean repeat number of one. These statistics are expected to be informative in that the slope should be positive under the linear model and near zero under the constant value, and the intercept should be low under the linear model and high under the constant model.
Figure 4 displays the predictive distributions of versus under both models. The observed data statistics are indicated by a cross (). If the cross does not lie within the body of the predictive distribution, this suggests that the model and data are inconsistent with respect to aspects of the data captured by these statistics. The lower four panels present these diagnostics for artificial data generated under both models. The linear data (lower images) can be seen to be inconsistent with the constant model, but consistent with the linear model. The constant data (middle images) appear to be consistent with both models. As such, these diagnostics are able to reject the constant model when the data is generated by the linear model. In terms of the actual empirical data, the top plots in Figure 4 are based on the data from Albania. Clearly, the constant model is insufficient to describe the variation in repeat numbers inherent in the data. The linear model is better able to account for the observed pattern of repeat variation, although it is still imperfect. The posterior predictive distributions using the data sets from the other three countries were very similar to those of the Albanian data set (not shown).
The question of whether the linear model is adequate is examined further in Figure 5 which shows a posterior predictive check of versus under the linear model for each of the analysed data sets. In each case, the observed data lie on the periphery of the predictive densities. Although the linear model is partially able to reproduce these statistics, this analysis shows that there is room for improvement.
We have analysed VNTR data from four tuberculosis studies using a model combining marker mutation and disease transmission processes, within a Bayesian framework. Our analysis shows that the VNTR mutation rate is likely to be relatively high – the posterior mean is higher than some previous estimates obtained in the literature [3], [19] and closer to more recent estimates [20]. The four data sets, which are from different geographic regions, yielded very similar estimates. Such agreement of estimates is expected if there is a common mechanism of mutation across data sets.
Previous work by two of us [20] used standard equilibrium results of the infinite alleles model to describe mutation at multiple VNTR loci, and used estimates of other markers (IS6110 and spoligotyping) to calibrate the VNTR rates. That population genetic approach did not account for evolution of VNTRs as a stepwise mutation process. It therefore did not account for homoplasy, though this problem is mitigated by the inclusion of multiple VNTR loci. Further, the underlying dynamics did not include any epidemiological details. Nevertheless, it allowed us to analyse a large number of data sets in the literature to provide a ballpark estimate of VNTR mutation rates. In contrast to that and other prior work, here we used a model that explicitly and simultaneously accounts for the mutation process of the marker and the disease dynamics, and we explored two alternative models of mutation. In addition, the stepwise mutation model used here allows mutation events to re-generate existing VNTR profiles, thereby accounting for homoplasy [48].
In the debate over the magnitude of VNTR mutation rates [3], [21]–[24] it has been noted that if loci are classified as less variable and more variable, then lower values would be estimated from the former category of loci. This raises the question of whether classification of loci into two categories of rates is supported by an underlying bimodal distribution whose modes correspond to low and high levels of polymorphism. In examining gene diversity, which is a measure of polymorphism, across loci in each data set (Figure 2) we did not observe any obvious break separating less and more variable loci. We have therefore pooled all loci and obtained an estimate of the rate of an arbitrary locus, rather than for a subset of slow or fast evolving loci. If hypermutable VNTR loci exist and are excluded from estimation procedures, using the remaining loci would clearly yield a lower mutation rate.
Our use of the linear model is a step towards resolving this issue. The linear relationship by which more units of a repeat are more prone to mutation naturally creates variation in rates. In fact, in assessing the ability of each of our two mutation models to describe the data, we found that the linear model performs better than the constant model (Figure 4). We note that the average mutation rate under the linear model was estimated to be very close to the mutation rate in the constant model; in this sense our analysis is robust to the exact form of the mutation model.
Despite the linear model outperforming the constant model, a posterior predictive goodness-of-fit analysis revealed some evidence that the linear model did not fit the data perfectly (Figure 5). While previous studies of eukaryote minisatellites agree with a linear relationship between repeat number and mutation rate [49], some studies of eukaryote microsatellites indicate a more complex relationship between repeat number and mutation rate [50]–[53]. We investigated a third model in which the mutation rate increases exponentially with repeat number, but the results are very similar to those of the linear model (Figure S3 in Text S1). Future work might adopt a per locus mutation rate that grows non-linearly with repeat number. A drawback of this possibility would be the added complexity and dimensionality of the model with the need to estimate further parameters in a framework that is already computationally intensive. An alternative approach might be to construct a hierarchical Bayesian model of mutation rates in which each locus is associated with its own rate according to some distribution, akin to the analysis of Bazin et al. [54].
We have used a simple model to avoid overfitting the data. However, it is possible to extend the model in future studies to incorporate further complexity and realism. One such detail is the reactivation of latent infection, which could be described by a susceptible-exposed-infected (SEI) model in which a proportion of cases progress directly to disease [38]. We performed preliminary simulations from a stochastic version of such a model (details in Text S1). We consider the number of distinct genotypes since this is one of the statistics we use in the inference and it is known to be informative for mutation rate in similar models [55], [56]. Figure S2 in Text S1 shows how the number of distinct genotypes in a sample varies with the mutation rate under both models. The latent reactivation model was able to generate statistics close to the observed statistic. The points in the region of the observed statistic are near the posterior density generated under the original model. While this is suggestive that a latency model would produce similar estimates, a full Bayesian analysis would be required to address this issue. The lack of latency is a limitation of our study which should be addressed in future research.
Migration is another factor which a more realistic multi-deme population model might incorporate. The interplay between migration and mutation may affect the resulting estimates of the mutation rate. For example, migration from regions with genetically very different clades of M. tuberculosis occurs at a high rate would lead to over-estimation of the mutation rate. Our approach based on the approximate Bayesian computation framework makes future directions such as this and those relating to the mutation process feasible.
|
10.1371/journal.ppat.1002471 | (Homo)glutathione Deficiency Impairs Root-knot Nematode Development in Medicago truncatula | Root-knot nematodes (RKN) are obligatory plant parasitic worms that establish and maintain an intimate relationship with their host plants. During a compatible interaction, RKN induce the redifferentiation of root cells into multinucleate and hypertrophied giant cells essential for nematode growth and reproduction. These metabolically active feeding cells constitute the exclusive source of nutrients for the nematode. Detailed analysis of glutathione (GSH) and homoglutathione (hGSH) metabolism demonstrated the importance of these compounds for the success of nematode infection in Medicago truncatula. We reported quantification of GSH and hGSH and gene expression analysis showing that (h)GSH metabolism in neoformed gall organs differs from that in uninfected roots. Depletion of (h)GSH content impaired nematode egg mass formation and modified the sex ratio. In addition, gene expression and metabolomic analyses showed a substantial modification of starch and γ-aminobutyrate metabolism and of malate and glucose content in (h)GSH-depleted galls. Interestingly, these modifications did not occur in (h)GSH-depleted roots. These various results suggest that (h)GSH have a key role in the regulation of giant cell metabolism. The discovery of these specific plant regulatory elements could lead to the development of new pest management strategies against nematodes.
| Parasitic nematodes are microscopic worms that cause major diseases of plants, animals and humans. During compatible interactions, root-knot nematodes (RKN) induce the formation of galls in which redifferentiation of root cells into multinucleate and hypertrophied giant cells is essential for nematode growth and reproduction. The importance of glutathione (GSH), a major antioxidant molecule involved in plant development, in plant microbe interaction and in abiotic stress response, was analyzed during the plant-RKN interaction. Our analyses demonstrated that the gall development and functioning are characterized by an adapted GSH metabolism and that depletion of GSH content impairs nematode reproduction and modified sex ratio. This phenotype is linked to specific modifications of carbon metabolism which do not occur in uninfected roots indicating a peculiar metabolism of this neoformed organ. This first metabolomic analysis during the plant-RKN interaction highlights the regulatory role played by GSH in this pathogenic interaction and completes our vision of the role of GSH during plant-pathogen interactions. RKN sex ratio modification has previously been observed under unfavorable nematode feeding conditions suggesting that the GSH-redox system could be a general sensor of gall fitness in natural conditions.
| Glutathione (GSH) is a tripeptide, γ-glutamyl-cysteinyl-glycine, present in a wide range of organisms. It is a low molecular weight thiol which in plants is involved in antioxidant defense, detoxification of xenobiotics and tolerance to abiotic and biotic stresses [1]. GSH regulates the expression of stress defense genes and is involved in plant resistance to oomycete and bacterial pathogens and insect herbivores [2]–[4]. GSH is also involved in organ development and its role in seed maturation and root and leaf growth has been established [5]–[7]. In certain legumes, a GSH homolog, homoglutathione (hGSH), is also present instead of, or in addition to, GSH [8]–[10]. The synthesis of GSH is a two-step process. In the first step, γ-glutamylcysteine synthetase (γ-ECS) produces the dipeptide γ-glutamylcysteine (γ-EC) from L-glutamic acid and L-cysteine and regulates the accumulation of GSH and hGSH [(h)GSH]. The formation of GSH and hGSH is determined by the substrate specificity of the enzyme catalyzing the second step. Glutathione synthetase (GSHS) catalyses the addition of glycine to γ-EC, whereas homoglutathione synthetase (hGSHS) catalyses the addition of β-alanine to γ-EC.
In the model legume Medicago truncatula, we have shown that (h)GSH deficiency alters the nitrogen fixing symbiotic interaction and reduces the formation of root nodules [11]. The transcriptomic response of (h)GSH-deficient plants to Sinorhizobium meliloti infection showed a downregulation of genes involved in meristem formation and an increased expression of several genes involved in the early plant defense reaction against abiotic or biotic stresses [12]. Thus (h)GSH may regulate both nodule neoformation and the plant defense response during symbiosis [12].
Plant-parasitic nematodes that infect M. truncatula and other legumes have emerged as models for studying the molecular dialogue during plant-nematode interactions and investigating whether beneficial plant symbionts and biotrophic pathogens induce distinct or overlapping regulatory pathways [13]–[17]. Root-knot nematodes (RKN, Meloidogyne spp.) are obligate root pathogens that interact with their hosts in a remarkable manner. During a compatible interaction, infective second stage RKN juvenile (J2) migrate intercellularly towards the vascular cylinder and induce the redifferentiation of root cells into specialized nematode feeding cells named giant cells (GCs). GCs are hypertrophied and multinucleate. They are the result of successive nuclear division without cell division and isotropic growth [18]. Mature GCs are metabolically very active, and act as transfer cells between vascular tissues and RKN. They are the sole source of nutrients for the feeding nematode and are thus essential for nematode growth and development [19]. Hyperplasia of neighboring cells (NCs) leads to the gall, the characteristic symptom of RKN infection. Once sedentarized, J2 molt three times to reach the adult stage. The reproduction of M. incognita is parthenogenetic: males migrate from the root and are not required for reproduction whereas the pear-shaped females produce and extrude eggs in a gelatinous matrix onto the root surface.
The formation of both nodule and gall requires root cell dedifferentiation and modification of their cell cycle [20], [21]. Moreover, both nematodes and rhizobia seem to actively modulate the host plant defense, so as to allow the compatible interaction [22], [23]. The modifications to the plant defense and organogenesis observed in these plant-microbe interactions led us to analyze (h)GSH metabolism in galls. We studied the involvement of these tripeptides in the M. incognita development cycle in M. truncatula and tested for modifications of gall metabolism under (h)GSH deficiency.
The development cycle of M. incognita in M. truncatula is 6–7 weeks long. We analyzed (h)GSH metabolism during the RKN life cycle. First, the expression of M. truncatula γECS, GSHS and hGSHS genes was evaluated by qRT-PCR (Figure 1A). The expression of γECS and hGSHS was significantly lower in galls than in uninfected roots from 2 wpi (Figure 1B and D). In contrast, no significant difference in the expression of GSHS was observed between galls and uninfected roots (Figure 1C). We tested whether the changes in the expression of the genes involved in (h)GSH synthesis correlated with the GSH and hGSH pools (Figure 2A). The quantification of (h)GSH pools by HPLC analysis (Figure 2) showed that hGSH was significantly less abundant in galls than in uninfected roots during the first two wpi corresponding to the period of GC formation (Figure 2A). By contrast, the GSH pool was significantly larger in galls than in uninfected roots 3 and 5 weeks post infection (wpi) with 4 fold-higher level in mature galls 5 wpi (Figure 2B).
To assess the involvement of (h)GSH in the plant-nematode interaction, we analyzed the production of egg masses by the nematode in (h)GSH-depleted plants. The plant (h)GSH pool was depleted pharmacologically with L-buthionine-[S–R]-sulfoximine (BSO), a specific inhibitor of (h)GSH synthesis. The effect of BSO treatment on nematode fitness was analyzed by treatment with 1 mM BSO supplemented with 1% resorcinol, a compound shown to induce solute uptake in nematodes [24]. No difference in nematode reproduction was observed between BSO-treated nematodes and controls (Figure S1). Treatment with 0.1 mM BSO applied one week before infection led to an 85% reduction of total (h)GSH in roots as previously described [11]. The primary root of each control and (h)GSH-depleted plants was then inoculated with M. incognita and the production of egg masses at 7 wpi was used as a measure of nematode reproduction efficiency (Figure 3). A mean of 23 egg masses was produced in control plants at 7 wpi (Figure 3A). BSO treatment led to a 75% reduction in the (h)GSH content and a 95% diminution of egg mass production in (h)GSH-depleted plants relative to control plants.
To verify that this reduction in egg mass production was related to the decrease in (h)GSH content and not to another secondary effect of BSO, RNAi was used to deplete (h)GSH in M. truncatula roots of composite plants [25]. The transgenic roots carrying the γecs-RNAi construct were compared with transgenic roots expressing an RNAi construct for the Green Fluorescent Protein (GFP) as a control. The number of egg masses and the (h)GSH content of composite plants were analyzed for each individual root at 7 wpi (Figure 3B). Both (h)GSH content and the number of egg masses were significantly lower in the γecs-RNAi roots than the control gfp-RNAi plants. These experiments demonstrate that the reduction in (h)GSH content in galls correlates with a decrease in nematode egg mass production.
As pharmacological and genetic (h)GSH depletion resulted in similarly impaired nematode reproduction, we mainly used BSO treatment to produce sufficient amounts of homogeneous material for further experiments. However, the major results of (h)GSH depletion were confirmed on genetically-modified material. To determine whether the reduction of egg masses was linked to a delay or an arrest in nematode development, galls were dissected at 4 and 7 wpi and the number of nematodes at each developmental stage (juvenile, male and female) was counted (Figure 4). At 4 wpi, an average number of 27 nematodes were detected per control plant whereas only 18 nematodes were observed in each (h)GSH-depleted root, suggesting that nematode infection is affected by (h)GSH depletion. Thirteen of the nematodes were at the female stage in control galls. In contrast, no female was observed in (h)GSH-depleted galls: almost all nematodes were at the juvenile stage and few males were identified in (h)GSH-depleted galls (Figure 4). Under genetic (h)GSH depletion, a significant lower number of nematodes at the female stage was also observed in the γecs-RNAi roots than in the gfp-RNAi control ones (Figure S2A). Moreover, the proportion of males was also significantly increased in γecs-RNAi roots (Figure S2B). At 7 wpi, the proportion of females was much lower in (h)GSH-depleted galls (23%) than in control galls (90%) and more than half of the nematodes were still juvenile; (h)GSH-depleted galls also contained a large proportion of males (23% vs 0.1% in controls) (Figure 4). Thus, the (h)GSH depletion substantially reduced the number of females per gall (from 27 for controls to 2.3 for (h)GSH-depleted galls) consistent with its effects on egg mass formation. In addition, the numbers of nematodes in (h)GSH-depleted galls at 7 wpi shows that juveniles present at 4 wpi mostly molted into males, or were eliminated from the gall.
The dissection analysis showed that nematode development was impaired at four wpi as no female was detected in (h)GSH-depleted plants at this time point. To test the effect of (h)GSH deficiency on the formation of GC, molecular and cellular analyses were performed at 2 wpi (Figure 5). First, the expression of marker genes involved in GC development was evaluated by qRT-PCR (Figure 5A). The establishment of RKN infestation is associated with the suppression of plant defense responses and the induction of genes encoding proteins involved in both cell wall and DNA metabolism [22]. We therefore studied the expression of the defense-related genes, Pathogenesis-Related 1 protein and patatin, of expansin and of histone H3. During the interaction between M. truncatula and M. incognita, the expression of both Pathogenesis-Related 1 protein and patatin was significantly weaker in galls at 2 wpi than in uninfected controls; however, the expression of both expansin and histone was higher in galls than controls. No significant difference was observed between the expression of these four marker genes in (h)GSH-depleted galls and that in the controls. GC morphology was analyzed to detect potential morphological effects (Figures 5B and 5C). Microscopic analysis at 2 and 3 wpi revealed GCs with dense cytoplasm, multiple small vacuoles and nuclei observed in both (h)GSH-depleted galls and control galls (Figures 5B, 5C and Figure S3). Thus, there was no significant molecular or cellular difference between (h)GSH-depleted galls and control galls strongly suggesting that GC ontogenesis was unaffected by (h)GSH depletion.
Depletion of (h)GSH was thus associated with impaired nematode development and in particular the absence of females at 4 wpi. We performed a metabolomic analysis of control and (h)GSH-depleted roots and galls at 3 wpi to assess primary metabolic effects. We investigated the major compounds of the primary metabolism in roots and galls through an untargeted proton Nuclear Magnetic Resonance (1H-NMR) analysis approach. Eighteen primary and secondary metabolites were identified in the 1H-NMR spectra of each extract (Figure S4) after peak assignment using chemical shift reported in the literature and metabolomic databases, with assistance from 2D NMR and/or by spiking samples with commercial compounds. Eighteen additional metabolites remained unidentified. Clear differences between uninfected roots and galls were obvious on visual inspection of spectra and were confirmed by quantification of metabolites. Eighteen of the 37 quantified metabolites were related to sugar, organic acid and amino acid metabolism (Table 1). Starch, an important sugar reservoir in nematode-induced syncytia [26], was assayed enzymatically. Principal component analysis (PCA) was used to provide an overview of sample grouping and metabolic differences between uninfected roots and galls: we used a matrix containing the data for the 18 identified and quantified polar metabolites plus starch (Figure 6). The first principal component (PC1) of the score plots (Figure 6A), explaining 56% of the total variability, clearly separated galls (on the negative side) from uninfected roots (on the positive side). The loading analysis (Figure 6B) suggested that the major metabolites contributing to this separation along PC1 were sucrose, trehalose, malate, fumarate, six amino acids (aspartate (Asp), glutamate (Glu), isoleucine (Ile), phenylalanine (Phe), tyrosine (Tyr) and valine (Val)) and trigonelline on the negative side and glyoxylate on the positive side. PCA score plot also showed that the second principal component (PC2), explaining 17% of the total variability, clearly separated (h)GSH-depleted (on the negative side) from control galls (on the positive side) (Figure 6A). Observation of PC2 loading (Figure 6B) suggested that this separation along PC2 mainly involved γ-aminobutyrate (GABA), Ile, Val, Glu, Asp and Asn on the negative side and glucose, starch and proline-betaine on the positive side.
The PCA was confirmed by univariate analyses of metabolite data (Table 1). Relative to control roots, galls exhibited a significantly higher content of starch, sugars (sucrose, glucose), organic acids (malate, fumarate) and amino acids (Phe, Tyr, Val, Glu, Asp). However, the increase in amounts of these metabolites was not related to a similar modification of the expression of primary metabolism genes, the expression of which was maintained (Sucrose synthase 1, ADP-glucose pyrophosphorylase, starch synthase, α 1–4 glucan phosphorylase, pyruvate kinase) or even decreased (cell wall-invertase, mitochondrial malate dehydrogenase, malate synthase, phosphoenolpyruvate carboxylase, phosphoenolpyruvate carboxykinase) in galls (Figure 7). Interestingly, the asparagine (Asn) content of control galls was significantly lower than that of control roots (Table 1). This was related to a significant decrease in expression of the asparagine synthetase and an increase in that of asparaginase (Figure 7). Proline-betaine, production of which in plants is related to the water stress response, accumulated significantly more in galls than control roots (Table 1). Trigonelline, another aminated compound related to secondary metabolism and potentially involved in salt-stress response [27], was also more abundant in galls than control roots (Table 1). Finally, trehalose accumulation may also be related to a modification of the osmotic status in galls [28]. We used these metabolite and gene expression data to establish a metabolic pathway scaffold (sucrose and starch metabolism, glycolysis and the tricarboxylic cycle connected branch points toward organic acid and amino acid synthesis) highlighting the significant differences observed between roots and galls (Figure 8A). Generally, there is little correlation between increased accumulation of quantified metabolites and the expression of the associated primary metabolism genes.
Depletion of (h)GSH modified the metabolism of roots and galls in different ways. Most metabolites in roots were not significantly modified by (h)GSH-depletion. PCA analysis of the 18 identified and quantified polar metabolites plus starch showed that (h)GSH-depleted and control uninfected roots had a similar composition of polar metabolites (Figure 6A). Significant variations were observed only for the hydrophobic amino acids Ile, Phe and Tyr, for trigonelline and for starch (Table 1). Indeed, the starch content was decreased 3-fold and starch synthase was significantly down regulated by (h)GSH depletion, suggesting that starch metabolism in roots is regulated by (h)GSH content or metabolism. Unlike the findings for uninfected roots, (h)GSH depletion had substantial effects on the metabolism of galls (Figure 6). The content of nine metabolites differed significantly between (h)GSH-depleted galls and control galls (Table 1). Starch and glucose contents were significantly lower in (h)GSH-depleted galls, whereas that of sucrose was not significantly different. With the exception of malate, the abundance of which was significantly decreased, the organic acid content of galls was not significantly affected by (h)GSH depletion. In the γecs-RNAi galls, starch, glucose and malate contents were also significantly lower than in the control gfp-RNAi galls (Table S1). Consistent with these findings, the relative expression in galls of most genes involved in the metabolism of sugars and organic acids was not significantly modified by (h)GSH depletion. The amino acids content was slightly increased by (h)GSH depletion, with the exception of Asn which was increased two-fold to the range of that found in roots (Table 1). This increase in Asn content was associated with a significant decrease in asparaginase gene expression and a two-fold increase in asparagine synthetase gene expression associated with (h)GSH depletion (Figure 7). A similar trend was also observed for proline-betaine, the content of which in (h)GSH-depleted galls was close to that in uninfected roots. Thus, our data indicate that (h)GSH depletion partially reversed the effect of nematode infection on starch, glucose and Asn metabolism, and on proline-betaine accumulation. Interestingly, GABA, a compound associated with biotic and abiotic stresses [29], was markedly more abundant in (h)GSH-depleted than control galls. The significant differences between control and (h)GSH-depleted galls are summarized in a metabolic pathway scaffold (Figure 8B). As observed for the comparison between galls and uninfected roots, differences in metabolite contents were more marked than the differences between gene expression levels. This implies post-transcriptional regulatory mechanisms, such as post-translational modifications or metabolic controls, in the metabolic modifications associated with (h)GSH depletion of galls.
(h)GSH play a major role in plant development and plant adaptation to biotic and abiotic stresses [30]. A threshold (h)GSH concentration is necessary for plant and organ development [5]–[7], [31], [32]. (h)GSH is also involved in plant responses to pathogens [2], [3] and to symbiotic microorganisms [11]. Here, we report an analysis of the involvement and roles of (h)GSH metabolism in the M. truncatula-M. incognita interaction.
M. truncatula roots contain two low molecular weight thiols, GSH and hGSH [9]. The (h)GSH content is significantly higher in galls than in roots at later stages of gall functioning. Surprisingly, γECS transcript level was lower in galls than in roots whereas this gene should regulate the level of (h)GSH. The post-transcriptional regulation of γECS [33], [34] may explain the discrepancy between γECS transcript level and GSH accumulation. (h)GSH accumulation has been observed in several developmental conditions involving endoreduplication and enhanced metabolic capacity such as in symbiotic nitrogen fixation [9] and in trichomes [35], both physiological modifications also occurring during gall formation and function [18], [22], [36]. In addition, the accumulation of GSH in galls may be caused by the nematode, as a GSHS has been identified amongst the proteins secreted by M. incognita [37]. Finally, (h)GSH accumulation is also associated with the nematode secreting multiple redox- and (h)GSH-regulated proteins, including thioredoxin, glutathione peroxidases and glutathione-S-transferases, required for the completion of nematode life cycle [37], [38]. Indeed, the control of the plant cell redox status through the modification of the (h)GSH content may be a key regulator of the GC effectiveness.
We show here that root (h)GSH deficiency strongly impairs nematode reproduction. This reduction of egg masses seems to be largely a consequence of the nematode sex ratio in galls. At 7 wpi, galls in (h)GSH-depleted plants harbored only one third as many nematodes as controls, suggesting that most juveniles developed into males and therefore migrated from the gall to soil such that they were not found in the gall by dissection. An hypothesis might be that BSO would be involved in direct impairment of GSH production in nematodes [39], [40] and thus modify their development and egg mass production. Analysis of M. incognita genome using Caenorhabditis elegans γECS and GSHS sequences shows that the GSH biosynthesis genes are present in M. incognita. Moreover, HPLC analysis shows that GSH is produced in M. incognita J2 larvae (unpublished data). The effect of GSH depletion on M. incognita development could not be directly tested as it is a plant obligatory parasite. However, data provided on WormBase (http://www.wormbase.org) showed that GSH does not play a major role in both development and health in C. Elegans. GSH depletion induced by γECS-RNAi or gene deletion is not larval or embryonic lethal and does not induce slow growth and female sterility [41]–[43]. Finally, the lower egg mass production, the modifications of the nematode sex-ratio and metabolite contents observed in both BSO-treated plants and transgenic roots expressing a plant specific γECS-RNAi construct show that these modifications are not linked to direct impairment of nematode function by BSO.
During symbiosis between M. truncatula and S. meliloti, the (h)GSH depletion reduces the formation of nodule meristems [11]. Transcriptomic analysis evidences the involvement of (h)GSH regulation both in plant development and defense responses [12]. In contrast, under similar conditions, the development of the feeding site was not significantly affected and the expression of defense-related and development-related genes was not modified. Therefore, M. incognita is able to manipulate plant metabolism under (h)GSH depletion to avoid the defense and developmental phenotype observed during the establishment of nitrogen-fixing symbiosis.
Root and gall metabolomic profiling showed that most of the analyzed metabolites were significantly more abundant in galls than in uninfected roots. These modifications, and the analysis of the expression of numerous genes involved in primary metabolism, indicate that the gall metabolism differs substantially from that in uninfected roots. One of the striking differences concerning general metabolite accumulation is the significantly lower Asn content in galls than in roots. This, and the associated upregulation of asparaginase and the down regulation of asparagine synthase, shows that nitrogen metabolism is modified in galls. Asn is the major nitrogen transporting compound in temperate legumes such as Medicago [44], [45]. The primary site of Asn synthesis is the root and it follows that, through loading into the xylem, Asn is the principal nitrogen source for amino acids and protein synthesis in leaves. Thus, the decrease in Asn content upon nematode infection is likely to result in nitrogen deprivation for the plant. This metabolic modification may thus reduce nitrogen supply to leaves and increase carbon and nitrogen accumulation in galls. Our findings show that gall metabolism involves the fine-tuning of metabolism involving both the up regulation of some metabolic pathways and the down regulation of others, so as to enhance nutriment availability for the nematodes.
Analysis of metabolite contents shows that (h)GSH depletion significantly affect gall metabolism. A significant difference in metabolite content between control and (h)GSH-depleted galls was detected for half of the metabolites quantified. In contrast, (h)GSH depletion did not significantly modify the content of the major primary metabolites in uninfected roots. This result is in agreement with our previous findings that a 85% depletion of (h)GSH does not significantly affect root growth [11]. The metabolic modifications observed in (h)GSH-depleted galls include a significant reduction of malate (40%), glucose (60%) and starch (84%). Starch accumulation during the interaction between A. thaliana and the parasitic nematode Heterodera schachtii is crucial for the nematode infection and development. It may serve as long- and short-term carbohydrate storage for the feeding needs of the parasites [26]. Glucose and malate are likely substrates and probably essential for nematode nutrition. Thus, the diminution of these three metabolites under (h)GSH depletion may impair nematode carbohydrate nutrition. The development of M. incognita juveniles into males rather than females has previously been observed under unfavorable nematode feeding conditions such as low concentrations of sucrose in the growth medium, defoliation and complete removal of the host plant above-ground parts [46]–[48]. These conditions also trigger carbon starvation of the galls. Carbohydrate nutrition deficiency has been also involved in the modification of sex ratio and development of cyst nematode [49]. The development of M. incognita juveniles into males rather than females in (h)GSH-depleted galls is similar to that observed during carbohydrate deficiency. This is consistent with (h)GSH being involved in the modulation of nematode differentiation through regulation of gall carbon metabolism.
Interestingly, GABA was specifically detected in (h)GSH-depleted galls. In plants, GABA accumulates in response to abiotic and biotic stresses [50]. During biotic stress induced by invertebrate pests, GABA accumulation in plant tissues reduces the feeding capacity of the pests [51]. Strikingly, the reproduction of Meloidogyne hapla is affected by GABA accumulation: egg mass production by M. hapla infecting transgenic plants accumulating GABA is lower than that by the pests infecting control plants [52]. Thus, the accumulation of GABA in (h)GSH-depleted galls may also contribute to the altered nematode reproduction.
The substantial primary metabolite modifications in (h)GSH-depleted galls with reference to control galls were not associated with corresponding modification in the expression of primary metabolism genes. This suggests that (h)GSH regulates gall metabolism at levels other than transcriptional. Redox state and GSH affect the function of many enzymes through post-translational modifications such as disulfide bond reduction and cysteine glutathionylation [53]. For instance, thioredoxins and glutaredoxins, which are involved in the formation/reduction of disulfide bonds between proteins, have been implicated in the regulation of chloroplast metabolism [54], [55]. ADP glucose pyrophosphorylase, a key enzyme in the biosynthesis of starch was also shown to be redox regulated [56], [57]. Cysteine glutathionylation is an important regulatory mechanism of photosynthetic metabolism [58]. More generally, in vivo control of many glycolytic and/or TCA cycle enzymes by disulfide-dithiol interconversions (NAD-dependent GAPDH, citrate synthase, PPi-dependent phosphofructokinase, PEPC kinase, etc) has been reported in plants [59]. Thus, a redox-based control of the gall metabolism by (h)GSH may be proposed to explain our results.
We cannot exclude that GSH may also be used as a nutrient by the nematode as the GSH content was strongly increased in mature galls compared to roots. However, the impact of (h)GSH depletion on gall metabolism is not in favour of a trophic role for (h)GSH. Moreover, during nitrogen fixing symbiosis in which GSH is not used as nutrient to feed the bacteroids, modifications of the (h)GSH content affects the nitrogen-fixing capacity of the nodule also showing the regulatory role of glutathione in this interaction [60].
In conclusion, we report that (h)GSH metabolism differs between galls and uninfected roots. A deficiency in (h)GSH impairs nematode reproduction by mainly altering its sex determination. This alteration in sex ratio is associated with modifications in the gall metabolism under (h)GSH depletion which have been shown to impair nematode development. Thus, we reveal a completely new role of (h)GSH in this biotrophic interaction. Interestingly, these modifications in metabolite content do not seem to occur in (h)GSH-depleted roots suggesting that (h)GSH depletion provokes metabolism modifications specific to the gall. Therefore, the reduction of (h)GSH availability in galls is a potentially useful strategy for pest management.
M. truncatula ecotype A17 was used for all the experiments. Sterilized seeds were germinated for 3 days onto 0.4% agar at 14°C. Seedlings were plated onto modified Fahraeus medium with 2 mM nitrogen [25] with 1.4% agar and grown for 7 days before infection. Plants were germinated in the presence or absence of 0.1 mM L-buthionine sulfoximine (BSO). For nematode infection, 100 surface-sterilized freshly hatched M. incognita J2 larvae were added on each one week old seedling as previously described [13]. One infection per plant was performed on the primary root. For BSO treatment, nematodes were incubated for 4 hours in M9 buffer (43.6 mM Na2HPO4, 22 mM KH2PO4, 2.1 mM NaCl, 4.7 mM NH4Cl) with 1% resorcinol and 1 mM BSO and with 1% resorcinol as control. The gall corresponds to one infection point and contains multiple GCs. After infection, plants were grown 3 weeks onto Fahraeus medium with 2 mM nitrogen in the presence or absence of BSO (0.1 mM) in a growth chamber with a day temperature of 23°C and night temperature of 20°C and with a photo-period of 16 h. Then, plants were transferred in soil mixture (30% vermiculite-70% fine gravel) until the end of the experiment. As reference samples, uninfected, primary root fragments of similar age were collected from seedlings grown under the same conditions. For gall dissection, galls were digested in a mixture of 30% Pectinex (Novozymes, Bagsvaerd, Denmark) and 15% Celluclast BG (Novozymes, Bagsvaerd, Denmark) for 12 h, dissected and nematode development stages were analyzed under a stereomicroscope. For metabolite and gene expression analyses, biological samples of galls and roots were harvested at different time points post-infection, frozen and ground in liquid nitrogen and stored at −80°C. One biological sample was issued from 20 galls or roots from 20 plants.
Thiols were extracted with perchloric acid, derivatized with monobromobimane, and quantified after separation on reverse-phase HPLC as described previously [61]. Commercial GSH (Sigma, St. Quentin, France) and γ-EC (Promochem, Molsheim, France) were used as standards. The hGSH used as a standard was synthesized by Neosystem (Strasbourg, France).
Total RNA of galls and uninfected root fragments were reverse-transcribed using the OmniScript cDNA Synthesis Kit (Qiagen, Courtaboeuf, France). Quantitative PCR reactions were performed using a DNA Engine Opticon 2 Continuous Fluorescence Detection system (MJ Research, Waltham, USA) and a qPCR MasterMix Plus for SYBR green I (Eurogentec, Angers, France). In each reaction, 5 µl of 100 fold-diluted cDNA and 0.3 µM primer (sequences used are described in Table S2) were used. The PCR conditions were 50°C for 5 min, 95°C for 10 min, followed by 40 cycles of 95°C for 30 s, 60°C for 1 min. Each reaction was performed in triplicate and the results represented the mean of three independent biological experiments. The specificity of the amplification was confirmed by a single peak in a dissociation curve at the end of the PCR reaction. Data were quantified by using Opticon Monitor 2 (MJ Research, Waltham, USA) and normalized with the 2−ΔΔCT method [62]. Two constitutively expressed genes Mtc27 (TC106535) and 40S Ribosomal Protein S8 (TC100533) were the endogenous controls [63]. The use of these housekeeping genes were validated by using the GeNorm VBA applet for MS Excel which determines the most stable housekeeping genes from a set of tested genes in a given cDNA sample panel [64]. PCR reactions for each of the three biological replicates were performed in technical triplicate. The absence of genomic DNA contaminations in the RNA samples was tested by PCR analysis of all samples using oligonucleotides bordering an intron in M. truncatula GSHS gene.
To generate the γ-ECS-RNAi construct, a 502-bp region was amplified from the cDNA using gene-specific primers (Supplemental Table S2 on line) and cloned into the pDONR207 vector, subcloned in pENTR4 and integrated into the RNAi vector pK7GWIWG2DII,(0) [65]containing kanamycin resistance and the p35S:eGFP for selection and screening. M. truncatula plants (A17) were transformed with A. rhizogenes containing precedent construct as described previously [25] and transformed roots were selected by resistance to kanamycin and screening of eGFP. Control plants were transformed with A. rhizogenes containing the pKGWIWG2DII,(0) vector containing an eGFP DNA fragment to rule out the potential side effects linked to plant transformation or the RNAi vector.
Polar metabolites were quantified using 1H-NMR of polar extracts. For the preparation of extracts and NMR acquisition parameters, special care was taken to allow absolute quantification of individual metabolites. Briefly, polar metabolites were extracted on lyophilized powder (30 mg DW per biological replicate) with an ethanol–water series at 80°C as described previously [66]. The lyophilized extracts were titrated with KOD to pH 6 in 100 mM potassium phosphate buffer in D2O and lyophilized again. Each dried titrated extract was solubilized in 0.5 mL D2O with (trimethylsilyl)propionic-2,2,3,3-d4 acid (TSP) sodium salt (0.01% final concentration) for chemical shift calibration and ethylene diamine tetraacetic acid (EDTA) disodium salt (0.5 mM final concentration). 1H-NMR spectra were recorded at 500.162 MHz on a Bruker Avance spectrometer (Bruker, Karlsruhe, Germany) using a 5-mm dual 13C-1H cryoprobe and an electronic reference for quantification [66]. Sixty-four scans of 32 K data points each were acquired with a 90° pulse angle, a 6000 Hz spectral width, a 2.73 s acquisition time and a 25 s recycle delay. Preliminary data processing was conducted with TOPSPIN 1.3 software (Bruker Biospin, Wissembourg, France). The assignments of metabolites in the NMR spectra were made by comparing the proton chemical shifts with literature [66]–[68] or metabolomic database values (MeRy-B 2009, HMDB), by comparison with spectra of authentic compounds recorded under the same solvent conditions and/or by spiking the samples. For assignment purposes, 1H-1H COSY, 1H-13C HSQC and 1H-13C HMBC 2D NMR spectra were acquired for selected samples.
The metabolite concentrations were calculated using AMIX (version 3.9.1, Bruker) and Excel (Microsoft, Redmond, WA, USA) softwares. The metabolites were quantified using the glucose calibration curve and the proton amount corresponding to each resonance for all compounds. The metabolite concentrations were calculated from concentrations in the NMR tube and sample dry weight.
The 15 1H-NMR spectra of the data set were converted into JCAMP-DX format and deposited with associated metadata into the Metabolomics Repository of Bordeaux MeRy-B (http://www.cbib.u-bordeaux2.fr/MERYB/projects/home.php?R=0&project_id=28).
To explore the metabolite multidimensional data set, we used principal component analysis (PCA) on mean-centered data scaled to unit variance (MATLAB version 7.4.0, the MathWorks Inc, Natick MA).
Starch was recovered from the insoluble fraction of the extracts used for polar metabolite extraction after ethanol–water series at 80°C (see above, Moing et al. 2004). Insoluble residues were incubated for 1 h, at 55°C, in a 0.5 ml reaction medium containing 0.1 M sodium acetate, and 1.25 mg amyloglucosidase (Sigma-Aldrich, Saint-Quentin Fallavier, France). Reaction was stopped for 5 min at 100°C. Supernatants were collected and evaporated over night under vacuum. Dry residues were taken up with 0.5 ml of 0.3 M Hepes, pH 7.5, and 30 mM MgSO4. Glucose, issued from starch hydrolysis, was measured as followed: 200 to 400 µl of samples were mixed with a reaction medium containing 0.3 M Hepes, pH 7.5, 30 mM MgSO4, 2.5 mM ATP, and 2 mM NAD. Initial OD was red at 340 nm. Next, 2 Units of both hexokinase (Sigma-Aldrich, Saint-Quentin Fallavier, France) and glucose-6-phosphate dehydrogenase from Leuconostoc mesenteroides (Sigma-Aldrich, Saint-Quentin Fallavier, France) were added, and samples were incubated for 1 h, at room temperature in the dark. Final OD was red at 340 nm. The difference between the final and initial OD was used to calculate the glucose content. Starch was expressed as nmol of glucose equivalent per dry weight unit.
Malate was quantified by ionic chromatography and conductimetry. Separation was performed on an IonPac AS 11 column (4×250 mm, Dionex, Sunnyvale, CA, USA) and a IonPac AG11 guard column (4×50 mm, Dionex) with a NaOH gradient including 16% of methanol. Calibration was performed with commercial standards using gravimetric method.
DNA sequences were analyzed using BLAST [69] against the databases of the NCBI (http://blast.ncbi.nlm.nih.gov/), MtGI (http://compbio.dfci.harvard.edu/cgi-bin/tgi/gimain.pl?gudb=medicago) and the IMGAG (http://www.medicago.org/genome/). The accession numbers of the genes used in this study are indicated in the Supplemental Table S2.
All the data presented are given as means with the standard error of three or four independent biological experiments. The significance of the results was tested using Student t-test (P value ≤0.05).
|
10.1371/journal.pgen.1000520 | Avoiding Dangerous Missense: Thermophiles Display Especially Low Mutation Rates | Rates of spontaneous mutation have been estimated under optimal growth conditions for a variety of DNA-based microbes, including viruses, bacteria, and eukaryotes. When expressed as genomic mutation rates, most of the values were in the vicinity of 0.003–0.004 with a range of less than two-fold. Because the genome sizes varied by roughly 104-fold, the mutation rates per average base pair varied inversely by a similar factor. Even though the commonality of the observed genomic rates remains unexplained, it implies that mutation rates in unstressed microbes reach values that can be finely tuned by evolution. An insight originating in the 1920s and maturing in the 1960s proposed that the genomic mutation rate would reflect a balance between the deleterious effect of the average mutation and the cost of further reducing the mutation rate. If this view is correct, then increasing the deleterious impact of the average mutation should be countered by reducing the genomic mutation rate. It is a common observation that many neutral or nearly neutral mutations become strongly deleterious at higher temperatures, in which case they are called temperature-sensitive mutations. Recently, the kinds and rates of spontaneous mutations were described for two microbial thermophiles, a bacterium and an archaeon. Using an updated method to extrapolate from mutation-reporter genes to whole genomes reveals that the rate of base substitutions is substantially lower in these two thermophiles than in mesophiles. This result provides the first experimental support for the concept of an evolved balance between the total genomic impact of mutations and the cost of further reducing the basal mutation rate.
| Spontaneous mutations are key drivers of evolution and disease. In microbes, most mutations are deleterious, some are neutral (without significant impact), and a few are advantageous. Because deleterious mutations reduce fitness, there should be constant selection for antimutator mutations that reduce rates of spontaneous mutation. However, such reductions are necessarily achieved at some cost. Therefore, a mutation rate should converge evolutionarily on a value that reflects this trade-off. For DNA microbes, the observed genomic mutation rate is remarkably (and mysteriously) invariant, in the neighborhood of 0.003–0.004, with a range of less than two-fold despite huge variation per average base pair in organisms with a wide diversity of life histories. Would an environmental condition that increased the average deleterious impact of a mutation be balanced by additional investments in antimutator mutations? It is widely observed that many mutations with mild impacts become strongly deleterious at higher temperatures, so mutation rates were measured in two thermophiles, a bacterium and an archaeon. Remarkably, both displayed average mutation rates reduced by about five-fold from the characteristic mesophilic value, most of the decrease reflecting a 10-fold reduction in the rate of base substitutions.
| It has become increasingly clear that the basal rate of spontaneous mutation per genome per replication is remarkably invariant in DNA microbes: using a classical correction factor for estimating the ratio of all base-pair substitutions (BPSs) to detected base-pair substitutions, genomic mutation rates (mutations per genome per replication) vary by less than twofold while genome sizes vary by ≈6,000-fold (Table 1). Thus, when mutation rates are expressed per average base pair, they also vary by a similarly large factor. Therefore, basal mutation rates characteristic of unstressed microbial populations can evolve to finely tuned values. The theory of mutation rates has its roots in Haldane's 1927 formulation of the impact of selection and mutation on fitness [1], followed by Sturtevant's 1937 conjecture that the deleterious character of most mutations would generate selective pressures that should lower mutation rates indefinitely [2]. In 1967, Kimura offered the hypothesis that there would be a “physiological cost” to each reduction in rate, leading to an equilibrium value when that cost outweighs the gain in fitness [3]. The surprise has been that the observed genomic rates are so narrowly distributed among DNA microbes despite a wide variety of life histories and genome sizes. An even deeper mystery, not to be addressed here, is why the particular microbial genomic rate of about 0.003–0.004 has been adopted by microbes of such diverse life histories and genome sizes.
If the Kimura conjecture is correct, then increasing the average deleterious impact of a spontaneous mutation (and thus converting many neutral or nearly-neutral mutations to deleterious mutations) would lower the rate of mutation, at least on an evolutionary time scale. The concept of an equilibrium basal mutation rate is difficult to test in a laboratory context because any imposed resetting of the equilibrium would probably require numbers of generations large even by microbial standards, and is difficult to test convincingly because only one or a few habitats could be explored. However, it has recently proven possible to test the concept by examining a natural evolutionary experiment, life at high temperatures. Those who gather mutants for fun or profit have often observed that the most common class of mutations is to temperature sensitivity, indicating that many missense mutations are well tolerated at the standard growth temperature but become much more deleterious, often to the point of lethality, at a temperature only 5°C–10°C higher. This widespread anecdotal observation implies that macromolecular stability becomes increasingly dependent on structural integrity as temperatures rise, a reasonable conjecture in keeping with the considerable constraints observed in the proteins of thermophilic microbes (e.g., [4]). It is therefore likely that the average missense mutation harms thermophiles more than mesophiles (the hypothesis of dangerous missense). This simple prediction was supported by the observation that missense mutations accumulated to a lesser extent (compared to synonymous mutations) in thermophiles than in mesophiles during the course of molecular evolution (dN/dS falling from 0.14 to 0.09), implying stronger purifying selection in thermophiles [5]. Here, direct measurements of the rate and character of spontaneous mutation are compared for mesophilic and thermophilic microbes.
The first phase of determining genomic mutation rates involves measuring a mutation frequency, converting the frequency to a rate, and taking precautions to exclude or take into account the impact of perturbations such as differential growth rates of mutants versus wild type and delayed expression of the mutant phenotype. In addition to measuring rates, it is crucial to identify the kinds of mutations that arise in order to exclude biases due to massive mutational hotspots or to bizarre classes of mutations. The typical result is a rate for a mutation-reporter gene, which is then extrapolated to the whole genome provided that the spectrum of mutations is fairly ordinary. However, there is a substantial problem here: while most indels are detected, most BPSs fail to produce a phenotypic change detectable in the laboratory. One must therefore estimate their full frequencies. (An exception is the still rare case that mutation detection is achieved with phenotype-blind genomic DNA sequencing.) Two methods have been applied. Both make the reasonable assumption that almost all indels and chain-termination (CT) BPSs are detected with high efficiency in protein-coding sequences. (Although exceptions occur, they are infrequent and tend to occur at the extreme downstream end of a gene.) The first method was based in part on the average relative frequencies of CT and non-CT BPSs in a handful of spectra and provided a correction factor for base substitutions of 4.726 [6]. This method was used for almost all of the entries in Table 1; however, the range of values averaging to 4.726 was large, reducing reliability. The second method is based exclusively on CT mutations. It involves examining the reporter sequence for all possible BPSs capable of generating CTs and then dividing the observed CT mutation frequency by that reduced target size and multiplying by 3 (to account for the three BPSs that can arise at any site) to obtain an average mutation rate per base pair. The CT method also has drawbacks. First, it cannot report A·T→G·C mutations, but these generally arise at approximately average BPS rates, suggesting a minimal problem. Second, CT mutations are typically a minority of all mutations, so that many spectra sport only a few CTs, reducing sampling accuracy.
The other major barrier to accurate extrapolation from a mutation-reporter gene to the whole genome becomes manifest when sequencing reveals a major hotspot. Mutation rates at particular sites vary greatly, but most mutational spectra display a range of site-specific numbers of mutations ranging from 1 to hotspots with from several percent to even a quarter of the whole collection. The impact of a hotspot containing a quarter of all the mutations is modest, but some genes contain single hotspots bearing the large majority of mutations; the classic example is the E. coli lacI gene, where ∼72% of all mutations are indels arising at a stretch of 13 BPSs consisting of 3.25 repeats of a tetramer [7]. However, such massive indel hotspots are infrequent among genes, and it is reasonable to post occasional genomic rates both including and removing them.
All informative microbial mutation rates obtained before 2000 were for mesophilic species, but rates and spectra are now available for two genes in each of two very different thermophiles, the crenarchaeon Sulfolobus acidocaldarius [6] and the bacterium Thermus thermophilus [8], both growing at close to 75°C. In the first study, with S. acidocaldarius, BPSs were a smaller fraction of the spectrum than in mesophiles, and this observation prompted the hypothesis of dangerous missense. Note, however, that if greater fractions of missense mutations are phenotypically detectable in thermophiles than in mesophiles, then the historical method of correcting for undetected BPSs becomes inappropriate when based on mesophiles. It is therefore advisable to resort exclusively to the CT method for estimating total BPS rates, which is the central result for this report.
Table 2 lists genomic mutation rates estimated using the CT method (or its lacZα equivalent), sometimes based on the same sources as for Table 1 but excluding some reports whose sequencing information was inadequate for the CT method. The nine entries at the top are for mesophiles and reveal no significant departures from the values in Table 1, providing empirical confidence in the robustness of the CT method. The two entries at the bottom are for thermophiles, whose numbers of CTs are small. (The data for the two mutation-reporter genes are combined in each organism because of the small number of CTs.) The thermophile BPS rates are substantially lower, by about 10-fold, than their mesophile counterparts. When major indel hotspots are included, indel rates are less than twofold lower in thermophiles, while total genomic rates are about fivefold lower. (When the indel hotspots are removed from the analysis, the indel rate decrease is three-fold and the total genomic rate decrease is seven-fold.) Although these ratios are somewhat uncertain because of the small numbers of CTs for five of the seven mesophiles and both thermophiles, the mean difference is large enough to support the inference that BPS rates are lower in thermophiles. The mesophile and thermophile values were compared using randomization t-tests [9], a nonparametric test that requires no assumptions about normality or equal variances of the mutation rates. The resulting one-sided p values are 0.018 for both the total mutation rate and its BPS component, and 0.27 for the indel values that include the hotspots.
Genomic mutation rates have long been suspected to evolve as a balance between the deleterious impact of the average mutation and the cost of further reducing the mutation rate. A test of this conjecture on the evolutionary scale could consist of estimating mutation rates in organisms whose environment increases the impact of the average mutation. Because many base substitutions do greater harm at higher temperatures, thermophiles were suitable candidate organisms. For both a bacterium and an archaeon, the thermophiles display sharply reduced rates of base pair substitutions compared to the typical mesophile.
The lower mutation rates in thermophiles are likely to reflect their higher optimal growth temperatures. There is no obvious hint of a particular aspect of life history other than temperature that sets the two thermophiles apart from the mesophiles. The %(G+C) values for the ten organisms in Tables 1 and/or 2, listed monotonically with the two thermophile values in bold, are 35–36–37–38–41–50–50–51–68–69, providing no hint of a role for this variable, as also noted in the earlier molecular-evolution study [5]. Thus, the Kimura conjecture, that the equilibrium mutation rate reflects a balance between the impact of the average mutation compared to the cost of keeping mutations in check, is supported in a natural experiment.
The hypothesis of dangerous missense predicts that BPS rates will be reduced in thermophiles but does not speak directly to indel rates. However, indel rates are also reduced, although less strongly than are BPS rates and with a p value of 0.27 for these data. One candidate explanation for this difference is that the reduction in BPS rates is achieved by the accumulation of modifiers selected to target BPS mutagenesis but at most incidentally targeting indel mutagenesis. Because single-base additions and deletions tend to be the large majority of indels in mesophiles (35 single-base indels/38 total indels in phage λ, 20/23 in phage T4, 45/45 in Herpes simplex virus, 604/641 in E. coli, 88/97 in S. cerevisiae, and 24/32 in S. pombe) and are similarly frequent in thermophiles (84/95 in S. acidocaldarius and 46/54 in T. thermophilus), these small indels must be the main targets of antimutagenic modifiers acting on indels generally. Both single-base indels and BPSs result from errors of insertion followed by failures of proofreading and DNA mismatch repair in well studied model organisms such as E. coli and S. cerevisiae, but little is known about the sources of spontaneous mutations in S. acidocaldarius and T. thermophilus.
Are there likely to be other outliers with informative deviations from the mutational pattern that is consistently displayed among the mesophilic microbes examined to date with respect to either the mutation rate or the BPS:indel ratio?
Mutations to cold sensitivity are rarely reported and are anecdotally described as difficult to discover. If they are indeed rare, perhaps fewer missense mutations produce mutant phenotypes in psychrophiles than in mesophiles. One evolutionary consequence might then be a relaxation to a higher spontaneous rate of BPS mutation, perhaps with little effect on the rate of indel mutation.
Because of incomplete buffering against the impacts of their environments, halophiles and acidophiles experience relative high internal concentrations of Na+ and H+, respectively, compared to other microbes. These ionic environments might be unusually stressful to mutants carrying missense mutations, resulting in adjustments to their mutational patterns in the same direction as seen for thermophiles. Although without significance because of sampling constraints, Table 2 attributes a five-fold lower BPS mutation rate to the acidophile S. acidocaldarius than to the non-acidophile T. thermophilus. Unfortunately, an attempt to characterize mutation in the halophilic archaeon Haloferax volcanii failed, probably because this mesophile is highly polyploid [10].
The lactic acid bacterium Oenococcus oeni, used in wine making to convert malic acid to lactic acid, lacks the usual bacterial DNA mismatch repair (MMR) system and has a high mutation rate as judged by mutations conferring resistance to rifampin and erythromycin, as does Oenococcus kitaharae [11]. These results suggest a powerful genus-wide mutator condition, which would normally be highly deleterious. The question then arises whether the lack of MMR is so strongly adaptive in these species as to outweigh the sharply decreased fitness of the mutator condition, or whether the species have been unable to re-acquire the MMR genes by horizontal transfer.
Whereas the above two species lack MMR function and display mutator phenotypes, the crenarchaeons as a whole, including S. acidocaldarius, lack all known bacterial MMR genes, but S. acidocaldarius, at least, displays an antimutator phenotype compared with mesophiles. How can this be? In Escherichia coli, the mutation rate per average base pair ≈8×10−10 (Tables 1 and 2). Based on the strengths of mutator mutations, replication infidelity can be estimated as the product of three components during DNA replication: insertion errors ≈0.9×10−5, proofreading failures ≈1.7×10−2, and MMR failures ≈5×10−3 [12],[13]. In bacteriophage T4, which does not employ a general MMR system, the mutation rate per average base pair ≈2×10−8 (Tables 1 and 2). Based on the strengths of mutator mutations, replication fidelity can be estimated as the product of two components during DNA replication: insertion errors ≈1×10−5 and proofreading failures ≈2×10−3 [13]. Thus, T4 makes up for the lack of MMR by a proofreading potency about an order of magnitude greater than that operating in E. coli. The mutation rate per base pair for S. acidocaldarius ≈3×10−10, which might be achieved by a product of factors applied to the T4 insertion and proofreading accuracies that together produce a 70-fold improvement. Alternatively, S. acidocaldarius may possess an MMR system so distinct from the standard mutHLS model as to have escaped recognition by genomic scans. Note also that both thermophiles have genomes about twofold smaller than the E. coli genome.
We begin in possession of values for the following:
G = the genome size in bases or base pairs.
T = the number of bases or base pairs in the target (the mutation-reporter sequence).
μT = the measured mutation rate at T, corrected where necessary for mutants expressing the characteristic phenotype but revealed by sequencing to lack mutations in the reporter gene, but not corrected for mutants with two or more mutations (which are infrequent and sometimes absent). In many cases, μT = f/ln(μTN) where f = the measured mutation frequency for the given target, N = the final population size, and the median μT over several cultures is used [14], a method that is robust compared to the classical fluctuation test provided the average number of mutational events per culture is ≥30 [15].
M = number of sequenced mutants = B+I, where B = number of BPS mutants and I = number of indel mutants, the latter also including complex mutants (a minority, if present at all) regardless of their components.
For the “historical” method, we correct for undetected BPSs by multiplying the number of detected BPSs by 4.726 [6]. Then the average mutation rate per base or base pair μb = [μT corrected upwards by (I+4.726B)/M]/T = (I+4.726B)( μT/MT). The genomic mutation rate μg = Gμb.
For the “CT” method, the indel genomic mutation rate μg(I) is calculated as above ignoring the BPS component, B becomes BCT = number of mutations to a chain-terminating codon (TAG, TGA, or TAA), and P = number of possible mutational pathways to a CT mutation within T (there being three mutational BPS pathways per base or base pair). Then the BPS genomic mutation rate μg(B) = μT (3BCT/MP)G. The total genomic rate μg = μg(I)+μg(B).
|
10.1371/journal.pgen.1005561 | PPP2R5C Couples Hepatic Glucose and Lipid Homeostasis | In mammals, the liver plays a central role in maintaining carbohydrate and lipid homeostasis by acting both as a major source and a major sink of glucose and lipids. In particular, when dietary carbohydrates are in excess, the liver converts them to lipids via de novo lipogenesis. The molecular checkpoints regulating the balance between carbohydrate and lipid homeostasis, however, are not fully understood. Here we identify PPP2R5C, a regulatory subunit of PP2A, as a novel modulator of liver metabolism in postprandial physiology. Inactivation of PPP2R5C in isolated hepatocytes leads to increased glucose uptake and increased de novo lipogenesis. These phenotypes are reiterated in vivo, where hepatocyte specific PPP2R5C knockdown yields mice with improved systemic glucose tolerance and insulin sensitivity, but elevated circulating triglyceride levels. We show that modulation of PPP2R5C levels leads to alterations in AMPK and SREBP-1 activity. We find that hepatic levels of PPP2R5C are elevated in human diabetic patients, and correlate with obesity and insulin resistance in these subjects. In sum, our data suggest that hepatic PPP2R5C represents an important factor in the functional wiring of energy metabolism and the maintenance of a metabolically healthy state.
| After a meal, dietary glucose travels through the hepatic portal vein to the liver. A substantial part of this glucose is taken up by liver, which converts it to glycogen which is stored, and lipids which are in part stored and in part secreted as VLDL particles. The rest of the organs receive whatever glucose the liver leaves in circulation, plus the secreted lipids. Hence the liver plays a crucial role in determining the balance of sugar versus lipids in the body after a meal. This balance is very important, because too much glucose in circulation leads to diabetic complications whereas too much VLDL increases risk of atherosclerosis. Little is known about how the liver strikes this balance. We identify here a phosphatase—the PP2A holoenzyme containing the PPP2R5C regulatory subunit—as a regulator of this process. We find that knockdown of PPP2R5C in mouse liver specifically causes it to uptake elevated levels of glucose, and secrete elevated levels of VLDL into circulation. This leads to a phenotype of improved glucose tolerance and insulin sensitivity. The prediction from these functional studies in mice is that elevated levels of PPP2R5C expression should lead to insulin resistance. Indeed, we find that PPP2R5C expression levels are elevated in diabetic patients, or healthy controls with visceral obesity, raising the possibility that dysregulation of PPP2R5C expression in humans may contribute towards metabolic dysfunction.
| According to Greek mythology, Odysseus was forced to carefully navigate his ship between two dangers—Scylla and Charybdis—whereby passing too close to either one would lead to destruction. Likewise, in metabolic regulation, the liver strikes a difficult balance between glucose and lipid homeostasis. After a meal, dietary glucose travels through the hepatic portal vein to the liver. The liver uptakes a substantial part of this glucose, removing it from circulation, and converting it to glycogen for storage, or to lipids which are in part stored and in part re-secreted as VLDL particles [1–3]. Hence the liver plays a crucial role in determining the balance of sugar versus lipids in the body after a meal. While elevated circulating glucose is intimately linked to diabetes, elevated lipids are linked to atherosclerosis or non-alcoholic fatty liver disease [4–6]. Despite the critical importance of handling both glucose and lipid pathways in a coordinated and balanced manner, the molecular mechanisms regulating this balance remain to be investigated. We describe here a phosphatase that affects how the liver strikes this balance between glucose and lipid metabolism.
Phosphatases are an interesting class of enzymes because they can have specific, yet very strong cellular effects by dephosphorylating multiple proteins in one signaling pathway or biological process [7–13]. One such phosphatase is the PP2A holoenzyme, the most abundant serine/threonine phosphatase in the cell. The PP2A holoenzyme is composed of three subunits, a large scaffolding A subunit, a catalytic C subunit which performs the dephosphorylation reaction, and one of many possible regulatory B subunits which provide substrate specificity to the holoenzyme [10,14]. Four families of B subunits have been identified (B, B’, B” and B”‘) with each family containing multiple members [10], resulting in a large array of possible B subunits. Hence, although the PP2A catalytic subunit which is shared by all these holoenzymes has pleiotropic effects, the individual B subunits which direct the phosphatase to a subset of targets can have very specific effects [15,16]. We previously showed that Drosophila lacking one of the B’ subunits, PP2A-B’, are viable but have metabolic defects [17]. Starting from these observations in the fly, we hypothesized that one of the two mammalian homologs of PP2A-B’, called PPP2R5C or B56γ, might also be involved in metabolic regulation in mammals.
To date, PPP2R5C has been linked to cancer development. Several studies have reported either increased or decreased expression of PPP2R5C in various tumor types [18–21]. One mechanism by which PPP2R5C affects tumor development appears to be via dephosphorylation of p53 on Thr55 [22–24], leading to inhibition of cell proliferation and anchorage-independent growth [23,25]. PPP2R5C has also been proposed to have p53-independent mechanisms [26]. Indeed, mouse p53 appears to lack Thr55, suggesting that PPP2R5C acts via additional mechanisms. Whole-body PPP2R5C knockout mice are viable but display heart defects including an incomplete ventricular septum and reduced numbers of ventricular cardiomyocytes [27]. PPP2R5C knockout mice also display reduced locomotive coordination and gripping strength [27]. Together with the heart defects, this suggests a muscular function for PPP2R5C. Interestingly, PPP2R5C knockout mice also developed obesity with age [27], raising the possibility that PPP2R5C regulates metabolism either directly or indirectly as a consequence of its effects on locomotion. The involvement of PPP2R5C in the regulation of metabolic homeostasis, however, remains to be investigated.
We investigate here for the first time the consequences of tissue-specific PPP2R5C deficiency, and show that hepatic PPP2R5C plays a role in postprandial physiology, affecting the balance of conversion of sugar into lipids by the liver. PPP2R5C knockdown in hepatocytes (HepKD) leads to increased glucose uptake and lipid biosynthesis. Consequently, HepKD mice are highly glucose tolerant and insulin sensitive. Interestingly, we find that liver and adipose PPP2R5C expression levels correlate with diabetic state and glucose tolerance in humans in agreement with the functional studies in mouse models.
Since our previous studies in Drosophila identified a role for fly PPP2R5C (called PP2A-B’) in the regulation of organismal metabolism [17], we asked if we could also observe a link between PPP2R5C and metabolism in mice. Genes for metabolic regulators are often present in transcriptional regulatory feedback loops [28,29], so we first tested if organismal nutritional status affects PPP2R5C expression. Indeed, we found that PPP2R5C expression is nutritionally regulated in tissues of metabolic relevance such as liver, adipose tissue, and muscle, albeit in a complex way. In mouse liver, PPP2R5C expression increases upon fasting, and drops again upon refeeding (Fig 1A). This nutritional regulation of PPP2R5C is lost in obese diabetic db/db mice, which express elevated, constant levels of PPP2R5C. In abdominal white adipose tissue, PPP2R5C expression does not respond to changes in feeding status, but similar to liver, expression of PPP2R5C is significantly, tonically elevated in db/db mice compared to controls (Fig 1B). Opposite to liver, in gastrocnemius muscle PPP2R5C expression increased upon feeding (Fig 1C) and again this regulation is blunted in db/db mice (Fig 1C). These transcriptional responses to nutritional status did not translate into detectable changes in PPP2R5C protein levels in mouse liver, indicating that they are either buffered at the translational level, or our antibody is not sensitive enough to detect changes of this magnitude (S1A Fig). By screening a panel of drugs, we found that PPP2R5C expression is inhibited by stimulation with insulin or human FGF19 (homologous to mouse FGF15) in primary hepatocytes (S1B Fig) and is induced by a PPARα agonist to a degree similar to that of a canonical PPARa target gene, CPT1A (S1C Fig). In contrast, stimulation of primary hepatocytes with leptin did not cause a change in PPP2R5C expression (S1D Fig) whereas feeding mice a high-fat diet for 4 weeks led to a mild increase in PPP2R5C expression (S1E Fig), suggesting that the elevated expression of PPP2R5C in db/db mice might be an indirect consequence of their altered metabolic status, and not directly due to impaired leptin signaling. In sum, although these transcriptional changes may not result in functional consequences, they suggest there might be links between PPP2R5C and tissue-specific metabolic regulation.
To study if PPP2R5C regulates metabolism in mammals, we knocked-down PPP2R5C expression in vivo in the mouse and assayed if this leads to metabolic alterations. Since liver is a central organ for metabolic regulation, we focused specifically on hepatic PPP2R5C function. Tail-vein injection of adeno-associated virus carrying a miRNA under control of a hepatocyte-specific promoter [30] targeting all transcriptional isoforms of PPP2R5C led to a significant reduction in PPP2R5C mRNA and protein levels in the liver (S2A–S2A’ Fig). As a control, we injected equal amounts of an adeno-associated virus carrying a non-targeting miRNA. Body weight and serum ALT levels were not altered upon PPP2R5C hepatocyte-specific knockdown (“HepKD”) (S2B and S2C Fig), indicating that the liver is not experiencing severe stress in these conditions. Since the liver is important for maintaining euglycemia, we measured blood glucose levels in knockdown mice either in an uncontrolled feeding regimen (“Random”), or after 16 hours of fasting (“Fasting”), followed by 6 hours of refeeding (“Refed”). In all three conditions, PPP2R5C HepKD did not have a significant effect on blood glucose levels (Fig 2A). These same mice, however, displayed dramatically reduced levels of circulating insulin (Fig 2B), indicating that lower insulin levels are required to maintain euglycemia upon liver knock-down of PPP2R5C. Consistent with this, PPP2R5C HepKD mice defended circulating glucose levels more efficiently than control mice in a glucose tolerance test (2g glucose injected intraperitoneally per kg body weight, Fig 2C) despite lower insulin levels (S2D Fig). In sum, these data suggest PPP2R5C HepKD livers have elevated insulin sensitivity. Indeed HepKD livers maintain activation of insulin signaling, as judged by phosphorylation of Akt and the downstream target GSK3β, despite lower circulating insulin levels (Fig 2D and 2B). Strikingly, in a direct test of insulin sensitivity by measuring Akt phosphorylation 10 minutes after tail-injection of insulin we found that indeed PPP2R5C HepKD livers have elevated insulin sensitivity (Fig 2E).
To test if this is a cell-autonomous phenotype in the liver, we turned to cell culture. Knockdown of PPP2R5C (S2E Fig) caused Hepa 1–6 cells to deplete glucose from the medium more quickly than control knockdown cells (Fig 2F). Correspondingly, lactate production was also elevated in Hepa 1–6 cells upon PPP2R5C knockdown, indicating an increase in the overall glycolytic flux (Fig 2G). This was confirmed by measuring glycolytic flux using a Seahorse analyzer (Fig 2H). We also observed an increase in glucose uptake following a paradigm similar to the Glucose Tolerance Test, whereby Hepa 1–6 cells were starved overnight in serum-free DMEM and then treated with a fluorescent glucose analog for 20 minutes. Quantification by FACS revealed increased uptake upon induction of two independent shRNAs targeting PPP2R5C (Fig 2I, S2F Fig for knockdown efficiency control), thereby also ruling out possible off-target effects. Together, these data indicate that knockdown of PPP2R5C leads to a cell-autonomous increase in glucose uptake.
In addition to increased hepatic glucose uptake, one other factor that could contribute to improved glucose tolerance in vivo is reduced gluconeogenesis. However intraperitoneal injection of pyruvate, a gluconeogenic substrate, caused a similar ascension in blood glucose levels of PPP2R5C HepKD mice compared to controls (0–50 min, S2G Fig), indicating they do not have significantly reduced gluconeogenic capacity. Consistent with improved glucose clearance in PPP2R5C HepKD mice, blood glucose levels returned to baseline more quickly in knockdown mice compare to controls (the descending phase of the pyruvate tolerance test, S2G Fig), although the difference was not statistically significant. Consistent with the lack of change in gluconeogenic capacity, we also found no reduction in expression of gluconeogenic genes in PPP2R5C HepKD mice (S2H Fig). We did observe an increase in expression of G6PC (S2H Fig), which is a ChREBP target [31], as discussed below.
In sum, knockdown of PPP2R5C in liver leads to increased glucose uptake, improved glucose tolerance, improved insulin sensitivity and reduced insulin levels in vivo.
The increased glucose uptake upon PPP2R5C knockdown suggests that PPP2R5C knockdown might cause liver cells to shift towards a more anabolic metabolic profile. In agreement with this, quantification of liver weight revealed that PPP2R5C HepKD leads to increased liver mass in all three feeding regimens tested (Fig 3A), despite normal food intake (S3A Fig). Glucose is used by hepatocytes in part to synthesize glycogen [32]. Consistent with the increased glucose uptake, PPP2R5C knockdown livers had increased glycogen levels compared to controls, even when normalized to liver weight (Fig 3B). Most strikingly, although glycogen levels drop in control livers upon fasting, as they enter a catabolic state to provide the rest of the organism with glucose, PPP2R5C knockdown livers displayed almost no drop in glycogen upon fasting (Fig 3B).
Glucose is also used by hepatocytes for lipid biosynthesis. In the random feeding state, mice with PPP2R5C HepKD had significantly elevated TAG levels in their livers. Although this effect was visible 7 weeks after injection of PPP2R5C knockdown virus (Fig 3C), it was even more pronounced 2 weeks after virus injection (S3B Fig), perhaps due to compensatory regulatory mechanisms developing over time. One possible explanation for the increased liver TAG levels could be reduced liver fatty acid beta-oxidation, however circulating ketone bodies were not elevated in HepKD mice, suggesting this is not the case (S3C Fig). Furthermore, PPP2R5C knockdown in Hepa 1–6 cells did not lead to reduced levels of fatty acid beta-oxidation (S3 Fig panel D), which if anything were slightly elevated. Alternatively, HepKD livers might have elevated lipid biosynthesis rates. To test if PPP2R5C knockdown leads to increased lipogenesis in a cell-autonomous manner in hepatocytes, we turned once again to cell culture. Both Hepa 1–6 cells as well as primary mouse hepatocytes displayed increased TAG levels upon PPP2R5C knockdown (Fig 3D and 3E). These increased TAG levels could not be explained by an increase in free fatty acid uptake from the medium (S3E Fig), nor by reduced TAG secretion into the medium (since neither control nor PPP2R5C knockdown Hepa 1–6 cells secrete TAG into the medium, S3F Fig), indicating that PPP2R5C leads to both increased de novo lipogenesis and triglyceride formation in hepatocytes. PPP2R5C HepKD livers had reduced triglyceride levels upon fasting, when de novo lipogenesis is very low (Fig 3C), as discussed below. Upon refeeding, however, they re-accumulated triglycerides more rapidly than controls, reaching control levels within 6 hours of refeeding (Fig 3C), consistent with elevated de novo lipogenesis in PPP2R5C knockdown livers when dietary glucose is available. Taken together, PPP2R5C knockdown livers take up more glucose than control livers, thereby producing more triglycerides,
Surprisingly, triglyceride levels dropped significantly in PPP2R5C HepKD livers upon fasting when de novo lipogenesis in liver is very low (Fig 3C). This was accompanied by a strong elevation of circulating VLDL levels in PPP2R5C HepKD mice upon fasting or brief refeeding (Fig 3F, 3G and S3G Fig). One possible explanation consistent with reduced liver triglycerides and increased circulating VLDL is that HepKD livers secrete more VLDL than controls. Indeed consistent with this explanation, PPP2R5C HepKD led to a concomitant drop in liver cholesterol levels upon fasting (S3H Fig), without a significant change in circulating free fatty acid levels (S3I Fig). However, other explanations for this phenotype are also possible, such as reduced VLDL re-uptake by HepKD livers, as discussed below.
In sum, PPP2R5C knockdown leads to increased glucose uptake and increased de novo lipogenesis in cell culture and in vivo when dietary glucose is available, and elevated VLDL in circulation upon starvation.
PPP2R5C is a regulatory subunit of PP2A, thought to provide substrate specificity to the phosphatase holoenzyme. Therefore, we aimed to identify target substrates that bind PPP2R5C. PPP2R5C has been shown to target Thr55 of human p53 for dephosphorylation, but this residue is not present in mouse p53, prompting us to search for additional substrates. Since protein-protein interactions between phosphatases and substrates are notoriously transient and difficult to detect via co-immunoprecipitation strategies, we employed the BioID method [33] to identify PPP2R5C interacting proteins. We expressed in Hepa 1–6 cells a fusion between PPP2R5C and the biotin ligase BirA (Fig 4A). This leads to biotinylation in vivo of PPP2R5C interacting proteins, which can subsequently be purified by cell lysis and streptavidin binding. We fused BirA to either the N-terminus or the C-terminus of PPP2R5C (Myc-BirA-PPP2R5C and PPP2R5C-BirA-HA respectively), and used Myc-BirA or BirA-HA alone as negative controls. In addition, we also introduced a mutation into the catalytic subunit of PP2A (PPP2CA) known to eliminate phosphatase activity [34], with the aim of generating a ‘substrate-trapping’ mutation to extend the duration of interaction between PP2A and substrate proteins. In this manner, we tested if PPP2R5C binds to various known regulators of liver metabolism, and found specific interactions with the beta-1 subunit of AMPK (Fig 4B and S4A Fig), HIF1α, STAT3 and S6K (Fig 4B’). In contrast, we could detect no binding of PPP2R5C to SREBP-1, PPARα, LXR, or a panel of negative control proteins (RpL26, TSC1, YAP, HSP90, Tubulin, and Actin, Fig 4B–4B’, S4A Fig). Since phosphatases are known to dephosphorylate multiple substrates, these results suggest that the metabolic effects of PPP2R5C might be due to the combined effects on several metabolic regulators. We tried to confirm the interaction between PPP2R5C and AMPK by co-immunoprecipitation of endogenous proteins, but were unsuccessful, possibly due to the transient nature of phosphatase–substrate interactions which rarely survive biochemical purification. Instead, we employed Proximity Ligation Assay (PLA) [35] which detects protein-protein interactions in situ by fixing cells, staining with two antibodies recognizing the two proteins of interest, and detecting complex formation of the two antibodies. We could detect a strong PLA signal when tagged PPP2R5C and AMPK-β1 were expressed in Hepa 6–1 cells, but not when only one of the two proteins was expressed, indicating specificity of the interaction (S4B Fig).
We next tested if binding of PPP2R5C to AMPK or HIF1a leads to changes in their phosphorylation state and activity. If the PPP2R5C-PP2A holoenzyme dephosphorylates AMPK, we would expect increased AMPK phosphorylation upon PPP2R5C knockdown. To test this, we generated Hepa 1–6 cell lines transfected with two independent, inducible shRNAs targeting PPP2R5C. We used two independent shRNAs targeting PPP2R5C to avoid possible off-target effects. Upon induction of the PPP2R5C-targeting shRNAs, AMPK phosphorylation on Thr172 increased significantly (Fig 4C). Concurrently, phosphorylation of two AMPK substrates, ACC1 and TBC1D1, was also elevated upon PPP2R5C knockdown (Fig 4C) suggesting that PPP2R5C knockdown leads to elevated AMPK activity.
Since PPP2R5C also interacts with HIF1α (Fig 4B’), we next studied the effect of PPP2R5C knockdown on HIF1α phosphorylation. To our knowledge, however, phospho-specific antibodies are not available to detect HIF1α phosphorylation. Therefore, we employed phos-tag gels, which contain functional groups that specifically bind phosphate, causing phosphorylated proteins to migrate more slowly compared to their respective un-phosphorylated forms [36]. In this manner, we observed an up-shift of HIF1α upon PPP2R5C knockdown in Hepa 1–6 cells, that could be reversed by treating the cell lysates with CIP phosphatase prior to PAGE (Fig 4D), suggesting that PPP2R5C affects HIF1α phosphorylation. To test if this has any functional consequences, we looked at expression of HIF1α target genes and found them to be up-regulated in primary hepatocytes upon PPP2R5C knockdown (Fig 4E).
To analyze more broadly the changes occurring within hepatocytes upon PPP2R5C knockdown, we performed microarray expression profiling of polyA mRNA from mouse primary hepatocytes in the presence and absence of PPP2R5C knockdown. This identified 11 down-regulated and 49 up-regulated genes upon PPP2R5C knockdown (2-fold cut-off, S1 Table). Since PPP2R5C is part of a phosphatase complex, these transcriptional changes likely occur as a secondary consequence of altered activity of transcription factors in signaling pathways targeted by PPP2R5C. To identify these transcription factors, we used the TFactS software, which predicts transcription factors that are dysregulated by comparing lists of up-regulated and down-regulated genes to annotated catalogs of transcription factor target genes [37]. This analysis identified PPARα and SREBP-1 as the two up-regulated transcription factors upon PPP2R5C knockdown (p-value<0.05, Fig 5A). Since SREBP-1 promotes lipid biogenesis, which is up-regulated in PPP2R5C knockdown hepatocytes (Fig 3), we analyzed this in more detail. Knockdown of PPP2R5C in primary hepatocytes led to elevated expression of a panel of SREBP-1 target genes (Fig 5B). Elevated expression of SREBP-1 target genes was also observed in vivo in PPP2R5C knockdown livers, especially in the refed condition (Fig 5C), suggesting elevated SREBP-1 activity. PPP2R5C knockdown livers also had elevated levels of SREBP-1 precursor as well as mature SREBP-1 protein (Fig 5D), in agreement with SREBP-1 positively auto-regulating its own expression [29]. In sum, elevated SREBP-1 activity may be contributing to the increased lipogenesis phenotype of PPP2R5C knockdown hepatocytes.
Another transcription factor promoting conversion of carbohydrates into triglycerides in liver is ChREBP [38]. We found that PPP2R5C HepKD livers have elevated expression of several but not all ChREBP targets (Fig 5E and G6PC in S2H Fig), suggesting ChREBP activity might also be elevated in knockdown livers. In contrast, we did not see strong changes in expression of PPARα target genes in knockdown livers compared to controls (S5 Fig panel A), although a few PPARα target genes such as CPT1A and ADFP were significantly reduced. Finally, expression of two LXR targets was increased, but not expression of LXR itself (S5B Fig).
Taken together, these data suggest that PPP2R5C regulates AMPK, HIF-1α, and a yet-to-be identified target that affects SREBP-1 activity.
Since reduced PPP2R5C expression in liver leads to significantly improved glucose tolerance and improved insulin signaling (Fig 2C and 2E), we wondered if diabetic patients might have the opposite—elevated levels of PPP2R5C expression. Indeed, PPP2R5C expression was significantly increased in livers of type-2 diabetic patients (p = 0.0003, Fig 6A). Even in non-diabetic patients, PPP2R5C expression increased with increasing adiposity, in particular visceral obesity, an important risk factor for diabetes (Fig 6B). Both in diabetic and non-diabetic patients, PPP2R5C liver expression correlated inversely with insulin sensitivity, determined by glucose infusion rate (GIR) during hyperinsulemic-euglycemic clamp (Fig 6C), in agreement with our mouse data showing that reduced PPP2R5C liver expression leads to improved glucose tolerance (Fig 2C). In sum, these data raise the possibility that altered PPP2R5C expression in liver might contribute towards the etiology of type-2 diabetes. We extended our analysis to other tissues and found that PPP2R5C expression is also elevated in subcutaneous, but not visceral, white adipose tissue in type-2 diabetic patients compared to healthy controls (p = 0.04, S6 Fig). These data fit nicely with the expression data from mice, showing elevated levels of PPP2R5C expression in livers and adipose tissue of diabetic mice (Fig 1A and 1B).
Since PPP2R5C HepKD leads to improved insulin sensitivity and glucose tolerance, we asked if PPP2R5C HepKD could have beneficial effects in db/db mice, which are hyperphagic and consequently become obese and diabetic [39]. Consistent with the results of PPP2R5C knockdown in wildtype mice, PPP2R5C HepKD in leptin receptor-deficient db/db mice ameliorated their diabetic phenotypes, reducing circulating glucose levels (Fig 7A) and improving their response in an insulin tolerance test (Fig 7B). PPP2R5C HepKD in wildtype mice, however, indicated that the down-side of improved glucose handling is elevated liver or circulating triglycerides. Indeed, also in db/db mice, PPP2R5C HepKD led to elevated body mass accumulation (Fig 7C) and whole body fat content (Fig 7D) which was mainly due to increased triglycerides in liver (Fig 7E) but not adipose tissue (Fig 7F). In sum, PPP2R5C HepKD in diabetic mice worsened their dyslipidemia but ameliorated their hyperglycemia and improved their insulin response. This phenotype is interesting in light of the fact that obesity and insulin resistance, which often correlate in humans, can be uncoupled, with 20% of obese people displaying a ‘healthy obese’ phenotype with good insulin sensitivity and no diabetes [40].
We identify here PPP2R5C as a modulator of liver metabolism. Reduced PPP2R5C expression in hepatocytes leads to increased glucose uptake and increased de novo lipogenesis in cell culture. These phenotypes are reiterated in vivo whereby hepatocyte-specific knockdown (HepKD) of PPP2R5C yields mice with improved glucose tolerance but elevated liver triglyceride or circulating VLDL levels. Hence PPP2R5C modulates the balance the liver needs to strike between preventing circulating glucose levels from becoming too elevated after a meal, and yet not flooding the circulatory system with lipids (Fig 5F).
Interestingly, PPP2R5C HepKD mice have reduced levels of circulating insulin (Fig 2B) but normal levels of circulating glucose (Fig 2A). This is likely because the rheostat for euglycemia, the pancreas, is not affected in our HepKD mice. In the ‘random fed’ and ‘refed’ states, PPP2R5C HepKD livers uptake more glucose than normal (Fig 2). Upon starvation, PPP2R5C HepKD livers have reduced glycogen mobilization (Fig 3B) and normal gluconeogenesis (S2G Fig), indicating reduced glucose output in total compared to control animals. Hence, under all feeding conditions, PPP2R5C HepKD livers contribute towards reduced blood glucose (either via elevated glucose clearance or reduced glucose release) compared to control livers. Thus, to maintain euglycemia peripheral tissues such as muscle must be compensating by uptaking less glucose, likely as a result of reduced insulin from the pancreas.
The phenotypes we report here were obtained by targeting PPP2R5C with multiple different miRNA or shRNA sequences. For instance, the increased glucose clearance was observed in cell culture using two independent shRNAs (Fig 2G) and a third independent target sequence in vivo (Fig 2C). This allows us to exclude that the phenotypes could arise from possible off-target effects.
Although the phenotypes of PPP2R5C knockdown are quite specific both in cell culture and in vivo, they likely result from an effect of PPP2R5C on multiple downstream targets. Hence it will likely be difficult or impossible to identify a single downstream target as the main one mediating the effects of PPP2R5C. We identify here 4 protein complexes as PPP2R5C interactors—AMPK, HIF1a, STAT3 and S6K. We tested whether phosphorylation of these proteins increases upon PPP2R5C knockdown, as would be expected of a PPP2R5C target. For S6K, we analyzed phosphorylation on Thr389 using phospho-specific antibody, and did not find it elevated upon PPP2R5C knockdown in hepatocytes. For this reason, we did not pursue S6K further, although this does not exclude that S6K phosphorylation on another site or in a different cell type could be regulated by PPP2R5C. Indeed, the fact that PPP2R5C HepKD renders the liver more sensitive to insulin (Fig 2E) suggests it might be affecting insulin signaling by counteracting phosphorylation of a component of the insulin pathway. For STAT3, we analyzed phosphorylation of Ser727, but saw no change upon PPP2R5C knockdown. We also tested STAT3 motility on a phos-tag gel in an approach similar to what we did for HIF1α (Fig 4D), and did not observe any changes in its motility, although an important caveat from our experience is that phos-tag gels only resolve phosphorylations on roughly half of the proteins we have tested and know to be phosphorylated. For these reasons, we focused on AMPK and HIF1α. For AMPK and HIF1α we see both elevated phosphorylation and elevated activity upon PPP2R5C knockdown. Both AMPK and HIF1α can acutely increase glycolytic flux in response to stress conditions—either reduced energy supply or impaired mitochondrial function [41,42]–and therefore the two may act in concert to drive glucose uptake and glycolysis upon PPP2R5C knockdown. Indeed, the functional role of AMPK has been carefully studied in vivo in mouse liver, with increased liver AMPK activity leading to decreased blood glucose and fatty liver [43] and reduced liver AMPK activity leading to glucose intolerance and hyperglycemia during fasting [44], in agreement with what we see here. The phenotypes of liver-specific PPP2R5C knockdown and AMPK activation do not overlap completely, as expected if PPP2R5C also affects other metabolic pathways in addition to AMPK. It is unclear to what extent HIF1α could be contributing to the phenotypes observed in PPP2R5C HepKD mice, given that the mice we were studying were housed under normoxia. To our knowledge, whether HIF1α plays a role in mouse liver metabolism in such circumstances has not been studied. One study has shown that HIF1α can drive glycolysis and lipogenesis in cancer cells also under normoxic conditions [45]. Mice with liver-specific ablation of HIF1β develop diabetic phenotypes under normoxia [46], however HIF1β also binds other partners besides HIF1α. In our cell culture experiments with Hepa 1–6 cells, also conducted under normoxia, HIF1α is indeed functionally relevant because we can detect HIF1α protein by western blot (Fig 4B’ and 4D) and we see induction of HIF1α target genes upon PPP2R5C knockdown (Fig 4E). HIF1α is known to be frequently up-regulated and functionally relevant in cancers, therefore HIF1α may be a relevant downstream PPP2R5C target in this pathological context. Of note, however, although we see activation of AMPK and HIF1α upon PPP2R5C knockdown, due to technical limitations we have not been able to directly assay the contributions of these two targets to the PPP2R5C knockdown phenotype, for instance by performing genetic epistasis experiments. Further work will be required in this regard.
We observe activation of SREBP-1 in PPP2R5C HepKD livers, which likely contributes to their increased lipogenesis. SREBP-1 can be activated downstream of HIF1α [47] but additional mechanisms are likely to link PPP2R5C to SREBP-1 activation. We also observe increased expression of some, but not all ChREBP targets that we tested in PPP2R5C HepKD livers. This suggests elevated ChREBP activity could also be contributing to the increased lipogenesis in PPP2R5C HepKD livers. Unfortunately, however, we could not find a good antibody to detect endogenous ChREBP to test whether ChREBP might be a direct PPP2R5C target. Another important regulator of lipid metabolism is PPARα, which ppromotes fatty acid beta-oxidation. Although we observed reduced expression of a few PPARα targets, we do not see reduced fatty acid beta-oxidation in Hepa 1–6 cells upon PPP2R5C knockdown, suggesting that reduced PPARα activity might not contribute to the PPP2R5C knockdown phenotype.
PPP2R5C has been linked to cancer development. One mechanism appears to be via dephosphorylation of p53 on Thr55 [22–24]. The results described here suggest that upon PPP2R5C knockdown, cells also increase their glucose uptake, glycolytic rate, and lipid biosynthesis—all of which are metabolic hallmarks for cancer cells. Therefore, it will be interesting to study in the future whether these metabolic effects might be contributing towards the tumor suppressive properties of PPP2R5C.
One phenotype requiring further investigation is that liver triglycerides drop in PPP2R5C HepKD mice upon starvation compared to the random feeding regimen, instead of increasing as they do in control animals. Upon starvation, adipose tissue mobilizes lipid stores, releasing them as free fatty acids into circulation. These are taken up by the liver and converted to triglycerides, leading to liver steatosis upon starvation. Together with the stored de-novo synthesized triglycerides, they are packaged and secreted as VLDL particles. Some of these particles are re-uptaken by the liver, whereas others travel to other organs. One explanation of the observed phenotype is that HepKD livers secrete elevated levels of VLDL, thereby depleting TAGs stored in liver. A second possible explanation is that PPP2R5C HepKD livers do not efficiently re-uptake VLDL that is in circulation, leading to reduced liver triglycerides and elevated VLDL. A third possible explanation is that PPP2R5C HepKD livers do not uptake sufficient fatty acids from circulation upon starvation. However, this does not explain why VLDL levels would be elevated, and would predict free fatty acid levels in circulation should be elevated upon starvation, which we do not see (S3I Fig). A fourth possible explanation is that peripheral organs in PPP2R5C HepKD mice uptake less VLDL upon starvation. This, however, does not explain why liver triglyceride levels drop in PPP2R5C HepKD mice, and would require a secreted signaling molecule from the knockdown liver to peripheral tissues to alter their behavior. A fifth possible explanation is the HepKD livers have elevated fatty-acid beta-oxidation, thereby depleting TAG stores. However, this also would not explain the elevated VLDL levels. Hence, in sum, we think the most likely explanations are that PPP2R5C HepKD livers either secrete elevated levels of VLDL or re-uptake reduced levels of VLDL, but further work will be required to look at this aspect carefully. A second phenotype requiring further investigation is that glycogen levels in PPP2R5C HepKD livers do not drop upon starvation as in control animals. Since our analysis of PPP2R5C targets by BioID was done in the ‘fed’ state (ie cells growing with glucose and insulin), a similar analysis in the ‘starving’ state might shed more light on both of these starvation-associated phenotypes.
We previously identified the fly homolog of PPP2R5C, PP2A-B’, as a metabolic regulator in the fly, which dephosphorylates S6K [17]. Interestingly, there are both similarities and differences between our results in the fly and in the mouse. One similarity is that PPP2R5C and PP2A-B’ both appear to bind S6K. One difference is that we do not observe an increase in S6K phosphorylation upon PPP2R5C knockdown in mouse hepatocytes. Since we previously observed an increase in S6K phosphorylation upon PPP2R5C knockdown in HeLa cells [17], this difference is less likely due to evolutionary divergence and more likely due to incomplete PPP2R5C knockdown in our setup. A second similarity is that both in mice and in flies PPP2R5C/PP2A-B’ regulates organismal metabolism. However, PPP2R5C liver knockdown has opposite consequences to the full-body PP2A-B’ knockout which makes flies lean. This may be due to tissue-specific effects. For instance, PP2A-B’ knockdown in fly muscle might contribute towards fly leanness be elevating metabolic rate, as could be expected by elevated S6K activity. If this effect predominates in the whole fly knockout, it would make the fly lean. Further work will be required to further understand these similarities and differences.
Our knockdown experiments show that PPP2R5C expression in liver inhibits glucose uptake and reduces insulin sensitivity. Astoundingly, we find that increasing PPP2R5C expression in human liver also correlates with insulin resistance. Type-2 diabetic patients have significantly elevated PPP2R5C liver expression levels compared to controls (Fig 6A). In fact, even within the control population PPP2R5C expression correlates with reduced insulin sensitivity (Fig 6C), raising the hypothesis for future studies that increased PPP2R5C expression might play a causative role in insulin resistance.
In sum, we identify we identify here PPP2R5C as a novel metabolic regulator in the liver.
For transient cell culture knockdowns, shRNA targeting a common region of all PPP2R5C transcript variants, or a non-targeting scrambled shRNA, were packaged into adenovirus by using the BLOCKiT adenoviral RNAi system from Invitrogen (K4941-00) and produced in HEK293A cells. Adenovirus was purified from HEK293A lysates with a CsCl gradient and checked for titer using a TCID50 assay. For hepatocyte-specific PPP2R5C knockdown in mouse liver, miRNA targeting PPP2R5C, or a non-targeting scrambled miRNA, were packaged into adeno-associated virus (AAV) as described [30,48,49]. AAV production and purification was done by Vector Biolabs. Target sequences are listed in a table below.
Two independent shRNAs were cloned into an inducible piggyBac shRNA expression system (System Biosciences) and transfected into Hepa 1–6 cells. Single clones of stably transfected cells were selected with 3μg/ml puromycin for 2 weeks. Knockdown efficiencies were evaluated by qPCR and western blot analysis on individual clones and the best clones were selected. Target sequences are listed in a table below.
HEK293A, HEK293T and Hepa1-6 cells were maintained in DMEM medium with 10% FBS and 1x penicillin/streptomycin (100IU and 100ug/ml). HEK293A/T cells also required 1x Non-essential amino acids (Sigma M7145). Primary hepatocytes were isolated and cultured as described [50]. For virus infections, MOI (multiplicity of infeciton) of 100 was used and cells were assayed 3 days after infection.
RNA was isolated using Trizol (Invitrogen) and cDNA was generated by reverse transcription with RevertAid reverse transcriptase (Fermentas). Relative gene expression was measured using a StepOnePlus machine (Applied Biosystems). Expression profiling was performed using Illumina Mouse Sentrix-6 chips and the data were analysed in Chipster.
8–10 week C57BL/6J male mice were purchased from Charles River and maintained with unlimited water and normal chow food in a 12 hour light-dark cycle. After 1 week of adaptation, mice were tail vein-injected with 1x1011 viral particles/mouse of either control or knockdown AAV diluted in PBS. After 7 week of infection, mice were subjected to ad libitum feeding, 16 hour fasting or 16 hour fasting + 6 hour refeeding. Food intake for C57BL/6J mice was monitored in an automated metabolic measurement system (TSE Systems).
9 week db/db male mice were also purchased from Charles River and maintained under the same condition as C57BL/6J mice. For virus injection, 2x1011 viral particles/mouse were tail-injected. After 5 week of infection, db/db mice were subjected to ad libitum feeding and sacrificed. Mouse body weight and blood glucose levels were monitored regularly. Whole body composition analysis was performed for db/db mice using an Echo MRI system.
For assaying liver PPP2R5C expression in mice on a high-fat versus low-fat diet (S1E Fig), 8-week old male C57BL6/N were fed either a low fat diet (Research Diets Inc. D12450B) or a high fat diet (Research Diets Inc. D12492) for 4 weeks after which they were sacrificed in the fed state and total liver RNA was extracted by Trizol and gene expression was quantified by quantitative RT-PCR.
Glucose levels in cell culture medium were measured using the Glucose HK assay kit from Sigma (GAHK20-1KT) whereas lactate was measured using the D/L-Lactic acid kit from Roche (11 112 821 035). Non-esterified fatty acids in culture media were first extracted with methanol-chloroform as previously described [51], and then measured with the Free Fatty Acid Fluorometric Assay Kit from Caymen Chemical (700310). Triglycerides were first extracted from cell lysate or tissues with methanol-chloroform as previously described [51] and then digested with lipase (Genzyme, LIPA-70-1461) overnight. The released glycerol was measursed using the Free Glycerol Kit from Sigma (F6428-40ML). Glycogen was extracted from liver in 30% KOH at 95°C for 30 min, and then precipitated in 60% EtOH and re-suspended in water for measurement. Glycogen was first converted into glucose by overnight amyloglucosidase digestion (Sigma 10115) and the resulting glucose was quantified with the Glucose HK kit from Sigma (GAHK20-1KT). Cholesterol was first extracted from tissues with methanol-chloroform as previously described [51], and then measured using the Cholesterol kit from Randox (CH201). Glycolysis rate was measured using the Seahorse glycolysis stress kit on a XF96 analyzer (Seahorse bioscience)
C57BL/6J mice 4 weeks after infection with control or knockdown AAV were starved for 6 hours and then injected intraperitoneally with glucose (2g/kg body weight). Blood glucose was measured at 0, 20, 60, 90, 120, and 150 min after injection using a glucose meter from One-Touch (Lifescan). Circulating insulin was measured from 50 μl of blood at 0, 20, and 60 min after glucose injection. For Insulin sensitivity test, C57BL/6J mice with 4 week PPP2R5C HepKD were starved for 6 hours, and then 1IU/kg insulin (Humulin, Eli Lilly) was injected via the tail. Liver samples were collected 10 min after injection. Pyruvate tolerance test was done similarly to the glucose tolerance test. 2g/kg pyruvate was injected intraperitoneally and blood glucose levels were monitored at 0, 20, 40, 60, 90, 120, and 180 minutes after injection.
Db/db mice with 4 week of virus infection were subjected to insulin tolerance test. 1.5IU/kg Insulin (Humulin, Eli Lilly) was injected intraperitoneally and blood glucose levels were monitored at 0, 20, 40, 60, 90, 120, 150, 180 and 240 minutes after injection.
200μl of pooled serum from 5 or 6 mice was mixed with 100μl PBS and then injected onto a Superose 6 10/300 GL column in an AKTA FPLC purifier for size exclusion chromatography. Separated VLDL, IDL, LDL and HDL were collected in 500μl/fraction and 160 ul and 40 ul of each fraction was used to measure triglyceride and cholesterol levels as described above.
Serum insulin was measured with an ELISA kit from Alpco (80-INSMSU-E01). Serum ALT levels were measured with the Infinity ALT/GPT Reagent (Thermo Scientific TR71121).
Stable cell lines with inducible shRNAs or control hepa 1–6 cells were induced with 30μg/ml cumate for 3–4 days and then starved in serum-free DMEM overnight. Then cells were sensitized in KRPH buffer (20 mM HEPES, 5 mM KH2PO4, 1 mM MgSO4, 1 mM CaCl2, 136 mM NaCl, 4.7 mM KCl, pH to 7.4) for 1 hour and treated with 100 μM 2NBDG for 20 min to allow 2NBDG uptake. Uptake was stopped by washing with PBS 3 times. Cells were then trypsinized for 3 min to detach them, and trypsinization was stopped by adding an equal volume of fetal bovine serum. All cells were resuspended and washed in PBS with 2% FBS for 3 times before FACS measurement. 2NBDG intensity was recorded on BD’s FACSCanto II. FACS data were analysed either in FlowJo or R.
For measurement of glucose consumption from medium, Hepa 1–6 cells were infected by adenovirus carrying shRNA targeting all mouse PPP2R5C isoforms (PPP2R5C KD) or a negative-control scramble shRNA (Control KD). After 48h, cells were given fresh DMEM medium. Glucose levels in the medium were quantified using the Glucose HK assay kit from Sigma (GAHK20-1KT) both prior to incubation with the cells, as well as after 24 hours. Glucose consumption was normalized to total cell protein.
Hepa 1–6 cells were detached by trypsinization, washed once with 180 μL assay medium (DMEM (sigma D5030) with 143mM NaCl, 2 mM L-Glutamine, pH 7.35 ± 0.05. Adjust pH at day of assay) and incubated with 175 μL assay medium at 37°C for 1 hour in XF96 glycolysis stress kit 96-well plates (Seahorse Bioscience). Then 25 μL of glucose (10 mM in assay medium), oligomycin (2.5 or 1 μM in assay medium) and 2-Deoxyglucose (100 mM in assay medium) each were injected into the plate reservoir and glycolysis rate was recorded on XF96e Extracellular Flux Analyzer from Seahorse Bioscience. All the data collection and analysis was done on XF96 built-in software. Data were normalized to cell number, counted using a Cell Profiler after DAPI staining.
Fatty acid beta-oxidation activity in Hepa 1–6 was measured using the Mito Stress kit (Seahorse Bioscience) +/- Etomoxir treatment. Hepa 1–6 cells were replated in XF96 cell culture microplates 24 hours before the experiment and incubated with substrate-limited medium (DMEM with 0.5 mM glucose, 1 mM GlutaMAX, 0.5 mM carnitine, and 1% FBS.). Cells were washed in FAO assay medium (111 mM NaCl, 4.7 mM KCl, 1.25 mM CaCl2, 2 mM MgSO4, 1.2 mM NaH2PO4, 2.5 mM glucose, 0.5 mM carnitine, and 5 mM HEPES pH 7.4) 45 minutes before the experiment. OCR (oxygen consumption rate) measurement was performed by using the Mito Stress kit with 2uM Oligomycin, 2μM FCCP and 0.5μM Retenone/Antimycin A. 400 μM Etomoxir was used to inhibit beta-oxidation activity.
Insulin sensitivity was assessed with the euglycemic-hyperinsulinemic clamp method using a previously described protocol [52]. In brief, after an overnight fast and resting for 30min in supine position, intravenous catheters were inserted into antecubital veins in both arms of the study participants. One was used for the infusion of insulin and glucose, the other was used for frequent blood sampling. After a priming dose of 1.2nmol/m2 insulin, infusion with insulin (Actrapid 100 U/ml, Novo Nordisk, Bagsvaerd, Denmark) was started with a constant infusion rate of 0.28nmol/m2 body surface per min and continued for at least 120min. After 3min, the variable glucose infusion rate (20% glucose) was added and adjusted during the clamp to maintain the blood glucose at 5.0mmol/l. Bedside blood glucose measurements were carried out every 5min. Glucose infusion rate (GIR) was calculated from the last 45 min of the clamp, in which glucose infusion rate could be kept constant in order to achieve the target plasma glucose concentration of 5.0(± 5%) mmol/l. Therefore, the duration of the clamp varied between individuals (range 120-200min). In premenopausal women, clamp studies were performed during the luteal phase of the menstrual cycle.
PPP2R5C mRNA expression was investigated in liver tissue samples obtained from 66 Caucasian men and women (BMI range: 22.7–45.6 kg/m2) with (n = 26) or without (n = 40) type 2 diabetes who underwent open abdominal surgery for Roux-en-Y bypass, sleeve gastrectomy, explorative laparotomy or elective cholecystectomy. A small liver biopsy was taken during the surgery, immediately frozen in liquid nitrogen, and stored at −80°C until further preparations. All baseline blood samples were collected between 8 and 10 am after an overnight fast. All study protocols have been approved by the ethics committee of the University of Leipzig (363-10-13122010 and 017-12-230112) in accordance with the principles of the WMA Declaration of Helsinki. All participants gave written informed consent before taking part in the study. Human PPP2R5C mRNA expression was measured by quantitative real-time RT-PCR in a fluorescent temperature cycler using TaqMan assay-on-demand kits (Hs00604899_g1, Applied Biosystems, Darmstadt, Germany), and fluorescence was detected on an ABI PRISM 7000 sequence detector (Applied Biosystems, Darmstadt, Germany). PPP2R5C mRNA expression was calculated relative to the mRNA expression of 18S rRNA (Hs99999901_s1, Applied Biosystems, Darmstadt, Germany).
Adipose tissue samples were obtained from a biobank collection of the University Hospital Joan XXIII (Tarragona, Spain). The Hospitals’ ethics committee approved the study and written informed consent was obtained from all participants. We used paired subcutaneous and visceral adipose tissue samples from healthy and type 2 diabetic patients, matched for age, gender and body mass index (see S6A Fig). All subjects were Caucasian and reported steady body weight for at least 3 months prior to the study. Subjects were scheduled for an elective surgical procedure (cholecystectomy or surgery for abdominal hernia); and had no metabolic diseases other than type 2 diabetes. They had been free of any infections in the month preceding the study. Exclusion criteria were: presence of liver or renal diseases, malignancy, chronic inflammatory disease and pharmacological treatments that may alter the lipid profile. All patients had fasted overnight, for at least 12 h, before surgical procedure. Adipose tissue samples were obtained during the surgical procedure, washed in PBS, immediately frozen in liquid N2 and stored at -80°C.
Total RNA was extracted from adipose tissue using the RNeasy lipid tissue midi kit (QIAGEN Science). One microgram of RNA was reverse transcribed with random primers using the reverse transcription system (Applied Biosystems). PPP2R5C (Hs00604899_g1) quantitative gene expression was evaluated with TaqMan low-density arrays (Applied Biosystems; microfluidic cards). Results were calculated using the comparative cycle threshold (Ct) method (2-ΔΔCt) and expressed relative to the expression of the housekeeping genes cyclophilin 1A (PPIA).
For immunoprecipitation, cells were lysed in IP buffer (150mM NaCl, 50mM Tris pH7.5, 1% Triton X-100) clarified by centrifugation at 14,000 rpm for 15 minutes, then incubated with primary antibody for 2 hours and then incubated with Protein A-Agarose beads (Roche) for 30 min before washing in IP buffer 3 times. Protein samples for immunoblotting were lysed in 1x Lamelli buffer and subjected to SDS-PAGE with a tris-glycine buffer system. Rabbit antibodies against phospho-GSK3β(S9), phospho-S6K(T389), S6K, phospho-AKT(T308), phospho-AKT(T473), AKT, phospho-AMPKα(T172), AMPKα, AMPK β1, phospho-ACC1, ACC1, phospho-TBC1D1(S700), TBC1D1, HSP90, YAP, TSC1, and Rpl26 were from Cell Signaling. PPARα antibody (3B6/PPAR) was purchased from Alexis (Now part of Enzo life sciences). LXRα/β was from Santa Cruz. Mouse antibody anti-SREBP-1 was from BD Biosciences. Anti-HA antibody was from Roche. Anti-FLAG(M2) was from Sigma. Monoclonal mouse anti-Actin was from Developmental Studies Hybridoma Bank. Anti-PPP2R5C was generated by immunizing guinea pigs with NCBI Variant 3 of mouse PPP2R5C.
Potential PPP2R5C substrates were identified by proximity biotinylation using the BioID method described in [33]. PPP2R5C was fused to a mutated biotin ligase (BirA R118G) either to its C terminal (Myc-BirA-PPP2R5C) or N terminal ends (PPP2R5C-BirA-HA). These were co-transfected with or without a catalytic dead version (D85N) of the PP2A catalytic C subunit [34]. After 24 hours of expression, cells were washed with PBS twice and lysed in BioID lysis buffer (50 mM Tris, pH 7.4, 500 mM NaCl, 0.4% SDS, 5 mM EDTA, 1 mM DTT, 2% Triton X-100, and 1x Complete protease inhibitor (Roche)). The lysates were mixed with an equal volume of 50mM Tris (pH 7.4) before incubating with pre-washed streptavidin magnetic beads (Invitrogen MyOne Streptavidin C1) overnight at 4°C. Then the beads were washed twice with 2% SDS, once with BioID wash buffer 1 (0.1% deoxycholate, 1% Triton X-100, 500 mM NaCl, 1 mM EDTA, and 50 mM Hepes, pH 7.5), once with BioID wash buffer 2 (250 mM LiCl, 0.5% NP-40, 0.5% deoxycholate, 1 mM EDTA, and 10 mM Tris, pH 8.1), and twice with wash buffer 3 (50 mM Tris, pH 7.4, and 50 mM NaCl). Finally, all biotinylated proteins were eluted in 1x Laemmli buffer with saturated biotin (around 1mM) at room temperature for 15min and following 15 min boiling at 95°C. Eluted proteins were probed by immunoblotting.
Mouse PPP2R5C variant 1 and AMPK beta1 were tagged with HA and FLAG tags respectively in pcDNA3 and transfected into Hepa 1–6 cells. Two days after transfection, cells were fixed in 4% PFA and permeabilized in 1x PBX (1xPBS, 0.2% Triton X-100) and stained with anti-HA and anti-FLAG antibodies as per manufacturer instructions (Duolink). Duolink in situ detection reagent red (Duolink) was employed to detect the AMPK beta1-PPP2R5C interaction signal. Nuclei were stained with DAPI.
The Student t-test was employed to test the significant difference in various mRNA levels or metabolic phenotypic data among different nutritional statuses, or between control knockdown and PPP2R5C knockdown, or human liver PPP2R5C mRNA levels between healthy control and type 2 diabetic patients. Human adipose tissue PPP2R5C mRNA levels was tested by Mann-Whitney test. Liver triglyceride was by one-way ANCOVA with liver NEFA as covariate. Time course related studies, including GTT, ITT, PTT, seahorse experiments, body weight or glucose changing profiles, were analysed by 2-way ANOVA with time and PPP2R5C knockdown as factors. The difference at individual time points for GTT was also tested by Wilcoxon signed-rank test. Correlation analysis between human liver PPP2R5C mRNA levels and glucose infusion rate was determined by Pearson’s correlation method. All data analysis was performed in R 3.0.1.
Sequences of oligos, shRNAs and miRNAs used are provided in S1 Table.
For PPP2R5C expression in human liver, all study protocols were approved by the ethics committee of the University of Leipzig (363-10-13122010 and 017-12-230112) in accordance with the principles of the WMA Declaration of Helsinki. Adipose tissue samples were obtained from a biobank adipose tissue collection (internal approval code: 54c/2009) of the University Hospital Joan XXIII (Tarragona, Spain), according to the biomedical research law 14/2007 from Spain, and with written informed consent from participants. Animal experiments were conducted according to local, national, and EU ethical guidelines and approved by local regulatory authorities (Regierungspräsidium Karlsruhe).
|
10.1371/journal.ppat.1000277 | The Cell Adhesion Molecule “CAR” and Sialic Acid on Human Erythrocytes Influence Adenovirus In Vivo Biodistribution | Although it has been known for 50 years that adenoviruses (Ads) interact with erythrocytes ex vivo, the molecular and structural basis for this interaction, which has been serendipitously exploited for diagnostic tests, is unknown. In this study, we characterized the interaction between erythrocytes and unrelated Ad serotypes, human 5 (HAd5) and 37 (HAd37), and canine 2 (CAV-2). While these serotypes agglutinate human erythrocytes, they use different receptors, have different tropisms and/or infect different species. Using molecular, biochemical, structural and transgenic animal-based analyses, we found that the primary erythrocyte interaction domain for HAd37 is its sialic acid binding site, while CAV-2 binding depends on at least three factors: electrostatic interactions, sialic acid binding and, unexpectedly, binding to the coxsackievirus and adenovirus receptor (CAR) on human erythrocytes. We show that the presence of CAR on erythrocytes leads to prolonged in vivo blood half-life and significantly reduced liver infection when a CAR-tropic Ad is injected intravenously. This study provides i) a molecular and structural rationale for Ad–erythrocyte interactions, ii) a basis to improve vector-mediated gene transfer and iii) a mechanism that may explain the biodistribution and pathogenic inconsistencies found between human and animal models.
| In most cases, adenoviruses are thought to initially enter the host via contact with epithelial cells and spread within the host via an unknown mechanism. Most adenovirus serotypes use a cell adhesion molecule dubbed “CAR” to attach to cells. To assess, predict and understand adenovirus biology and vectorology, many in vivo studies use mice and monkeys. These animal models have been considered reliable models in the realm of viral pathogenesis and gene transfer. One of the implications of our study suggests that the rat may be a more appropriate model during intravenous adenovirus delivery because like humans, and unlike mice and monkeys, they also express CAR on their erythrocytes. The identification of CAR on human erythrocytes explains a 50-year-old enigma of adenovirus hemagglutination, helps us better understand adenovirus in vivo biology and may open new avenues to understand the role of cell adhesion molecules during erythropoiesis.
| Adenoviruses (Ads) are nonenveloped double-stranded DNA pathogens that infect all vertebrate classes. To date, all Ads display a characteristic, icosahedral symmetry in which 240 subunits of the trimeric hexon protein form the facets and 12 copies of the penton, comprising the pentameric penton base protein and the externally projecting trimeric fiber, form the vertices [1]. While the stoichiometry of the penton base and hexon is apparently conserved, the fiber can exist as a single or double copy at each vertex [2]. At least in vitro and for most cell types, the fiber mediates the initial attachment to primary receptors, such as the D1 domain of the coxsackievirus and adenovirus receptor (CAR), sialic acids, CD46, and others (for review see Zhang & Bergelson [3]). Interaction with auxiliary receptor(s), in particular some of the dimeric integrins via the Arg-Gly-Asp sequence (an integrin-interacting motif) on the penton base, may induce internalization of some serotypes. However, other auxiliary receptors or mechanism of internalization may exist for human serotypes 40 and 41 (HAd40/41), and canine serotype 2 (CAV-2), which have no identifiable integrin-interacting motif in the penton base [4],[5].
Greater than 150 Ad serotypes have been isolated. Approximately 50 of these are currently classed as human pathogens that, in most cases, generate subclinical ocular, respiratory and gastrointestinal tract infections. In the immunocompromised host however, lethal HAd infections can spread, via unknown mechanisms, to the kidney, liver and brain [6],[7]. The human Ads (HAds) are divided into subgroups (or species or subgenera) A - F. The triage of the human serotypes into subgroups is based in part on serotype-specific ex vivo erythrocyte cross-linking (or hemagglutination) [8]. The clinical hemagglutination assays use lysates from virus-infected cells to crosslink erythrocytes from a handful of species. This highly heterogeneous lysate contains whole virus particles, empty capsids, penton monomers, penton dodecahedrons, fiber monomers, hexon etc. By using fractionated infected cell lysate, a handful of laboratories found that the multivalent complexes containing fiber were responsible for hemagglutination [9]–[12].
Erythrocyte membranes contain highly sialated glycoproteins and glycolipids. One of the most abundant glycoproteins on erythrocytes is glycophorin A (∼105 copies/cell). With its high sialic acid content, glycophorin A is the main contributor to the net negative cell-surface charge and is critical for minimizing cell–cell interactions and preventing erythrocyte aggregation [13]. Sialic acid is a collective term for a family of 9-carbon monosaccharides, which are often found as terminal sugar residues on glycans of glycoproteins and glycolipids (usually α2-3, -6 or -8 linked). In addition to some HAds, a number of viruses, including orthomyxoviruses, paramyxoviruses, picornaviruses, papovaviruses, coronaviruses, reoviruses and parvoviruses bind to sialic acids [14]–[18]. Possibly because erythrocytes from different species vary in their sialic acid content, the hemagglutination properties of sialic acid-binding viruses may also diverge [19].
HAd subgroup D (serotypes 9, 15, 19p and 37) and B:2 (serotypes 11a, 11p and 34a) erythrocyte binding depends on the fiber head and several attempts have been made to define the region(s) responsible [20]–[23]. Among the HAds, serotype 37 is unusual: its fiber head can bind CAR, CD46 and sialic acid, but the virus appears to use only the latter two as functional receptors [24]–[26]. Burmeister et al. found that the sialic acid moiety bound to a basic patch close to the center of the trimeric fiber heads of HAd37 and HAd19p [27]. Interestingly, the putative hemagglutination domain in subgroup D heads partially aligns with the sialyl-lactose binding site, which on the basis of sequence alignments is likely to be conserved in other members of this subgroup. These observations raise the question as to whether Ad-erythrocyte interaction is a consequence of fiber head-sialic acid interaction.
In this study we initially characterize the erythrocyte binding of unrelated serotypes, HAd5, HAd37 and CAV-2. Although both HAd37 and CAV-2 agglutinate human erythrocytes at low particle-to-cell ratios, they use different receptors (sialic acid and CD46 vs. CAR) [5],[25],[26], have different clinical tropisms (ocular vs. respiratory tract) and infect different species. We found that HAd37-erythrocyte interaction is primarily due to sialic acid binding. We show by structural analysis that the CAV-2 fiber head also contains a sialic acid binding site, but in contrast to HAd37 this site is modestly involved in hemagglutination. Unexpectedly, our biochemical and competition analyses suggested that CAV-2-erythrocyte interactions also depend on binding to CAR on human and rat erythrocytes. Using a transgenic mouse that expresses CAR on erythrocytes [28] we demonstrate that CAR-binding Ads can be sequestered by CAR-expressing erythrocytes, and prevent liver infection. Our study provides a molecular and structural rationale for the 50-year-old enigma of ex vivo Ad-erythrocyte interactions. In addition to the relevance for Ad pathogenesis and vector biology, the expression of CAR by human erythrocytes may shed light on the role of cell adhesion molecules during erythropoiesis.
Consistent with previous reports, we found that at a modest physical particle (pp)-to-erythrocyte ratio (∼130) HAd37 efficiently agglutinates human erythrocytes (Figure 1A). When compared to HAd37, CAV-2 hemagglutinates at ∼20-fold lower ratio (i.e. more efficiently, Figure 1B). In our hands, and consistent with others [29], HAd5 also agglutinated human erythrocytes, but at a higher ratio (>800 pp/erythrocyte) (Figure 1C). To assay the roles of HAd37 and CAV-2 capsid proteins, we incubated erythrocytes with chimeric HAd5 vectors harboring the fiber from HAd37 [30] or the fiber head from CAV-2 [31]. We found that the HAd5-HAd37F and HAd5-CAV-2H hybrid capsids induced agglutination at lower ratios than HAd5 but not quite to those of HAd37 and CAV-2 (Figure 1D and E). Pre-incubating erythrocytes with neuraminidase, which removes sialic acid from their membranes, eliminated agglutination by HAd37 and a HAd5-HAd37F hybrid capsid (Figure 1, right hand column). In contrast, removing sialic acid from erythrocyte membranes only modestly reduced CAV-2 and the hybrid HAd5-CAV-2H agglutination (data summarized in Table 1).
Together, these data suggest that the CAV-2 fiber head, like in subgroup D and B:2 HAds, is responsible for hemagglutination and that sialic acid binding plays a more significant role in HAd37 binding of human erythrocytes than it does for CAV-2.
To mimic the hemagglutination caused by HAd37 fiber head in a virion-like context, we generated a multivalent protein complex similar to the “complete hemagglutinin” described for HAd9 by Norrby et al. [9]. For this purpose HAd3 penton dodecahedra were incubated with chimeric “mini fibers” consisting of HAd3 fiber tail and shaft motifs 1+2 and HAd37 fiber head. In addition, we generated HAd37 fiber heads containing mutations in the sialic acid binding site [27] (see Table 2 for a list of fiber head mutants). The HAd3 penton dodecahedra without a fiber did not cause hemagglutination (data not shown). The dodecahedra containing a wild type HAd37 fiber head (HAd37Hwt) (Figure S1) agglutinated erythrocytes at low protein concentrations (about 104-fold less mass than HAd37 virions) (Table 2). Similar to HAd37, this “complete hemagglutinin” poorly agglutinates neuraminidase-treated erythrocytes (Table 2). The dodecahedra containing a HAd37 fiber with a Lys to Glu mutation at amino acid 345 in the sialic acid binding site (HAd37HSA−1), had at least a 105-fold reduced hemagglutination activity compared to the dodecahedra containing HAd37Hwt. The chimeric wild type CAV-2 fiber head (CAV-2Hwt) did not bind to HAd3 penton dodecahedra, which precluded equivalent experiments.
These data suggest that the HAd37 fiber shaft, hexon and pIX do not play key roles during binding and that the sialic acid binding site is involved in agglutination of human erythrocytes.
To assay the binding of recombinant fiber heads, we incubated HAd37Hwt and mutant fiber heads with erythrocytes and then added anti-fiber head antibodies or antiserum. We then quantified attachment by flow cytometry. HAd37Hwt (Figure 2A) and mutants carrying point mutations in the CAR binding site (HAd37HCAR−1 and HAd37HCAR−2, mutation in Glu351 and Ser299, data not shown) [24] bound to erythrocytes in a dose-dependent manner. However, HAd37HSA−1 and HAd37HSA−2 (a Tyr to Ala mutation at amino acid 312), which harbor mutations in the sialic acid binding site, poorly bound erythrocytes (Figure 2A). In addition, all of the HAd37 fiber heads poorly bound neuraminidase-treated erythrocytes (Figure 2A, bottom row and data not shown).
Together, these data suggest that the HAd37 fiber head sialic acid binding site is a key determinant in the attachment to human erythrocytes.
Using the above approach, we detected minimal binding of CAV-2Hwt to erythrocytes (data not shown), possibly because the affinity of single CAV-2 fiber heads is low, the high affinity polyclonal antibody binding out-competed the weaker erythrocyte binding, and/or the number of CAV-2 fiber head receptors on erythrocytes is low. To address the potential competition, we incubated fluorescently labeled CAV-2 fiber heads with erythrocytes to circumvent the use of antibodies. Here we detected low, but reproducible, binding of CAV-2Hwt to mock-, as well as neuraminidase-treated, erythrocytes (Figure 2B).
Together, these data suggest that CAV-2 erythrocyte binding is notably less dependent on sialic acid than HAd37, and the lower binding may be due to reduced affinity or fewer receptors.
To address sialic acid-fiber head interaction using a cell-free system, glycophorin or asialoglycophorin (asialoGP) were immobilized on BIAcore sensor chips and increasing concentrations of HAd37 fiber heads were assayed. All binding curves had a square shape, suggesting high on and off rates (Figure S2). HAd37Hwt bound to glycophorin, but less efficiently to asialoGP. Compared to HAd37Hwt, HAd37HSA−2 bound equally to asialoGP, but less to glycophorin. Interestingly, HAd37HSA−1 (Lys345Glu) bound less efficiently to glycophorin and asialoGP compared to HAd37Hwt and HAd37HSA−2, which suggests that HAd37 agglutination may be charge-dependent.
Consistent with the erythrocyte binding data, we found very little binding between CAV-2Hwt and glycophorin or asialoGP (data not shown). Overall, the SPR results reflect the results obtained using flow cytometry with mock or neuraminidase-treated erythrocytes.
Our results suggested that HAd37 hemagglutination was due to sialic acid binding via the sialic acid binding site in the fiber head. Although we have no evidence suggesting that CAV-2 can use sialic acid as a functional receptor, it is possible that the efficient CAV-2 agglutination of human erythrocytes is due to multiple and coordinated binding of sialic acid. For example, each of the 12 homotrimeric fiber heads on the end of the flexible CAV-2 shaft [32] could bind three sialic acid moieties (theoretically up to 36/capsid). Because the predicted subgroup D fiber head hemagglutination domains (in the CD and GH loops) appear to be well conserved, we tried to identify the corresponding domain in the CAV-2 fiber head using a mutagenesis strategy based on sequence homology (Figure S3). This approach was unsuccessful (data not shown), suggesting that Ad hemagglutination sites are not strictly conserved.
To address possible sialic acid binding via an alternative approach, we soaked CAV-2 fiber head crystals in solutions containing 2-3 sialyl-D-lactose. The crystal diffracted to 1.9 Å and the structure was solved by molecular replacement using our model of CAV-2 fiber head [24] (crystallographic details are summarized in Table S1). The electron density maps showed clear density for six N-acetyl neuraminic acid (Neu5AC) moieties in the asymmetric unit, three per fiber head trimer. Sialyl-lactose is composed of three sugar rings, the sialic acid moiety being linked to lactose (galactose-glucose). The density for the lactose moieties was weak, implying that these are flexible within the crystal. Only one of the six galactose rings in the asymmetric unit could be modeled. This molecule is located in between the two fiber head trimers in the asymmetric unit and forms a distant hydrogen bond (distance 3.25 Å) to Lys503 via the galactose oxygen 6. It is unlikely that this interaction is enough to immobilize the galactose ring because the other five galactose moieties in the asymmetric unit are not equally well visible in the electron density. More likely, the stabilization is due to the spatial restrictions imposed by the proximity of the other fiber head trimer.
We found that the sialic acid binding site on CAV-2 head is distinct, both in sequence and location, from that found on the HAd37 fiber head. On the CAV-2 fibers head, sialic acid binds further away from the three-fold symmetry axis at the periphery of the trimer (Figure 3A–D). The residues involved in binding (Asn435, Ser419, Ser416, Gln417 and Arg515) do not align with those involved in sialic acid-binding by HAd37 fiber head (Figure S2). The HAd37 sialic acid binding site consists of three residues forming hydrogen bonds (Tyr213, Pro317 and Lys345) and two residues contacting sialic acid in hydrophobic interactions (Tyr308 and Val322) (Figure 3E). All seven interactions between CAV-2 fiber head and sialic acid are hydrogen bonds or salt bridges (Figure 3F), suggesting a relatively strong interaction compared to HAd37. Unlike HAd37 fiber head [33], no hydrophobic contacts contribute to sialic acid binding. Finally, sialic acid binds within a basic patch in each head (Figure 3C and D).
Our results showing that the CAV-2 fiber head contains a sialic acid binding site creates a paradox: both HAd37 and CAV-2 fiber heads contain CAR and sialic acid binding sites - but HAd37 uses sialic acid and CD46, while CAV-2 uses CAR (CAV-2 does not use CD46 to infect cells, unpublished data) to infect cells. To determine if sialic acid and CAR binding were mutually exclusive, we soaked CAV-2 or HAd37 fiber head crystals in complex with CAR in solutions containing sialyl-D-lactose. Crystals containing the complex with CAV-2 fiber head diffracted to 2.9 Å and contained 12 chains of fiber head bound to 12 chains of CAR D1 and 12 sialyl-D-lactose molecules in the asymmetric unit (space group I422). Crystals containing the complex with HAd37 fiber head diffracted to 1.55 Å and contained one chain of fiber head in complex with one CAR D1 molecule and one sialyl-D-lactose molecule in the asymmetric unit (space group I23). For both structures the electron density maps showed clear density for sialic acid, but not lactose, at the expected positions on the fiber heads (Figure 3G and H).
Together, our data demonstrate that CAV-2 and HAd37 sialic acid binding sites do not overlap with the CAR binding sites, and both fiber heads can bind CAR and sialyl-D-lactose simultaneously.
Arnberg and colleagues previously showed that HAd37 interaction with sialic acid was inhibited at high salt concentration [33]. To determine if a CAV-2-erythrocyte interaction was at least partially charge dependent, we incubated a CAV-2 vector expressing GFP (CAVGFP) or a HAd5 vector expressing GFP (AdGFP) with mock- or neuraminidase-treated erythrocytes in PBS containing increasing concentrations of ions. We then pelleted the erythrocytes by centrifugation, removed an aliquot of the supernatant, added it to cells, and assayed the cells for GFP expression by flow cytometry 24 hr post-infection. Using mock-treated erythrocytes (Figure 4), we found that at physiological salt concentrations ∼80% of CAVGFP was removed from the supernatant, while at higher salt concentration (300 mM NaCl) ∼60% of CAVGFP was removed. When using neuraminidase-treated erythrocytes, we again found that CAVGFP was efficiently removed from the supernatant. All the test samples were significantly (P<0.01) different from the control, as well as from the mock-treated 150 mM NaCl (P<0.05). Consistent with other studies [29],[34] AdGFP was also efficiently (90%) removed from the supernatant after the incubation with mock- (or neuraminidase-) treated erythrocytes.
These data suggest that, like HAd37, a modest degree of CAV-2-erythrocyte interactions may depend on electrostatic interactions. In addition, both the CAR-tropic adenoviruses (HAd5 and CAV-2) bind human erythrocytes at physiological salt concentrations.
While our results show that HAd37 hemagglutination is due primarily to sialic acid binding, CAV-2 hemagglutination appeared more complex. We therefore developed additional tests to address the molecular and structural basis. Due its sensitive and semi-quantitative potential, we returned to hemagglutination assays to understand the erythrocyte interactions.
Freshly purified CAV-2Hwt, which predominantly consists of individual trimeric fiber heads, did not cause hemagglutination, presumably because it is not sufficiently multivalent to crosslink erythrocytes. However, several His-tagged Ad fiber heads, including those of HAd41 short fiber [35] and HAd37, form multimers that dissociate into single trimeric fiber heads upon removal of the histidine tag. Similarly, His-tagged CAV-2 fiber head forms multimers of a defined size that are stable on a size-exclusion column and can be visualized with an electron microscope (not shown). Removal of the His-tag yields single fiber heads. We found that His-tagged CAV-2Hwt agglutinated erythrocytes, while His-tagged CAV-2HSA−1 and CAV-2HSA−2 (two CAV-2 fiber heads with one or two mutations in the sialic acid binding site, see Table 2) showed reduced hemagglutination titers. However, we could not exclude the possibility that the reduced hemagglutination in these latter CAV-2H constructs was due to electrostatic interactions (the mutations modified the charge of the fiber heads).
We next assayed CAV-2 hemagglutination using a competition assay (see Figure S4 for schema). In these competition assays, we pre-incubated erythrocytes with fiber heads, antibodies, salt and/or neuraminidase. Then the erythrocytes were incubated with CAV-2 and compared to mock-treated erythrocytes. We found that hemagglutination could be >256-fold reduced by pre-incubating CAV-2 with anti-fiber head antibodies, or pre-incubating erythrocytes with CAV-2Hwt (see Figure 5A for specific example and 5B for cumulative data). Pre-incubating the erythrocytes with the head from HAd5 or performing the assay in 225 mM NaCl led to modest 2 to 4-fold reductions. Unexpectedly, we found that like CAV-2Hwt, CAV-2HSA−1 notably reduced agglutination (∼16-fold), while a CAV-2 fiber head with a mutation in the CAR binding site (CAV-2HCAR−) had a modest ∼2-fold reduction. In most cases, pre-treating erythrocytes with neuraminidase had a fairly small additive effect of CAV-2 hemagglutination.
Together, these data suggest that the CAR binding site, which is present in CAV-2Hwt and CAV-2HSA−1 but not CAV-2HCAR−, is involved in CAV-2 agglutination of human erythrocytes. That the CAV-2 CAR binding site is involved in erythrocyte binding is consistent with the study by Nicol et al. [36], which showed that a CAR-ablated HAd5 vector no longer agglutinated human and rat erythrocytes.
The presence of CAR on erythrocytes would be inconsistent with other reports [34]. Among other functions, CAR acts as a homodimeric cell adhesion molecule at tight gap junctions [37]. However, the expression of cell adhesion molecules during erythropoiesis is not unprecedented [38]. Erythrocytes from some species express cell adhesion molecules during the early stages of differentiation that are thought to be involved in interaction with macrophages. We therefore incubated erythrocytes with anti-CAR antibodies that recognize the extracellular domain of CAR and assayed expression using flow cytometry. We found low, but reproducible, CAR expression on the cell surface of human erythrocytes (Figure 6A). To assay CAR expression using another approach, we used western blot analysis to screen erythrocytes from several species. Using an anti-CAR Ab that recognizes the cytoplasmic domain of CAR we found CAR expression on human and rat, but not on mouse, dog, rabbit and nonhuman primate erythrocytes (Figure 6B and data not shown). Again, our results showed a relatively low level of CAR on human erythrocytes, which is consistent with the low level of CAV-2Hwt binding to erythrocytes (Figure 2B).
These data suggest that CAR binding is a significant factor in CAV-2, and likely other CAR-tropic Ads, interaction with human erythrocytes. This also is consistent with the fact that CAV-2 agglutinates rat erythrocytes, but not erythrocytes from mice, dogs, rabbits or some nonhuman primates (as well as other species) (Figure S5).
If CAR binding plays a role in CAV-2 agglutination, then knocking down/blocking or artificially expressing CAR on erythrocytes should prevent/induce hemagglutination. Eliminating CAR on mature enucleated erythrocytes is technically challenging. Furthermore, to the best of our knowledge there are no known anti-CAR antibodies that completely block Ad attachment. On the other hand, CAR has been expressed on erythrocytes using a transgenic mouse line (GATA1-CAR) with the CAR cDNA downstream of the globin transcription factor 1 promoter [28]. Using flow cytometry and western blot analysis (Figure 6A & B), we compared the levels of human CAR expressed by GATA1-CAR and human erythrocytes. We then repeated the hemagglutination assays using erythrocytes from control (C57BL/6) and GATA1-CAR mice. We also found that CAV-2 agglutinated GATA1-CAR erythrocytes at a low particle-to-cell ratio (Figure 6C), while there was no agglutination of C57BL/6 erythrocytes. Equally important, competition assays with recombinant CAV-2 fiber heads gave profiles that were similar to when we used human erythrocytes (Figure 6D, and data not shown).
Together, our data demonstrate that CAR expression by erythrocytes can lead to agglutination by CAR-tropic adenoviruses.
In spite of recent notable advances [39],[40], in vivo Ad biodistribution, tropism and pathogenesis for the CAR-tropic HAds are still poorly understood. Group C HAd serotypes 2 and 5 are the prototype Ads in terms of structure, tropism and pathogenesis. However, tropism has been primarily studied in vitro, ex vivo or in animal models. Our results showing that human and rat erythrocytes harbor CAR on their external membranes while mice and nonhuman primates do not, suggests that these latter animals poorly mimic the in vivo environment that Ads encounter.
To better address in vivo biodistribution of CAR-tropic Ads, we injected GATA1-CAR and control C57BL/6 mice with a HAd5 vector and quantified viral genome blood half-life and tissue distribution. We found that the viral load in the blood was 1000-fold higher (P<0.01) in GATA1-CAR versus isogenic control mice (CAR-negative erythrocytes) during the first 72 hr post-injection (Figure 7A). These data are reminiscent of the studies where human blood cells are routinely positive by qPCR for wild type HAd sequences [6],[7]. In addition, notably absent from the AdGFP-injected GATA1-CAR mice was transgene expression in the liver. Lack of transgene expression was also consistent with the significant (P<0.01) ∼25-fold difference in the mean viral load as quantified by qPCR (Figure 7B). The lack of efficient liver infection is also consistent with the generally unexpected and poor infection of rat liver compared to mice and nonhuman primates following injection of CAR-tropic Ad vectors.
Together, these data suggest that CAR expression by rat and human erythrocytes plays a significant role in HAd in vivo distribution and, in turn, determining which tissues are susceptible to infection.
We initiated this study to examine a 50-year-old enigma of adenovirus biology. Our results demonstrate that HAd37 interaction with human erythrocytes is primarily due to sialic acid binding via a conserved sialic acid binding site on this subgroup D HAd fiber head. CAV-2, a serotype that like the prototype HAd5 is “CAR-tropic” and by most criteria is unrelated to HAd37, interacts with human erythrocytes via a mechanism depending on several factors, including most notably binding to CAR.
Most subgroup D HAd sialic acid binding site residues are conserved [27], and structural analysis suggest that they bind sialic acid in an equivalent fashion. Unexpectedly, we found that the CAV-2 fiber head harbors a sialic acid binding site. But, in contrast to the well-conserved location of the CAR binding domains on some Ad fiber heads, location of the sialic acid binding site is not conserved.
Following the structure-based identification of the amino acids in the CAV-2 and HAd37 fiber heads that interact with sialic acid, we introduced mutations to assay the role of these sites in sialic acid binding and erythrocyte interaction. In a number of conditions and approaches we showed that hemagglutination by HAd37 depends primarily on sialic acid binding. First, removing sialic acids from erythrocytes with neuraminidase eliminated erythrocyte cross-linking by HAd37 and a chimeric capsid harboring the HAd37 fiber head. Likewise, pre-incubation with neuraminidase significantly reduced hemagglutination caused by protein complexes containing multiple copies HAd37 fiber heads. Second, mutating single residues in the sialic acid binding sites reduced the hemagglutination activity of these protein complexes. Third, wild type HAd37 fiber heads bound more efficiently to erythrocytes or glycophorin than those carrying a point mutation in the sialic acid binding site. Our binding assays using wild type and mutated HAd37 heads also suggested that the sialic acid affinity is partially charge-driven. In addition, although HAd37 head also binds CAR [24], in the context of the virus it does not appear to use CAR as a receptor, possibly due to its relatively short fiber. Therefore, CAR on erythrocytes is likely to be less important for HAd37 biodistribution.
In spite of our structural data showing a well-defined sialic acid binding site, we could not detect notable binding between CAV-2 fiber head and glycophorin, a highly sialated protein. It was possible that the affinity of CAV-2 fiber head to erythrocytes or glycophorin was lower than that of HAd37 fiber head, and that the efficient erythrocyte binding of CAV-2 depended on the avidity of multiple fibers. As the CAV-2 fiber head is less positive than HAd37 fiber head (pI's of 8.4 and 9.2 respectively), this could have explained the lower affinity. The affinity between virus proteins and sialic acid is usually in the millimolar range; therefore it was conceivable that the affinity of single CAV-2 fiber heads to erythrocytes or glycophorin was low. Consistent with this assumption, we previously found that CAV-2 is more neutrally charged than other Ads [32]. Paradoxically though, CAV-2 agglutinates at a lower particle-to-cell ratio than HAd37 and our crystallography data suggested that sialic acid binding should be at least as strong as HAd37. However, one cannot reliably predict affinity from structural data.
The identification of CAR expression by erythrocytes from species that are agglutinated by CAR-tropic Ads is a crucial observation. Using i) competition assays with recombinant fiber heads harboring point mutations in the sialic acid and CAR binding sites and ii) transgenic mice expressing CAR on erythrocytes, we characterized the unexpected and significant role of CAR in Ad binding.
As mentioned previously, Ad serotypes differ in their ability to bind erythrocytes from various species. Our study suggests that this may be due to i) an interaction with different chemical variants, linkages and ratios of sialic acid, ii) the presence and affinity to CAR, and/or iii) the pI's of the head [33]. For example, some non-CAR-tropic Ads that agglutinate human erythrocytes may preferentially bind Neu5AC, the most abundant sialic acid on human erythrocytes. With respect to charge, both HAd37 and HAd19p bind sialic acid with equal affinity and the only two residues that differ between their heads are not close to the sialic acid-binding site [33]. With the limited amount of data available, subgroup D HAds tend to have heads with higher pI's and interact with sialic acid [33]. Interestingly, the residues that make up the CAV-2 sialic acid binding site are conserved in the CAV-1 fiber head (data not shown), suggesting a conserved sialic acid binding function. With respect to CAR affinity, the CAV-2 fiber head binds human CAR with the highest affinity of any fiber head known [24], which may favor its efficient hemagglutination. In contrast to other Ads, higher temperatures poorly inhibit CAV-2 hemagglutination [12], again suggesting a role for the high-affinity attachment to CAR. This is also consistent with a rather large difference between HAd5 and CAV-2 hemagglutination: to the best of our knowledge the HAd5 head does not harbor a sialic acid binding site and its pI (6.25) is less basic than HAd37 or CAV-2. Finally, we cannot exclude a role of the inter- and intra-species differences in the quantity of CAR expressed by erythrocytes.
Phylogenetically, CAR expression by erythrocytes appears to be random; we tested several other species (dogs, mice, rabbits, lemurs and monkeys) and did not find CAR expression. However, CAR levels on rat erythrocytes were relatively high (Figure 6) and consistent with our hypothesis that CAR-tropic Ads agglutinates human and rat erythrocytes via CAR binding. It is possible, but unlikely, that our lack of detection of CAR on erythrocytes on some species was due to the limit of sensitivity of our bank of anti-CAR antibodies, which nonetheless detected CAR expression on other cell types from these species (not shown). In the context of the significant amount of HAd5-mediated gene transfer data, our data highlights an important parallel. While a 30-gram mouse can be injected intravenously with up to 1012 pp of an HAd5 vector with minimal side effects, injection of the same dose in a 250-gram rat is normally lethal [41]. Whether our data have a bearing on the death following portal vein injection of a HAd5 vector during a phase I trial [42] is unknown, but deserves consideration.
Similar to the CAR binding domain in HAd37, we can only speculate about the importance of sialic acid binding in the biology of CAV-2. Although CAV-2 does not agglutinate dog erythrocytes (Figure S4), Canis lupus familiaris may not be the original host of CAV-2: seroprevalence against CAV-2 can be found in coyotes, bears, pandas, skunks, mongooses, raccoons and foxes [43]–[45]. It is paradoxical that HAd37 binds CD46, sialic acid and CAR - yet does not use CAR [25] - while CAV-2 binds both sialic acid and CAR and to the best of our knowledge uses only CAR as a receptor [5]. We cannot exclude the possibility that in some cell types sialic acid binding provides a first low-affinity attachment to the cell surface, while CAR-binding is followed in a second step, providing a high-affinity binding. Similar two-step mechanisms have been proposed for other viruses [46]. It is tempting to speculate that sialic acid binding may play a role in the preferential transduction of neurons by CAV-2 vectors [47]–[50]. Our crystal structure of CAV-2 fiber head in complex with CAR and sialyl-D-lactose demonstrates that the ternary complex of the three molecules is stable. There is no indication that this should not be the case also in vivo.
Our data and numerous reports describing the ex vivo interaction of CAR-binding HAd5 with human/rat erythrocytes [12], [21]–[23], [29], [36], [51]–[53] strongly suggest this interaction probably occurs after the intravascular injection of CAR-tropic vectors. Notably, Lyons et al. showed that that >90% HAd5 vector DNA was associated with blood cells following intratumoral injection during a clinical trial [34].
It is likely that erythrocyte binding occurs during wild type infection of some HAds (as well as coxsackie B viruses) that use CAR, which will certainly lead to altered biodistribution. HAd DNA is routinely found in human blood cells by PCR. It is also a common misconception that Ads are rapidly cleared following a classical immune response. Numerous clinical cases strongly suggest that latent HAds can readily resurface if the host is immunosuppressed. The fate of particles that stick to erythrocytes under natural or artificial (i.e. vector injections) conditions is probably complex. For example, HAd5-induced liver disease in immunocompromised humans is relatively common. In contrast, immunocompromised nonhuman primates are rarely diagnosed with simian Ad (SAV)-induced liver disease [54]. Does CAR expression on erythrocytes lead to an advantage for host or virus? Has CAR expression on erythrocytes put a selective pressure on Ads (and coxsackie B viruses that also bind CAR) to avoid an erythrocyte virus trap [28]? Or has CAR expression by erythrocytes allowed HAds to thrive because it has allowed open access to so many more tissues and cell types?
In summary, our results resolve a longstanding enigma of Ad-erythrocyte interaction in vitro and in vivo. In addition, we provide new insights into virus-erythrocyte interactions that will allow us to better understand HAd pathogenesis and facilitate the engineering of safer, more efficient gene transfer vectors. Although in most in vivo scenarios hemagglutination per se is unlikely to occur due to the turbulence that erythrocytes encounter in the circulation, sequestering of vector particles by erythrocytes must diminish gene transfer efficacy. Interest in Ad biology is continually growing due to the increasing incidence of HAd-induced morbidity and mortality during immunosuppression [55],[56], and because Ad-derived vectors are the most commonly used vectors in gene therapy clinical trials. The identification of fiber head mutants that do not bind human erythrocytes may be of interest. Equally important may be the need to screen fiber heads from other serotypes for sialic acid and CAR binding domains. Combined with HAd interactions with vitamin K-dependent coagulation factors [39], our study adds another critical element in the interaction with blood components, biodistribution and pathogenesis.
CAV-2 (CAVGFP), HAd5 (AdGFP), HAd37, and the hybrid HAd5-CAV-2H (Ad5Luc1-CK) and HAd5-HAd37F (Ad37f) vectors were prepared as previously described [30],[31],[57],[58]. Briefly, CAVGFP and AdGFP are E1-deleted vectors expressing GFP. The capsids contain no modifications. The hybrid HAd5-CAV-2H vector contains the fiber head of CAV-2 on the HAd5 capsid. HAd5-HAd37F vector contains the HAd37 fiber (shaft and head) on the HAd5 capsid. All vectors were purified by double banding on CsCl gradients, CsCl was removed using PD-10 columns (Pharmacia). The vectors were stored in phosphate-buffered saline (PBS) containing 10% glycerol at −80°C. Stock titers were >1×1012 physical particles/ml with >1 infectious particle/5 physical particles. Anti-CAR antibodies tested in this study included E1.1 a monoclonal mouse (S. Hemmi, University of Zurich), CAR1605 polyclonal rabbit (J. Zabner), MoAbE(mh)1 monoclonal mouse (S. Carson, University of Nebraska), and AF2654 (anti-mouse) and AF3336 (anti-human) polyclonal goat (R & D Systems).
The recombinant fiber heads and dodecahedral sample were loaded between the mica-carbon interface as described [59]. The samples were stained using 2% sodium silico tungstate pH 7.5 and air-dried. Images were taken under low-dose conditions in an EX1200-II JEOL electron microscope working at 100 kV and with a nominal magnification of 40,000. The images were scanned on a Z/I Imaging scanner (Photoscan TD) with a pixel size of 14 mm (3.5 Å per pixel at the sample level).
Fiber head constructs were cloned into pPROEX HTb (Life Technologies) and expressed with a cleavable His-tag as described previously [24]. HAd37 fiber head constructs contain residues 177–365 and CAV-2 fiber head constructs contain residues 358–542. Point mutations were introduced using the QuikChange Site-Directed Mutagenesis Kit (Stratagene) and polymerase chain reaction (PCR). Protein purification was performed as described [24]. Briefly, cells were incubated in lysis buffer (20 mM Tris-HCl pH 7.5, 300 mM NaCl, 20 mM imidazole, (Boehringer Complete EDTA-free protease inhibitor cocktail), centrifuged, and fiber head bound to a Ni-NTA column (Qiagen). Eluted protein was either directly loaded onto a Superdex200 column for hemagglutination experiments with tagged fiber head, or incubated overnight with 1/100 His-tagged tobacco etch virus (TEV) protease at 10°C. Proteins were then dialyzed against lysis buffer and uncleaved protein and TEV protease bound to a Ni-NTA resin. Untagged fiber head was loaded onto a Superdex200 column using the same buffer as for tagged protein (20 mM Tris pH7.5 and 300 mM NaCl). CAV-2HCAR− was labeled using Alexa488 Microscale Protein Labeling Kit (Molecular Probes). CAV-2Hwt was dialyzed in PBS 0.1 M NaCO3 pH 9.3 and labeled using mono-reactive dyes (Cy3, Cy5 or Alex488, Amersham Bioscience) for 45 min at room temperature. The elution of labeled protein was performed with 2 ml of PBS using NAP5 column (GE Healthcare) pre-equilibrated with 10 ml PBS. The final dye/protein ratios (∼2.4 for each) were determined using NanoDrop ND-100 spectrophotometer.
Untagged CAV-2 fiber head was concentrated to 17 mg/ml in crystallization buffer (150 mM NaCl, 20 mM Tris pH 7.5, 8 mM sialyl-D-lactose) and crystallized in hanging drops containing 1 µl protein solution and 1 µl well solution (5% PEG 4000, 5% isopropanol, 0.1 M HEPES pH 7.4). Single crystals were transferred and frozen in a cryoprotectant solution (65% well solution, 25% glycerol, approximately 80 mM sialyl-D-lactose). The sialyl-D-lactose (Sigma) corresponds to α2-3 N-acetylneuraminosyl-D-lactose (according to the supplier).
The cloning, expression and crystallization of HAd37 and CAV-2 fiber head CAR D1 complexes (space groups I23 and I422) was performed as described previously [24]. After crystal growth, sialyl-D-lactose (Sigma) was added to the drop to a final concentration of 10 to 50 mM. Crystals were frozen the next day without adding cryoprotectant solution to crystals with CAV-2 fiber head complex, and with 20% glycerol in the crystallization condition for crystals with HAd37 fiber head complex.
Details of the data collection and refinement of the three structures (CAV-2+sialyl-lactose, CAV-2+CAR D1+sialyl-lactose and HAd37+CAR D1+sialyl-lactose) are given in Table S1. All crystallographic data were collected at the European Synchrotron Radiation Facility (ESRF) and processed with XDS [60]. All structures were solved by molecular replacement using PHASER [61] and refined with REFMAC [62]. COOT [63] was used for the visualization of all models and electron density maps, and for the superposition of different models. PROCHECK [64] and MolProbity [65] were used for the validation of the obtained models. Representations of the protein models and the electron density were made with PYMOL [66] and GRASP [67]. Binding interfaces were visualized with DIMPLOT [68]. The structure-based sequence alignment was made with SARF, modified manually in SEAVIEW [69] and visualized with ESPript [70]. The CAV-2+sialyl-lactose structure contains two fiber head trimers in the asymmetric unit. The CAV-2+CAR D1+sialyl-lactose contains four trimeric fiber head - CAR D1 complexes in the asymmetric unit. Due to the modest resolution of this structure, NCS restraints and TLS refinement was used. The HAd37+CAR D1+sialyl-lactose structure has one fiber head monomer with bound CAR D1 in the asymmetric unit.
Chimeric mini-fiber constructs were cloned by PCR in two steps. First, DNA fragments were produced using as template HAd3 genomic DNA (for fragment 1) or pPROEX HTb vectors coding for wild type or mutant HAd37 fiber head (for fragments 2 and 3). Fragment 1 coded for HAd3 tail and shaft motives 1+2 (primers were AAT AAT CCA TGG CCA AGC GAG CTC GG and GTT CCA TGC TAC CAA GGA TCC ATC AGT AG). Fragments 2 and 3 coded for wild type and mutant (K345E) HAd37 fiber head (primers used were CTA CTG ATG GAT CCT TGG TAG CAT GGA AC and AAT AAT GAA TTC TCA TTC TTG GGC AAT ATA GG). In a second PCR reaction, fragment 1 was annealed to fragment 2 or 3 using primers AAT AAT GAA TTC TCA TTC TTG GGC AAT ATA GG and AAT AAT CCA TGG CCA AGC GAG CTC GG. The resulting longer fragments coded for complete mini-fiber containing HAd3 tail+shaft and either wild type or mutant (K345E) HAd37 fiber head. These were cloned into pPROEX HTb, and expressed at 25°C in E. coli strain BL21 Star (DE3) (Life Technologies) together with an N-terminal cleavable 6×His-tag. Cells were re-suspended in lysis buffer and sonicated. The cell lysate was centrifuged for 30 min at 25000g and the supernatant loaded on a Ni-NTA column (Qiagen). Protein bound to the resin was washed with 20 mM Tris-HCl pH 7.5, 300 mM NaCl, 50 mM imidazole and eluted in 20 mM Tris-HCl pH 7.5, 150 mM NaCl, 500 mM imidazole. To remove the His-tag, mini-fibers were incubated overnight with 1/100 His-tagged TEV protease at 10°C. Imidazole was removed by dialysis against lysis buffer. Uncleaved protein and TEV protease were removed by binding to a Ni-NTA resin. HAd3 penton forms multimers (dodecahedra) consisting of twelve penton bases each [9],[71]. P. Fender (Institut de Biologie Structurale, Grenoble) generously supplied purified HAd3 dodecahedra. Mini-fibers and dodecahedra were mixed at an approximate molecular ratio of 1:10 and incubated at 4°C overnight. The mixture was loaded onto a Superdex200 column (Amersham) to remove excess fiber. The buffer contained 20 mM Tris pH7.5 and 300 mM NaCl. Dodecahedra in complex with mini-fiber were eluted with the void volume and were visualized with an electron microscope. All mini-fiber constructs carrying HAd3 tail+shaft and a CAV-2 fiber head were unstable, possibly because they were misfolded, or they did not bind to HAd3 dodecahedra.
Non-treated, mock-treated or neuraminidase-treated erythrocytes and non-tagged fiber heads were incubated for 20 min at 4°C. Cell samples serving as negative control were incubated with PBS instead of fiber head solution. The cells were pelleted at 800g, re-suspended in PBS containing 1:100 rabbit anti-HAd37 fiber head serum (gift from N. Arnberg, University of Umeå) or 1:100 purified rabbit antibodies against CAV-2 fiber head, and incubated for 20 min at 4°C. The cells were pelleted as before, re-suspended in PBS containing 1:100 anti-rabbit FITC labeled antibody (Sigma) and incubated in the dark at 4°C for 20 min. The cells were pelleted again, re-suspended in PBS, and injected into a FACSCalibur apparatus (BD Biosciences). The results were analyzed using CellQuest and FlowJo software (BD Biosciences).
Human, rat (Wistar), dog (beagle), vervet (Chlorocebus pygerythrus), cynomologous monkey, rhesus macaque, guinea pig, rabbit (New Zealand White), mouse (C57BL/6) or GATA1-CAR [28] mouse blood was collected in EDTA, heparin or Alsever's solution. Erythrocytes were purified using Ficol gradients, washed twice in PBS EDTA (5 min at 2000 rpm) and stored less than 48 hrs in PBS containing 5 mM EDTA. For the human and GATA1-CAR mouse erythrocytes two fractions of packed erythrocytes were re-suspended separately in neuraminidase buffer each. Neuraminidase α(2-3,-6,-8 and-9) from Arthrobacter ureafaciens (Sigma) or α(2-3,-6 and -8) (Ozyme) was activated as recommended, and added to one of the two erythrocytes fractions. Both mock- (without neuraminidase) and neuraminidase-containing fractions were incubated for 1 hr at 37°C. The cells were then washed 5 times with PBS to remove neuraminidase and buffer, and re-suspended in PBS containing 5 mM EDTA. Virus, dodecahedra ± chimeric mini-fibers and CAV-2 fiber head multimers were diluted with PBS in 10-fold (in the first column) and then 2-fold (horizontal) steps and added to 96-well plates with cone-shaped bottoms. An equal number of purified erythrocytes (either non-treated, neuraminidase-treated, or mock-treated) was added to each well and incubated at room temperature or 37°C for at least 3 hrs. The blocking protein (∼1 µg/well) was incubated with erythrocytes for 1 h at 4°C with slow rotation. All assays were repeated at least 3 times.
In a volume of 100 µl, CAVGFP or AdGFP (1 particle/cell) was incubated with 2.5×107 mock-treated or with neuraminidase-treated erythrocytes with PBS, or PBS supplemented with NaCl to bring its final concentration to 225 or 400 mM. The samples were incubated at room temperature for 15 min, and then centrifuged for 5 min at 5000 rpm in a microfuge. Aliquots of the supernatant were removed and incubated with 1×105 cells in 12-well plates. The cells were trypsinized and assayed for GFP expression by flow cytometry 24 hrs post-incubation. Data were analyzed using CellQuest. The assays were performed twice and in quadruplicate.
Control cells and tissues were lysed with 100 µl SDS buffer (106 cells) and benzonase for 1 h at 37°C. The liver, peripheral blood mononuclear cells, NIH 3T3 cells (mouse fibroblasts) and 293 cell (human embryonic kidney) extracts (40 mg/ml) were resuspended in 100 µl of SDS-sample buffer. Erythrocytes (∼200 µl) were lysed in ddH2O, the membranes were pellet for 10 min at 14,000 RPM in a microfuge and resuspended in 100 µl of SDS-sample buffer. SDS-PAGE was performed using a 5% acrylamide/bis-acrylamide stacking gel and a 12% acrylamide/bis-acrylamide running gel. Membranes were blocked with TBS-Tween, 10% milk at room temperature. The rabbit anti-CAR antibody CAR1605 was diluted 2000-fold for 1 h at RT in TBS-Tween 10% milk. The secondary anti-rabbit antibody was used at a dilution of 1/5000 for 30 min at room temperature in TBS-Tween 10% milk.
Four adult GATA1-CAR and four control C57BL/6 mice were injected with 1.2×1011 pp of AdGFP via the tail vein. Blood (∼100 µl) was taken by tail vein bleeds at 0.25, 6, 24 and 72 hr. The mice were sacrificed at 72 hr by lethal injection, and the organs were perfused with PBS via cardiac puncture. The liver, lung and spleen were recovered, divided into parts for qPCR or histology. The organs used for histology were fixed in 4% PFA for 24 hr then placed in 20% sucrose for 24 hr, and embedded in OCT matrix (CellPath, Powys, UK). Sections (10-µm-thick) were stained with 0.2 µg/ml bisBenzimide Hoechst (Sigma-Aldrich) and 1 ng/ml phalloidin-TRITC (Sigma-Aldrich) before being mounted. Images were acquired using a Zeiss microscope and processes using the MetaMorph (Molecular Devices, Wokingham, UK).
The experimental protocols involving animals were approved by the University of Massachusetts Medical School Institutional Animal Care and Use Committee.
Total DNA from blood and liver were extracted by using the High Pure DNA Isolation kit (Roche Diagnostics). qPCR was performed with a Light Cycler (Roche Diagnostics) using the Platinum Taq DNA polymerase (Invitrogen) and SYBR Green qPCR master mix [72]. The primer pairs used for GAPDH were: GAPDH forward, 5′ ACA GTC CAT GCC ATC ACT GCC 3′; GAPDH reverse, 5′ GCC TGC TTC ACC ACC TTC TTG 3′; and the EGFP: forward, 5′ CAG AAG AAC GGC ATC AAG GT 3′; eGFP reverse, 5′ CTG GGT GCT CAG GTA GTG G 3′. Data are expressed as a ratio of GAPDH to EGFP.
Data were analyzed using a one-way ANOVA and post-hoc comparisons were made using an unpaired Student's t-test.
|
10.1371/journal.ppat.1003252 | Two-Component Signal Transduction System CBO0787/CBO0786 Represses Transcription from Botulinum Neurotoxin Promoters in Clostridium botulinum ATCC 3502 | Blocking neurotransmission, botulinum neurotoxin is the most poisonous biological substance known to mankind. Despite its infamy as the scourge of the food industry, the neurotoxin is increasingly used as a pharmaceutical to treat an expanding range of muscle disorders. Whilst neurotoxin expression by the spore-forming bacterium Clostridium botulinum appears tightly regulated, to date only positive regulatory elements, such as the alternative sigma factor BotR, have been implicated in this control. The identification of negative regulators has proven to be elusive. Here, we show that the two-component signal transduction system CBO0787/CBO0786 negatively regulates botulinum neurotoxin expression. Single insertional inactivation of cbo0787 encoding a sensor histidine kinase, or of cbo0786 encoding a response regulator, resulted in significantly elevated neurotoxin gene expression levels and increased neurotoxin production. Recombinant CBO0786 regulator was shown to bind to the conserved −10 site of the core promoters of the ha and ntnh-botA operons, which encode the toxin structural and accessory proteins. Increasing concentration of CBO0786 inhibited BotR-directed transcription from the ha and ntnh-botA promoters, demonstrating direct transcriptional repression of the ha and ntnh-botA operons by CBO0786. Thus, we propose that CBO0786 represses neurotoxin gene expression by blocking BotR-directed transcription from the neurotoxin promoters. This is the first evidence of a negative regulator controlling botulinum neurotoxin production. Understanding the neurotoxin regulatory mechanisms is a major target of the food and pharmaceutical industries alike.
| Botulinum neurotoxin produced by the spore-forming bacterium Clostridium botulinum is the most poisonous biological substance known to mankind. By blocking neurotransmission, the neurotoxin causes a flaccid paralysis called botulism which may to lead to death upon respiratory muscle collapse. Despite its infamy as the scourge of the food industry, the neurotoxin is attracting increasing interest as a pharmaceutical to treat an expanding range of muscle disorders. Whilst neurotoxin production by C. botulinum appears tightly regulated, to date only positive regulatory elements, thus enhancing the neurotoxin production, have been implicated in this control. The identification of negative regulators, responsible for down-tuning the neurotoxin synthesis, has proven to be elusive, but would offer novel approaches both for the production of safe foods and for the development of therapeutic neurotoxins. Here, we report a two-component signal transduction system that negatively regulates botulinum neurotoxin production. Understanding the neurotoxin regulatory mechanisms is a major target of the food and pharmaceutical industries alike.
| Botulinum neurotoxins are the most poisonous biological substances known to mankind. The neurotoxins are metalloproteases which block neurotransmission in cholinergic nerves [1], [2] in humans and animals to cause botulism, a potentially lethal flaccid paralysis. Botulinum neurotoxins are produced by vegetative cultures of the anaerobic spore-forming bacterium Clostridium botulinum which is widespread in the environment. The neurotoxins can enter the victim's body through intoxication with food or drink, or they can be produced from spores germinating and growing into active cultures in vivo, most likely in the gut of small babies with poorly developed gut microflora or in deep wounds. Despite their infamy, botulinum neurotoxins attract increasing interest as a pharmaceutical to treat an expanding range of muscular and other disorders [3], [4], such as torticollis, focal dystonia, inappropriate contraction of gastrointestinal sphincters, eye movement disorders, hyperhidrosis, migraine [5], genitourinary disorders [6], and even cancer [7]. Indications in cosmetic surgery are well known.
Seven antigenically distinct toxin types (A to G), and several subtypes therein, have been described [8]–[12]. Type A1 neurotoxins are the best characterized, a consequence both of their frequent involvement in human botulism worldwide and of their greater potency and, therefore, suitability for therapeutics [13]. Botulinum toxins are produced as a complex containing the neurotoxin itself and one or more non-toxic auxiliary proteins that protect the neurotoxin from environmental stress and assist in absorption [14]. Type A1 toxins are complexed with the non-toxic non-hemagglutinating (NTNH) protein and three hemagglutinins (HA17, HA33 and HA70) [15]–[18]. A typical A1-type gene cluster is transcribed in two operons, namely the ntnh-botA and ha operons [19] (Figure 1). Both operons have consensus −10 and −35 core promoter sequences, which are recognized by the alternative sigma factor BotR, directing RNA polymerase (RNAP) to transcribe the two operons [20]. The gene encoding BotR is located between the two operons within the neurotoxin gene cluster.
Botulinum neurotoxin production is affected by the availability of certain nutrients [21]–[23] and is associated with transition from late-exponential to early-stationary phase cultures. A peak in the level of neurotoxin gene cluster expression in late-exponential to early-stationary phase cultures [19], [24] suggests that neurotoxin production is tightly regulated. To date only positive regulatory elements have been implicated in this control. These include the participation of BotR [25] and an Agr quorum sensing system [26]. The identification of negative regulators of botulinum neurotoxin production has until now proved to be elusive.
Two-component signal transduction systems (TCS) are conserved in bacteria and differentially specialized to control a range of cellular events in response to environmental stimuli. The histidine kinases sense cellular or environmental signals through the N-terminus of their sensor domains. This interaction leads to autophosphorylation at a histidine residue in their C-terminus and the subsequent activation of their cognate response regulator present in the cytosol by transmission of the phosphoryl group to an N-terminal aspartate residue of the response regulator and further to the C-terminal output domain. Response regulators possess DNA-binding activity, ultimately resulting in a specific response in the expression of their target genes. The individual roles of most TCSs in C. botulinum are not known, but their involvement in control of virulence in other pathogenic bacteria has been demonstrated [27]. Antisense mRNA inhibition of genes encoding three TCSs caused decreased neurotoxin production in C. botulinum type A strain Hall [28], suggesting these TCSs may play a role in positive control of neurotoxin synthesis. The model strain C. botulinum ATCC 3502 (Group I, type A) [29] encodes 29 putative TCSs and a set of orphan histidine kinases and response regulators [30]. One of the intact TCSs, CBO0787/CBO0786 (Figure 1), shares over 90% amino acid identity with other C. botulinum Group I strains and in many strains is located in the vicinity of the neurotoxin genes (3.6 to 24 kilobases [kb] up- or downstream of the toxin genes). Here we show that the TCS CBO0787/CBO0786 negatively regulates botulinum neurotoxin gene expression. Understanding the regulatory mechanisms that control the production of botulinum neurotoxin is a major target of the food and pharmaceutical industries.
We used quantitative reverse transcription PCR (qRT-PCR) to measure the relative expression of cbo0787 and cbo0786 during the growth of C. botulinum Group I type A strain ATCC 3502, which is the most widely used model strain for genetic studies in C. botulinum [26], [31], [32]. The relative transcription levels of cbo0787 and cbo0786 followed an identical pattern, suggesting that the two genes are co-transcribed (Figure 2). In relation to growth, cbo0787 and cbo0786 were expressed at a relatively constant level throughout the logarithmic growth phase, and were down-regulated at the transition into stationary phase (Figure 2).
To address the role of CBO0787/CBO0786, we constructed single, insertional inactivation mutations in cbo0787 or cbo0786 using the ClosTron tool [33]. Single insertion of the group II intron from pMTL007 into the desired sites in cbo0787 or cbo0786 (Figure 1) was confirmed by PCR (Figure 3A) and Southern blotting (Figure 3B). Consecutive cultures showed the mutants to be erythromycin resistant and stable. No significant difference between the growth of the TCS mutants and the ATCC 3502 wild-type strain (WT) were observed (Figure 2), and the log cell counts per ml of WT and the cbo0787 and cbo0786 mutant cultures at early stationary phase (10 hours) were 8.9, 9.0 and 8.8, respectively.
We used qRT-PCR to measure the relative expression of botA encoding botulinum neurotoxin type A and ha33 encoding one of the three haemagglutinins in WT and the two TCS mutants at mid-exponential, late exponential and early stationary growth phases. The two genes were selected to represent the two operons present in the neurotoxin gene cluster, and the three time points used have been shown to associate with induction and repression of the neurotoxin gene expression [19], [34], while later time points typically involve other cellular events, such as sporulation or lysis, and were thus not considered relevant. As expected, the WT levels of botA and ha33 expression peaked at late exponential or early stationary growth phases (Figure 4). Strikingly, the cbo0787 mutant had a maximum of 10.4-fold (P<0.01) and 8.0-fold (P<0.01) higher relative botA and ha33 expression levels, respectively, than the WT (Figure 4), the most prominent differences being observed at early stationary growth phase. This suggests that the histidine kinase CBO0787 is required to sense an as yet unidentified signal for the efficient ‘switch-off’ of the neurotoxin gene expression at transition into stationary growth phase.
The cbo0786 mutant also showed significantly increased botA and ha33 expression levels in relation to WT (Figure 4). A maximum of 2.4-fold (P<0.05) induction for botA at early stationary phase and 3.8-fold (P<0.01) induction for ha33 at late exponential growth phase were measured. These data suggest that CBO0786 negatively regulates neurotoxin gene expression, either directly or indirectly.
To confirm that our findings at the transcription level apply to protein level, we analyzed the relative amounts of botulinum neurotoxin in the WT and mutant culture supernatants collected at mid-exponential, late exponential and early stationary growth phases. Measurements were made using an enzyme-linked immunosorbent assay (ELISA). At mid-exponential growth phase, neurotoxin production was slightly but not significantly increased in the cbo0787 and cbo0786 mutant cultures compared to WT culture (Figure 5A). At late exponential and early stationary growth phases, neurotoxin production was significantly increased (2.1 to 3.7-fold higher OD405 readings, P<0.05) in both the cbo0787 and cbo0786 mutant cultures relative to WT culture (Figure 5A). To validate this result, we complemented the cbo0786 mutation by introducing pMTL::cbo0787/0786, containing the TCS genes and their putative native promoter, into the cbo0786 mutant and showed that neurotoxin levels in early-stationary phase cultures were restored to the vector-only control (WT-pMTL) level (P<0.05, Figure 5B). No difference was observed in growth between the cbo0786-pMTL control and the complemented strain (Figure 5C).
To test the hypothesis that the TCS response regulator CBO0786 is a transcriptional repressor of the neurotoxin gene cluster in ATCC 3502, we examined the binding of the recombinant CBO0786 protein to probes that encompassed the intergenic region between ha33 and botR containing the promoter of the ha operon [35] (Pha33 probe), or the intergenic region between botR and ntnh containing the promoter of the ntnh-botA operon [35] (Pntnh-botA probe), by electrophoretic mobility shift assay (EMSA). CBO0786 caused a shift in the mobility of both probes, although its binding affinity to the Pntnh-botA probe appeared somewhat lower than to the Pha33 probe (Figure 6). The specific nature of binding was further confirmed by disappearance of both protein-DNA complexes using competition with a 200-fold excess of unlabeled probe. Moreover, no electrophoretic shift was observed with the negative control probe (Figure 6). EMSA with recombinant CBO0786 phosphorylated by acetyl phosphate yielded a similar shift (Figure 7), indicating that phosphorylation is not essential for CBO0786 binding to DNA in vitro. These results suggest that CBO0786 recognizes and binds to the promoter regions of the ha and the ntnh-botA operons.
To understand the DNA-binding specificity of CBO0786, we identified its binding sites using DNase I footprinting and fluorescently end-labeled Pha33 and Pntnh-botA probes (Figure 8). With the Pha33 probe, the region protected by CBO0786 (−51 bp to −31 bp upstream of ha33 and −196 bp to −176 bp upstream of botR) was present in both strands (Figures 8A and 8B). Greater protection by CBO0786 was observed in the antisense strand containing the promoter of the ha operon [35], suggesting that the antisense sequence (TATGTTATATGTTATATGTAA, Figure 8C) is the major CBO0786 binding site. Interestingly, the core promoter −10 region (GTTATA) of the ha operon, recognized by the alternative sigma factor BotR [20], appeared as a direct repeat in the CBO0786 binding site, suggesting that CBO0786 represses transcription of the ha operon by binding to its core promoter.
In the Pntnh-botA probe, a site of CBO0786 protection similar to that observed with the Pha33 probe was evident (Figures 8D and 8E). Both strands contained the same protection site (GGCTATGTTATAT) (−120 bp to −108 bp upstream of ntnh, Figure 8F). Accordingly, the −10 region (GTTATA) of the ntnh-botA operon core promoter was presented in the binding region, but in one copy only. In accordance with the EMSA analysis, the binding affinity of CBO0786 to the ntnh-botA promoter appeared lower than to the promoter of the ha operon.
To demonstrate the direct effect of CBO0786 on transcription of the neurotoxin genes, we carried out in vitro run-off transcription assays using RNAP reconstituted with E. coli RNAP core enzyme and the purified sigma factor BotR (Figure 9). DNA fragments containing the promoter of the ha or ntnh-botA operons were cloned and used as transcription templates. As expected, transcripts produced from the promoters of ha or ntnh-botA operons were observed only in the presence of both RNAP core enzyme and BotR. Addition of increasing concentrations of recombinant CBO0786 caused gradual inhibition of both transcripts, suggesting CBO0786 directly represses the transcription from ha and ntnh-botA promoters. Transcription from the ha promoter was more efficiently repressed than that from the ntnh-botA promoter in the presence of 4 µM CBO0786 (Figure 9). Taken together, CBO0786 exhibits a higher binding affinity to the direct repeat of the −10 region (GTTATA) in the ha promoter than in the ntnh-botA promoter, and consequently shows a more effective inhibition of in vitro transcription from the ha promoter than from the ntnh-botA promoter. These results indicate that CBO0786 represses the transcription of neurotoxin gene cluster by binding to the consensus core promoter −10 region of ha and ntnh-botA operons.
To test whether CBO0786 represses the transcription of botR, we also performed in vitro run-off transcription assays with DNA template containing the botR promoter. No clear transcript from botR promoter was observed in the presence of RNAP core with BotR. Consistent with a previous study [20], the result demonstrates that botR is not auto-transcribed in vitro.
We show that the TCS CBO0787/CBO0786 negatively regulates botulinum neurotoxin expression in C. botulinum Group I type A1 strain ATCC 3502. This was supported by enhanced toxin gene expression and increased toxin synthesis by cbo0787 and cbo0786 mutants, by specific binding of recombinant CBO0786 regulator to the neurotoxin promoters, and recombinant CBO0786 inhibiting in vitro transcription from the neurotoxin promoters. Identification of botulinum neurotoxin repressors has been an unattainable target of neurotoxin research for decades [13], [36], [37], and would open up novel strategies for controlling the public health risks caused by the toxin, but also for enhancing industrial processes for therapeutic neurotoxin preparations.
While the current work focused on the most well-characterized C. botulinum type A1 neurotoxin subtype, the role of CBO0787/CBO0786 homologs in strains of other subtypes will be an interesting line of future research. While the cbo0787/cbo0786 locus is found at an 11-kb distance from the neurotoxin gene cluster (cbo0801–cbo0806) in ATCC 3502, a TCS showing a strikingly high (>95%) amino acid identity to CBO0787/CBO0786 is similarly encoded by loci near the neurotoxin gene cluster in the genomes of type A2 strain Kyoto (3.6 kb distance), type A5 strain H04402 065 (12 kb), and type F strain Langeland (24 kb). This TCS, therefore, is an interesting candidate for a universal neurotoxin repressor in Group I C. botulinum.
Our biochemical data suggest that CBO0786 recognizes and binds to the consensus −10 region (GTTATA) of the ha and ntnh-botA operon promoters. The −10 region is also specifically recognized by the alternative sigma factor BotR [20], [38], suggesting that CBO0786 binding to the core promoter −10 region may prevent RNAP-BotR from binding and initiating the transcription. The in vitro run-off transcription assay demonstrated that CBO0786 directly inhibits RNAP-BotR-directed transcription from the ha and ntnh-botA promoters, supporting the hypothesis that CBO0786 inhibits the transcription of these operons.
BotR is specifically required for botulinum neurotoxin gene expression [25] but does not autoregulate its own expression [20]. While the −35 site (TTTACA) of the ha and ntnh-botA core promoters is also found upstream of botR, the −10 site of botR promoter is different (TTCGTA) [20]. Accordingly, we did not observe CBO0786 binding to botR promoter. Moreover, we did not find additional CBO0786 binding sites downstream of the botR promoter in vitro or in silico. Thus it is plausible that CBO0786 does not repress botR transcription.
While an early study detected both monocistronic botA and bicistronic ntnh-botA mRNA species and a putative transcription start site in the intergenic region between ntnh and botA [34], BotR was later shown to exclusively drive the bicistronic form of transcription [20]. Accordingly, we could not identify a CBO0786 binding site in the intergenic region between ntnh and botA, thus inhibition by CBO0786 of transcription from a putative botA-specific promoter is not likely.
BotR homologues in other pathogenic Clostridia, such as TetR controlling tetanus neurotoxin in Clostridium tetani [39], TcdR controlling toxins A and B in Clostridium difficile [40], and UviA controlling bacteriocins in Clostridium perfringens [41] all recognize the same −35 sequence as BotR [20]. Interestingly, TetR also recognizes the same −10 box (GTTATA) as BotR [36]. The closest homolog of CBO0787/CBO786 in C. tetani is CTC01420/CTC01421, with the response regulator CTC01421 showing 75% amino acid similarity to CBO0786. Thus CTC01420/CTC01421 may be an interesting candidate for a tetanus neurotoxin repressor.
Our qRT-PCR and ELISA analysis, albeit yielding significantly greater neurotoxin gene expression in both cbo0787 and cbo0786 mutants than in WT, suggest the mutant phenotypes to be somewhat less striking than reported for some well-characterized TCSs in other Gram-positive bacteria. For example, mutation of CsrS, reported to respond to environmental Mg2+ [42] and to repress virulence-related capsular polysaccharide production in group A Streptococcus, resulted in a 10-fold increase in cellular hyaluronic acid production in the presence of Mg2+ [42]. However, in medium lacking Mg2+, only a modest 1.5-fold induction in hyaluronic acid production was observed [42]. Thus the presence of the signal triggering a TCS kinase is a key to control the activity of the cognate response regulator and its subsequent effects on target gene expression. The signal triggering CBO0787 and the role of phosphotransfer between CBO0787 and CBO0786 remain to be elucidated. Although our EMSA analysis suggested a DNA-binding activity for CBO0786 both with and without phosphorylation in vitro, which is in line with reports on some other TCS response regulators showing similar DNA-binding properties regardless of phosphorylation state [43], in vivo phosphorylation of CBO0786 could lead to conformational changes that fine-tune its DNA binding affinity [44]–[46]. Identification of the signal triggering CBO0787 will be an important future task for optimization of the study conditions to detect maximal mutant phenotypes and thus to detect the maximal effect of CBO0786 on neurotoxin repression.
The sensor domain of CBO0787 is predicted to contain an extracellular loop consisting of 79 amino acid residues and is flanked by two transmembrane helices, indicating that the signal triggering it is most likely extracellular. Moreover, the increasing differences observed between the relative neurotoxin gene expression levels of the cbo0787 kinase mutant and WT towards stationary growth phase are consistent with cell density-dependent signals [37]. Neurotoxin overproduction associated with inability to sporulate by strain Hall A-hyper [13] suggests that regulation of these processes may be linked. Repression of toxin synthesis after logarithmic growth before initiation of sporulation may represent a survival mechanism when nutrient sources become limited. Moreover, uncontrolled synthesis of botulinum neurotoxin would waste energy since only nanogram quantities are sufficient for C. botulinum to kill mammals and thereby gain nutrients and establish anaerobiosis.
Previous studies have identified excess of arginine [21] and tryptophan [22] to repress botulinum neurotoxin formation. By contrast, glucose was shown to induce neurotoxin synthesis [21]. Interestingly, glucose has also been linked with regulation of toxin synthesis in C. difficile through carbon catabolite control [47], [48]. Further research will be required to clarify the mechanisms by which nitrogen sources or glucose control neurotoxin synthesis in C. botulinum; however, our preliminary data do not indicate that these compounds trigger CBO0787 (data not shown). Thus it is plausible that the regulation of neurotoxin synthesis is accomplished through a complex network and other regulators are involved in this control.
While the relative neurotoxin gene expression levels in the cbo0786 regulator mutant were significantly higher than those of the WT at logarithmic growth phase, this difference was smaller at early stationary growth phase. Moreover, the wild-type levels of cbo0786 and cbo0787 transcription, being stable throughout the logarithmic growth phase, collapsed at the transition into stationary phase. These observations further support the involvement of repressors other than CBO0786 in the ‘switching-off’ of the neurotoxin expression, ensuring efficient onset of stationary-phase cellular events.
Bearing in mind the emergence of repressor gene (tcdC) mutations in ‘hyper-virulent’ isolates of the notorious healthcare-associated pathogen C. difficile [49], [50], the emergence of C. botulinum strains with a mutated neurotoxin repressor would set challenges to public health and safety. Hence identification of neurotoxin regulators and their mutations is crucial.
In conclusion, we propose that the TCS CBO0787/CBO0786 negatively regulates botulinum neurotoxin gene transcription through the response regulator CBO0786 blocking the −10 core promoter sites of the ha and ntnh-botA operons, inhibiting transcription from the ha and ntnh-botA promoters. These data provide keys for controlling the production of botulinum neurotoxin, which is a major target of the food and pharmaceutical industries.
Bacterial strains and plasmids are described in Table S1. C. botulinum Group I type A1 strain ATCC 3502 [29] and derivative mutants were grown in anaerobic tryptone-peptone-glucose-yeast extract (TPGY) medium at 37°C under strictly anaerobic conditions. Cell counts were determined by plating serially diluted cultures on anaerobic TPGY agar plates. Escherichia coli conjugation donor CA434 [51] and E. coli TOP10 strain (Invitrogen) were grown in Luria-Bertani (LB) medium at 37°C. When appropriate, growth media were supplemented with 100 µg/ml ampicillin, 50 µg/ml kanamycin, 25 µg/ml chloramphenicol, 250 µg/ml cycloserine, 15 µg/ml thiamphenicol or 2.5 µg/ml erythromycin. All oligonucleotide primers are listed in Table S2.
Target genes were insertionally inactivated in C. botulinum ATCC3502 by using the ClosTron system as previously reported [33], in combination with the TargeTron gene knockout system kit (Sigma-Aldrich). Target sites in cbo0786 (between nucleotides 267–268) and cbo0787 (between nucleotides 603–604) were identified, and intron-retargeting PCR primers (Table S2) were designed by using the TargeTron algorithm (http://www.sigma-genosys.com/targetron/).
Plasmid retargeting was carried out as previously described [33] and the resulting plasmid pMTL007::cbo0786 or pMTL007::cbo0787 was transferred to C. botulinum ATCC3502 by conjugation from E. coli CA434 [51]. Successful transconjugants were screened on TPGY agar plates containing cycloserine (250 µg/ml) and thiamphenicol (15 µg/ml), and then resuspended in 1 ml of anaerobic TPGY containing 1 mM isopropyl-β-D-thiogalactopyranoside (IPTG) and thiamphenicol (7.5 µg/ml) and incubated at 37°C for 3 h. The bacteria were then harvested, resuspended in 1 ml of fresh TPGY and incubated for a further 3 h. The subsequent integrants were selected by plating bacteria on TPGY agar supplemented with erythromycin (2.5 µg/ml) and cycloserine (250 µg/ml) and incubated for 16 h at 37°C in anaerobic conditions to select clones harboring the spliced erythromycin retrotransposition activated marker (ErmRAM), which indicates intron integration.
To demonstrate the integration of the Ll.LtrB intron in the desired sites, PCR was performed using primers flanking the target sites in cbo0786 and cbo0787 (Table S2). PCR using ErmRAM primers demonstrated the spliced ErmRAM. In addition, to confirm that only a single intron insertion occurred in each mutant, genomic DNA from the ATCC3502 wild-type strain and mutants, and the pMTL007 plasmid DNA were digested overnight with HindIII and analysed by Southern blot probed with a DIG-labeled fragment derived from the Ll.LtrB intron sequence.
For complementation, a 2441-bp fragment encompassing cbo0786, cbo0787, and the 5′ noncoding region including their putative promoter, was cloned into plasmid pMTL82151 [52] to make pMTL::cbo0787/0786. pMTL::cbo0787/0786 or pMTL82151 (empty-vector control) was transferred to C. botulinum ATCC3502 or cbo0786 mutant by conjugation from E. coli CA434. The complementation strain of cbo0786 mutant-pMTL::cbo0787/0786 and control strains of C. botulinum ATCC3502-pMTL and cbo0786 mutant-pMTL were obtained.
Total RNA from C. botulinum ATCC 3502 and the two mutants was isolated using the RNeasy Mini Kit (Qiagen) as described [31]. Residual DNA was removed sequentially with RNase-free DNase set (Qiagen) and the DNA-free Kit (Ambion) according to the manufacturers' instructions. The RNA was dissolved in 50 µl of nuclease-free water (Sigma-Aldrich) and its concentration was determined using the NanoDrop ND1000 spectrophotometer (NanoDrop Technologies). The integrity of RNA was confirmed with the Agilent Technologies 2100 Bioanalyzer.
For qRT-PCR, samples were collected during mid-exponential, late exponential and early stationary growth phases (Figure 2). Duplicate cDNA samples were generated from 800 ng of RNA using the DyNAmo cDNA Synthesis Kit (Finnzymes). Quantitative real-time PCR reactions comprised of 1× DyNAmo Flash SYBR Green I Master Mix (Finnzymes), 0.5 mM of each primer (Table S2), and 4 µl of 102-fold (cbo0786, cbo0787, botA, ha33) or 105-fold (16S rrn) diluted cDNA template in a total volume of 20 ul. All PCRs were performed in duplicate for both cDNA replicates and three replicated experiments. Real-time PCR was performed using the Rotor-Gene 3000 real-time thermal cycler (Corbett Life Science). Cycling conditions included 7 minutes at 95°C, followed by 45 cycles of 95°C for 10 seconds and 60°C for 20 seconds, followed by 30 seconds at 60°C. PCR efficiencies were determined based on a standard curve made from serially diluted pooled cDNA for each primer pair. The calculated efficiencies were 0.97 for 16S rrn, 0.99 for botA, 0.92 for ha33, 1.04 for cbo0787 and 0.93 for cbo0786. Melting curve analysis was performed following the completion of the PCR to confirm specificity of the PCR amplification products. Target gene expression was normalized to the expression of 16S rrn based on the Pfaffl method [53]. All samples were calibrated against the wild type culture at mid-exponential growth phase.
Botulinum neurotoxin was quantified by using a commercial type A neurotoxin ELISA kit (Tetracore) in three independent WT and mutant culture supernatants collected at mid-exponential, late exponential, and early stationary growth phases [24], [26], [28]. The plates were read at 405 nm (Multiskan Ascent, Thermo Fisher). To keep the optical density readings in the dynamic range, the culture supernatants were diluted 1∶10 at mid-exponential, 1∶20 at late exponential, and 1∶30 at early stationary growth phase.
To construct the plasmids for the expression of N-terminal 6-histidine translation fusion to the response regulator CBO0786 or the alternative sigma factor BotR, PCR products were generated using the primers listed in Table S2. The PCR products of cbo0786 and botR were digested with appropriate restriction enzymes and cloned individually into plasmid pET28b (Novagen). The plasmids were then individually transformed into E. coli Rosetta 2(DE3) pLysS cells (Novagen).
CBO0786 expression was induced with 1 mM IPTG at 37°C for 5 h. Cells from a 500-ml culture were harvested, re-suspended in 10 ml of lysis/binding buffer (500 mM NaCl, 20 mM imidazole, 20 mM Tris-HCl, pH 7.9) and lysed by sonication. The lysate was centrifuged at 10 000 g for 15 min and filtered through a 0.45-µm filter. The lysate was loaded with 1 ml of Novagen His Bind affinity resin and allowed to bind for 30 min at 4°C. The resin was washed by 10 ml of lysis/binding buffer and 20 ml of wash buffer (500 mM NaCl, 60 mM imidazole, 20 mM Tris-HCl, pH 7.9). Bound protein was eluted by washing with 4 ml of elution buffer (500 mM NaCl, 500 mM imidazole, 20 mM Tris-HCl, pH 7.9).
BotR expression was induced with 1 mM IPTG overnight at 20°C and purified as described previously [20], with some modifications. Briefly, cells were lysed by sonication in lysis/binding buffer. The insoluble cell debris was separated by centrifugation and dissolved in denaturing lysis/binding buffer with 6 M guanidine hydrochloride. After centrifugation and filtration, the denatured solubilized proteins were collected and loaded onto column containing Ni-NTA affinity resin. The bound proteins were let to refold on column with a decreasing urea gradient (6 to 0 M) in lysis/binding buffer. Finally, the refolded proteins were obtained by washing with elution buffer.
Eluted fractions were examined by SDS-PAGE and fractions containing CBO0786 or BotR were pooled in the Novagen D-tube Dialyzer and dialysed against 1 l of dialysis buffer (300 mM NaCl, 20% glycerol, 50 mM Tris-HCl, pH 8.0) overnight at 4°C. Protein concentrations were determined by using the Bradford reagent (Bio-Rad) and BSA (Sigma-Aldrich) was used as a standard.
A 354-bp fragment (Pha33 probe) covering the intergenic region between ha33 and botR (−287 bp to 67 bp of ha33) and a 262-bp fragment (Pntnh-botA probe) covering the intergenic region between botR and ntnh (−210 bp to 52 bp of ntnh) were amplified by PCR using 5′-end biotin labeled primers (Table S2). CBO0786 phosphorylation by acetyl phosphate was performed as described [54]. EMSA was performed with 1 nM of 5′-end biotin labeled, double-stranded oligonucleotide probes, 0 to 5.4 µM of recombinant CBO0786, 1 µg of poly(dI-dC), 2.5% glycerol and 5 mM MgCl2 in binding buffer (LightShift Chemiluminescent EMSA Kit, Pierce). For competition assays, a 200-fold molar excess of unlabeled double-stranded oligonucleotide was added. Binding was allowed to proceed for 30 min at room temperature. Band shifts were resolved on a 5% native polyacrylamide gel run in 0.5× TBE at 4°C for 1 h at 110 V.
DNase I footprinting was performed in triplicate using a modification of [55]. The Pha33 and Pntnh-botA probes were amplified by PCR using 6-FAM-labeled forward primers and HEX-labeled reverse primers (Table S2). Binding reactions were performed as described for EMSA, with 20 nM of 5′-6-FAM-labeled probe and 10 µM protein in a final volume of 20 µl. After 20 min of incubation, DNA probes were partially digested by 0.002 to 0.2 Kunitz unit of DNase I (Sigma-Aldrich) for 5 min at room temperature. Reactions were stopped by addition of 22 µl of 0.5 M EDTA and heated at 70°C for 10 min. The digested DNA fragments were purified with the QIAquick PCR Purification Kit (Qiagen) and eluted in 25 µl of water. The purified fragments were separated in a capillary DNA analyzer (Applied Biosystems 3130×l DNA Analyzer) and the electropherograms were analyzed using the Peak Scanner software (Applied Biosystems). Protected regions were identified by sequencing of the fragments using the Thermo Sequenase Cycle Sequencing Kit (Affymetrix) and the same labeled primers as described above (Table S2).
The upstream region of the ha operon spanning the −191 to +185 sites relative to the transcription start, and the upstream region of the ntnh-botA operon spanning the −108 to +172 sites relative to the transcription start, were cloned into pBluescript II KS- (Stratagene) and then linearized by SpeI or PstI to produce the run-off transcription templates. In vitro transcription assays were carried out in 10-µl reaction mixtures, in the absence or presence of CBO0786. E. coli RNAP core enzyme (Epicentre) was preincubated for 30 min at 37°C with six-fold molar excess of purified BotR. After incubation, 0.5 U RNAP, 0 to 4 µM of recombinant CBO0786, and 15 nM of linearized plasmid DNA were added in transcription buffer containing 40 mM Tris-HCl, pH 8.0, 10 mM MgCl2, 50 mM KCl, 0.1 mg/ml BSA, 5% glycerol, 4 U of RNasin (Promega), and incubated for 10 min at 37°C. Transcription was initiated by the addition of 200 µM each of ATP, GTP and CTP, 50 µM of non-radioactive UTP and 2.5 µCi [α-32P]-UTP (3000 Cimmol−1, Perkin-Elmer). After further incubation for 30 min at 37°C, the reaction was quenched by adding 10 µl of RNA loading buffer (95% formamide, 0.025% bromophenol blue, 0.025% xylene cyanol FF, and 5 mM EDTA), followed by heating for 10 min at 80°C. Samples were resolved by denaturing 6% PAGE and visualized by autoradiography.
For any pairwise comparisons between the WT and one of the two mutant strains, Student's t-test was used. For a multiple comparison among the cbo0786-pMTL::cbo0787/0786, cbo0786-pMTL, and WT-pMTL, one-way ANOVA with Tukey's post hoc test was used.
cbo0787 (ID number 5185042), cbo0786 (ID number 5185041), cbo0803, ha33 (ID number 5185058), cbo0805, ntnh (ID number 5185060), cbo0806, botA (ID number 5185061), Clostridium botulinum type A strain ATCC 3502 genome (Accession number NC_009495.1).
|
10.1371/journal.pntd.0004602 | Steps Toward Creating A Therapeutic Community for Inpatients Suffering from Chronic Ulcers: Lessons from Allada Buruli Ulcer Treatment Hospital in Benin | Reducing social distance between hospital staff and patients and establishing clear lines of communication is a major challenge when providing in-patient care for people afflicted by Buruli ulcer (BU) and chronic ulcers. Research on hospitals as therapeutic communities is virtually non-existent in Africa and is currently being called for by medical anthropologists working in the field of health service and policy planning. This paper describes a pioneering attempt to establish a therapeutic community for patients suffering from BU and other chronic ulcers requiring long term hospital care in Benin.
A six-month pilot project was undertaken with the objectives of establishing a therapeutic community and evaluating its impact on practitioner and patient relations. The project was designed and implemented by a team of social scientists working in concert with the current and previous director of a hospital serving patients suffering from advanced stage BU and other chronic ulcers. Qualitative research initially investigated patients’ understanding of their illness and its treatment, identified questions patients had about their hospitalization, and ascertained their level of social support. Newly designed question–answer health education sessions were developed. Following these hospital wide education sessions, open forums were held each week to provide an opportunity for patients and hospital staff to express concerns and render sources of discontent transparent. Patient group representatives then met with hospital staff to problem solve issues in a non-confrontational manner. Psychosocial support for individual patients was provided in a second intervention which took the form of drop-in counseling sessions with social scientists trained to serve as therapy facilitators and culture brokers.
Interviews with patients revealed that most patients had very little information about the identity of their illness and the duration of their treatment. This knowledge gap surprised clinic staff members, who assumed someone had provided this information. Individual counseling and weekly education sessions corrected this information gap and reduced patient concerns about their treatment and the status of their healing process. This led to positive changes in staff–patient interactions. There was widespread consensus among both patients and staff that the quality of communication had increased significantly. Open forums providing an opportunity for patients and staff to air grievances were likewise popular and patient representative meetings resulted in productive problem solving supported by the hospital administration. Some systemic problems, however, remained persistent challenges. Patients with ulcers unrelated to BU questioned why BU patients were receiving preferential treatment, given special medicines, and charged less for their care. The idea of subsidized treatment for one disease and not another was hard to justify, especially given that BU is not contagious.
This pilot project illustrates the basic principles necessary for transforming long term residential hospitals into therapeutic communities. Although the focus of this case study was patients suffering from chronic ulcers, the model presented is relevant for other types of patients with cultural adaptation.
| Little is known about communication patterns and social relations between health staff and long -term patients in African hospitals. An ethnography of a reference hospital treating patients afflicted with Buruli Ulcer (BU) and other chronic ulcers in Benin was conducted. Sources of psychosocial distress and communication patterns compromising quality of care were documented. Based on this research, an intervention was mounted to transform the hospital into a higher functioning therapeutic community. Question: answer education sessions were introduced to provide patients the opportunity to inquire about their illness, it’s treatment and trajectory; weekly open- forums were established to give patients and hospital staff a chance to air grievances; patient representatives met with hospital staff to resolve problems in a non-confrontational manner, and psychosocial support for individual patients was provided through drop-in counseling sessions with social scientists in residence. Patients reported positive changes in the quality of their care and interactions with care providers, care providers reported that the problem solving process instituted was productive, and hospital administrators actively supported efforts to improve social relations and lines of communication. Systemic problems related to perceptions of preferential treatment for BU patients provided subsidized treatment supported by a national program remained contentious.
| Providing in-patient care for people afflicted with diseases requiring long-term hospitalization is a major challenge in low-income countries. In these countries, health staff must manage patients with limited resources. At the same time, patients struggle to maintain a positive attitude while far from their families and burdened by concerns about both the progress of their treatment and the welfare of their households during their absence. Patients and hospital staff live and work in close quarters, yet they are often socially distant, their interactions cordial yet primarily focused on disease management tasks. While considerable literature exists in developed countries on the hospital as a social system and on formation of therapeutic communities to care for long-term patients (primarily mental health and substance abuse patients) [1,2], hospital based research on other types of therapeutic communities is sparse, and virtually non-existent for Africa.
This paper describes a pioneering attempt to establish a therapeutic community for patients suffering from Buruli ulcer (BU) and other chronic ulcers requiring long-term care in Benin, West Africa. The hallmark of a hospital-based therapeutic community, as we define it in this paper, is a communication process that invites open dialogue between patients and health staff, patient participation in problem-solving associated with everyday living, ways and means of resolving conflicts that arise, and information exchange that fosters adherence, as distinct from one-sided directives demanding compliance. Our definition of therapeutic community is based on the principle of mutual respect and recognition that respect is only forthcoming when patients and staff better understand the works, responsibilities, challenges, and constraints each faces.
Buruli ulcer (BU) is the third most common mycobacterial disease in the world. A majority of cases are found in West Africa [3]. It is a neglected tropical disease affecting poor rural villagers in several West African countries. Cases diagnosed early can be cured with 56 doses of a combined regimen of intramuscular streptomycin and oral rifampicin. Treatment of advanced cases of BU often requires surgery and long-term residential treatment. During their stay in hospital, a patient’s dressings must be changed daily or at least three times a week, and the patient must undergo physical therapy to prevent disabilities and joint contractures [4,5].
The Allada Buruli Ulcer Treatment Center (CDTUB) is one of the four primary reference centers for BU care in Benin and a recognized center of excellence for clinician training in BU management. The hospital also treats patients suffering from other types of chronic ulcers of various etiologies such as sickle-cell disease, necrotizing fasciitis, and phagedenic or vascular ulcers. Since BU treatment is the primary vocation of the center, BU patients receive subsidized treatment thanks to the government and international NGOs. Patients with advanced BU residing at the hospital require extensive post-operative care. Other chronic ulcer patients have to pay for much of their therapy out of pocket.
When patients suffering from more advanced stages of BU and other chronic ulcers come to hospitals like Allada, they have to adapt to a new way of life in unfamiliar surroundings. They have to learn to get along with other patients who are members of groups they have had little contact with in the past. They then have to cope with the uncertainty of their illness trajectory, the demands of treatment, and the physical discomfort associated with the frequent changing of bandages and physical therapy sessions. For more advanced cases requiring skin grafts, the duration of treatment is uncertain and difficult to predict due to individual variability in wound healing.
Given that the duration of BU treatment is long, and patients are unable to care for themselves, family caretakers are asked to accompany patients and attend to their daily needs such as cooking, washing clothes, and daily assistance. One of the main conditions for being admitted to the hospital is identifying a suitable caretaker from one’s extended kin network. This is often difficult, as removing household members responsible for agricultural operations or child care at home can place the wellbeing of an entire household in peril [6]. In some cases, caretakers come and go, and in other cases they are not able to remain at the hospital and the patient is abandoned [7]. Food is partly provided free of charge for BU patients, but not for their caretakers, and not for patients suffering from other types of chronic ulcers. Although treatment is subsidized for BU patients, there are indirect costs related to hospitalization that can prove burdensome.
The CDTUB is located in Allada, a small city of 127, 493 inhabitants located in Benin (West Africa) [8]. It is staffed by four doctors, 18 nurses, eight laboratory technicians, six support staff, three maintenance workers and three drivers. The director of the hospital is a doctor actively engaged in the care of BU patients as well as BU-related research. He is assisted by an administrative staff composed of five secretaries and accountants. The hospital receives approximately 200 new patients a year, out of which around 40 are BU cases. BU patients typically remain in the hospital for 8–18 months, but some remain much longer. Patients in the hospital range from 2 to 70 years of age, with 60% being children. There is an even split between male and female residents in the hospital, with residents divided into nine wards segregated by gender. Caretakers range in age from 9 to 50 years of age, and an overwhelming majority (over 90%) are female.
At the CDTUB, all patients are required to obey rules put in place by the hospital administration to assure a sense of order as well as quality of care. Compliance with hospital policies is mandatory. At the time of the pilot project, patients were treated as passive recipients of care and not provided much knowledge about their disease beyond being told what medications, if any, they were required to take and how to assist in the cleaning and bandaging of their wounds. For patients, their stay at the hospital was a highly liminal experience marked with much apprehension and uncertainty.
Ethical approval was obtained from Benin’s National Ethical Committee of Health Research before the start of the research (IRB00006860 N° 148 /MS/DC/SGM/DFRS/CNPERS/SA). Informed consent procedures already in place at Allada hospital were strictly adhered to over the course of the project. All patients and staff interviewed were assured that their opinions would be kept confidential. Patients were assured that information volunteered would in no way affect the quality of their care at the hospital. Patients were also reassured that issues discussed at open forum meetings would not result in negative actions by the staff, and a grievance process was put into place to make sure this did not occur. Oral consent was documented by the presence of witness. The use of oral consent is approved by the ethical review board because many study participants were illiterate. When a participant was under 18 years of age, both the child/adolescent and his/her caretaker were informed about the nature and aim of study before being asked to give consent.
Three core challenges to establishing a therapeutic community were identified during the pilot project. The first challenge is how to establish an open forum where patients and staff feel comfortable enough to speak their minds without fear of reprisal. If staff feels they are being criticized and that this will have negative impact on their job performance, they will assume a defensive posture. This challenge requires the active support of the hospital director and hospital administration. In the present case, the hospital director let it be known that he viewed the airing of discontent as the first step of a problem solving process that was valued at the hospital. Establishing trust in this process took time and required change on the part of all members of the therapeutic community. By the end of the seven -month pilot project, all stakeholders interviewed had enough trust in the process to feel they could communicate their problems without compromising their position or the quality of care they received.
The second challenge faces social scientists attempting to establish a therapy facilitator/cultural broker role. It is important that they not be seen as the handmaiden of the hospital administration or an advocate for either health staff or patients. Trust demands a neutral position where the charge of the social scientist is to identify, investigate and present all sides of a dispute and to provide in depth understanding of issues affecting administration–staff–patient relations. During the project, there were times when various parties attempted to gain the support of a social scientist in opposition to another. It became important for the social scientist to be clear about what they can and cannot do as part of a process of problem solving. For example, when a patient became destitute because they lost a caretaker or the resources needed for treatment, the social scientists assisted the patient in presenting a case to the administration, but could not be seen as directly solving the resource problem themselves. During the community outreach program that preceded the therapeutic community intervention, the social science team created a resource assessment screening tool to facilitate patient referral to the hospital. The same assessment tool was used in the hospital when an economic crisis was revealed to a social scientist. The screener enabled the case to be systematically presented to the administration after all data necessary to make a decision had been acquired.
A third challenge is sustainability and cost-effectiveness of the social scientist role. The therapeutic community model presented in the study requires the presence of a social scientist and justification for the resources needed to support the position. Based on the results of the pilot study, the Allada hospital administration has decided to employ a social scientist to assist in therapy facilitation and community- based outreach activities, and to secure the services of a psychologist in cases where patients need to be treated for mental health problems requiring medication.
In this paper, we have described a pioneering attempt to transform an African hospital serving long-term residential patients into a therapeutic community. Although the focus of this case study is BU patients, the model and experience presented here are relevant for many other types of patients. It requires a rethinking of hospital staff–patient relations in concert with the tenets of patient-centered and humanized patient care [19–22] and people-centered health policy [23, 24]. For patients, it addresses their concerns, enhances their sense of well-being, and provides a sense of support and compassion during their long hospital experience. For staff, it leads to greater patient adherence and the resolution of conflicts that can compromise care. In addition, it provides staff as well as patients a forum to articulate their grievances. And for administrators, it provides them with a finger on the pulse of everyday life in the hospital such that tensions can be identified and resolved, policies revisited, and greater transparency provided when necessary.
The pilot project proved to be highly successful as assessed by patients, staff, and administrators. Communication patterns improved, patient uncertainty about the status of wound healing decreased, and patients became far more knowledgeable about their illness. Socially, petty disputes were resolved in a far more amicable fashion, and both patients and staff felt vindicated by expressing discontentment and being heard by others, who could then better understand their position.
The pilot project made use of two distinct but complementary forms of problem solving as a means to establish a therapeutic community in keeping with culturally meaningful modes of conflict resolution in Africa. Much has been written in the anthropological literature about the value of both collective and individual forms of conflict resolution in settings ranging from the settling of social disputes between factions in villages, to processes of divination used to air grievances both past and present [25,26,27]. An open forum both facilitated collective problem solving and enrolled public support for one’s position, serving to establish their moral identity [28]. Individual counseling provided a patient the complementary opportunity to speak to an empathetic witness [29] about difficulties that one would not like to share in public, for reasons ranging from embarrassment to spiritual danger.
Is it feasible to transform African hospitals serving long-term patients into therapeutic communities? We would argue that it is feasible given two conditions. First, hospital administrators need to recognize the utility of building a therapeutic community and be willing to engage in the problem-solving processes outlined in this paper. Second, health social scientists need to receive basic training in health systems analysis and conflict resolution as well as hospital ethnography [30,31] and an anthropological approach to patients’ illness experience attentive to their many “works of illness”. Treating patients as active agents in the hospital will serve as a corrective to paternalistic approaches to patient care that treat them as passive recipients of treatment whose only work is compliance with medical advice [32,33]. Life is far more complicated, and when both patient and staff needs are not met, discontent undermines quality of care.
We would end with one last observation. There is another important way establishing a therapeutic community benefits the hospital. Former satisfied patents are positive sources of information about both the hospital and the community based outreach program it has promoted to identify early category one BU cases. As the old adage goes: the best advertisement is a satisfied customer. This is particularly important in a disease like BU, where the reputation of the hospital is essential to the success of community outreach and the entire BU program. Patients educated in wound care as well as BU re-enter the community as a valuable resource and “go to” person for information about the disease and wound management. In Benin, former patients already play an active role in identifying cases of BU in some communities [34]. Increased patient education and a more positive experience in the hospital increases the likelihood that they will refer chronic ulcer patients to health staff they know and trust.
|
10.1371/journal.pbio.1000440 | A Novel Role for Dbx1-Derived Cajal-Retzius Cells in Early Regionalization of the Cerebral Cortical Neuroepithelium | Patterning of the cortical neuroepithelium occurs at early stages of embryonic development in response to secreted molecules from signaling centers. These signals have been shown to establish the graded expression of transcription factors in progenitors within the ventricular zone and to control the size and positioning of cortical areas. Cajal-Retzius (CR) cells are among the earliest generated cortical neurons and migrate from the borders of the developing pallium to cover the cortical primordium by E11.5. We show that molecularly distinct CR subtypes distribute in specific combinations in pallial territories at the time of cortical regionalization. By means of genetic ablation experiments in mice, we report that loss of septum Dbx1-derived CR cells in the rostromedial pallium between E10.5 and E11.5 results in the redistribution of CR subtypes. This leads to changes in the expression of transcription factors within the neuroepithelium and in the proliferation properties of medial and dorsal cortical progenitors. Early regionalization defects correlate with shifts in the positioning of cortical areas at postnatal stages in the absence of alterations of gene expression at signaling centers. We show that septum-derived CR neurons express a highly specific repertoire of signaling factors. Our results strongly suggest that these cells, migrating over long distances and positioned in the postmitotic compartment, signal to ventricular zone progenitors and, thus, function as modulators of early cortical patterning.
| Patterning of the cerebral cortex occurs early during embryonic development in response to secreted molecules or morphogens produced at signaling centers. These morphogens establish the graded expression of transcription factors (TFs) in progenitor cells and control the size and positioning of cortical areas in the postnatal animal. CR cells are among the earliest born cortical neurons and play a crucial role in cortical lamination. They are generated at signaling centers and migrate over long distances to cover its entire surface. We show that three different CR subtypes distribute in specific proportions in cortical territories. Genetic ablation of one subpopulation leads to a highly dynamic redistribution of the two others. This results in defects in expression of transcription factors and in progenitor cell proliferation, which correlate with the resulting changes in the size and positioning of cortical areas. Given our additional evidence that CR subtypes express specific repertoires of signaling factors, the ablation phenotypes point to a novel early role for CR cells as mediators of cortical patterning and suggest that CR cells are able to signal to progenitor cells. Our data thus add to the conventional model that morphogens act by passive diffusion and point to a strategy of morphogen delivery over long distance by migrating cells.
| Patterning is defined as the process by which an equipotent field of cells proliferates and organizes into a complex spatial arrangement of distinct cell types in response to positional information [1]. The coordinated growth and patterning of both adjacent and distant developing tissues along the anteroposterior and dorsoventral axes is crucial to ensuring the harmonious construction of a functional body. Patterning mechanisms in early embryos have been studied for many years. These studies have led to the identification of signaling centers and their secreted molecules, some of which mediate their function in a dose-dependent manner, such as Fgfs, Bmps, Shh, RA, or Wnts, as well as their antagonists. Morphogens appear to affect patterning (growth and cell fate) over long distances. In the developing cerebral cortex, they have been shown to control the graded expression of transcription factors (TFs) in progenitors along the rostrocaudal (RC) and mediolateral axes of the neuroepithelium [2]–[5]. Although a major effort has been directed towards understanding how signaling information travels from the source through the surrounding tissues in both vertebrates and Drosophila [6],[7], how this leads to graded expression of transcription factors and coordinates growth and cell fate over long distance is still an unresolved issue.
The cerebral cortex has a laminar organization in which earlier and later born neurons accumulate according to an inside-out sequence. It is divided into areas which serve distinct functions ranging from motor and sensory to cognitive processing. These territories have a specific size and are positioned at precise spatial coordinates relative to each other. The achievement of such a highly complex architecture requires the exquisite orchestration of the proliferation of progenitors, the spatio-temporal generation of distinct cell types, and the regulation of their migration and final positioning. Commitment to a cortical regional phenotype occurs during early stages of development, between E10.5 and E12.5 [8]–[10]. The Pax6, Sp8, CoupTF1, and Emx2 transcription factors have been shown to regulate the early regionalization of the cortical neuroepithelium and to play a crucial role in controlling the size and position of cortical areas in the postnatal cortex [2]–[5].
Cajal-Retzius (CR) cells are among the first neurons to be generated in the embryonic telencephalon. They start invading the preplate at E10.0–E10.5 in mice [11]–[13] and are subsequently localized in the most superficial layer (marginal zone (MZ)/layer I) of the developing cortex until the first postnatal weeks. Their best documented function is to control the radial migration of neurons and the formation of cortical layers by secreting the extracellular glycoprotein Reelin (Reln) [14]–[16]. Although additional functions for CR cells have been proposed at late stages of development, such as the regulation of the radial glia phenotype [17],[18] and the development of hippocampal connections [19], their molecular properties and function during early corticogenesis remain elusive.
Genetic tracing experiments have disclosed at least three sites of origin of CR neurons at the borders of the developing pallium: the pallial-subpallial boundary (PSB or anti-hem) laterally, the pallial septum (also called retrobulbar area, commissural plate, medial-PSB) rostromedially, and the cortical hem caudomedially [20]–[22]. Hem-derived CR neurons have been shown to predominantly populate the caudomedial and dorsolateral pallium at E12.5 [22]. We have shown that cells expressing the homeodomain transcription factor Dbx1 at the septum and the PSB give rise to two molecularly distinct subtypes of CR cells that migrate over long distances from their origins to primarily populate the rostromedial and lateral developing pallium, respectively [20]. The choroid plexus and the thalamic eminence have also been suggested to generate CR neurons which invade caudoventral telencephalic regions [23]–[25].
In this report, we use genetic ablation to study the role of CR subpopulations in cortical development. We show that changing the dynamic distribution of molecularly distinct CR subtypes in pallial territories influences early cortical regionalization and postnatal arealization by controlling proliferation of progenitors within the ventricular zone (VZ). Our results show that septum-derived CR neurons express a highly specific repertoire of signaling factors and suggest that their secretion might be one of the mechanisms by which CR neurons from the postmitotic compartment contribute to cortical patterning.
Reln, p73 and Calretinin have been used alone or in combination as specific markers to define CR neurons [12],[13],[20],[26],[27]. Whereas Reln appears to be expressed by all CR cells, the expression of these three proteins do not overlap completely in the early preplate or during later stages of development within the MZ. We have previously shown that septum- and PSB-derived CR cells are molecularly distinct and that the former do not express one of the two Calretinin isoforms [20] (recognized by goat anti-Calretinin, Swant). At E11.5, p73 has been reported to be an early marker for CR neurons derived from the cortical hem, prior to their onset of Reln expression, whereas PSB-derived CR cells only express Reln at this stage [26],[27]. While additional genes have been shown to be expressed in CR neurons [25],[28],[29], these were also expressed by other preplate cell populations.
In this study, we used E11.5 Dbx1nlsLacZ animals to genetically label the Dbx1 lineage of CR cells and carried out Reln/p73/βgal triple immunostaining and in situ hybridization for Reln and p73 (Figure S1) in order to better define the distribution of the different CR populations in the pallial MZ. In rostral coronal sections (level L1, Figure 1A), βgal+/Reln+ cells were found all around the telencephalic vesicle and represented most of the Reln+ cells showing that they predominantly derived from the Dbx1 lineage [20]. In dorsomedial (DM) regions, nearly all βgal+/Reln+ cells also expressed p73 (Figure 1B) whereas in the dorsal (D) pallium 60% coexpressed p73 (unpublished data). Progressing from the dorsolateral (DL) to the lateral (L) pallium, the number of βgal+/Reln+ cells that did not coexpress p73 increased (Figure 1C). Less than 10% of hem-derived CR cells (Reln+/p73+/βgal−) were found on such rostral sections. Together with the lack of p73 expression in the lateral pallium during earlier stages (Figure S1), these results show that Dbx1-derived CR cells originating from the septum express p73, whereas those from the PSB do not [26]. They appear to represent the main subpopulations in the medial and lateral pallium, respectively. At intermediate levels along the RC axis (level L2, Figure 1D), septum-derived (Reln+/p73+/βgal+) and hem-derived (Reln+/p73+/βgal−) CR cells represented, respectively, 60–70% and 20–30% of the total CR cells in DM regions (Figure 1E), whereas PSB-derived (Reln+/p73−/βgal+) represented the main CR cells population in L and DL territories (unpublished data and [20]). At caudal levels (level L3, Figure 1I), most dorsally located cells were derived from the hem (Figure 1J) whereas 50–60% of lateral cells originated within the PSB, as reported previously [20]. Co-labeling using βgal, Reln, and ER81, a protein expressed in rostral but not caudal CR cells at this stage [29], further confirmed the distribution of septum and PSB Dbx1-derived CR cells in the rostral pallium (unpublished data). Our results are consistent with a previous study showing that hem-derived CR neurons at E12.5 represent 90% and 60% of p73+ CR cells in the caudomedial and dorsolateral cortex, respectively [22].
Consistent with previous reports [26],[27], our data enable us to distinguish between hem-, PSB-, and septum-derived CR cells (respectively Reln+/p73+/βgal−, Reln+/p73−/βgal+, and Reln+/p73+/βgal+) and further confirm their ability to migrate over long distances from their generation site to become distributed in specific combinations in pallial territories (Figure 1K).
To study the function of Dbx1-derived CR cells, we used Dbx1loxP-stop-loxP-DTA mouse mutants [20], which allow the conditional ablation of Dbx1-expressing cells in a temporally and spatially regulated manner, upon Cre-mediated recombination. Dbx1 is present in young postmitotic CR cells at L2 levels in the pallial (dorsal) septum, as shown by colabeling with Tuj1, Tbr1, and p73, whereas it is expressed in progenitors (Dbx1+/Tuj1−/Tbr1−) at the PSB (unpublished data and Figure 1F–H) starting at E10.5–E11.0 [20]. Dbx1 subpallial expression in the caudal septum/POA area is also detected at more caudal levels [20]. We used the Emx1iresCre line which shows effective recombination in pallial progenitors at E10.5, but recombines in a salt and pepper manner in the VZ of the PSB [22],[30]. On L2 sections of E11.5 ROSA26YFP;Emx1iresCre telencephalons, coexpression of YFP, and Dbx1 was detected in cells located in the MZ at the pallial septum, but not at the PSB (Figure S2A–B). We performed TUNEL staining at E11.5 and E12.5 to visualize the distribution of apoptotic cells (Figure S2C–D and unpublished data). At E11.5 in Dbx1DTA;Emx1iresCre embryos TUNEL+ cells were found in the septum at L2 levels but not in more caudal sections. Effective ablation was confirmed by the loss of Dbx1+ cells in the dorsal septum, but not in the ventral and caudal septum (Figure S2E–H and unpublished data). Some TUNEL+ cells were also detected in the mantle zone of the PSB. The identity of cells lost upon ablation was analyzed using Reln and p73 in situ hybridization (Figure 2A–D, quantified in Figure 2E–F). At E11.5 in L1 sections, a strong reduction in the number of Reln+ and p73+ cells (a 40% and 45% loss, respectively) was detected in the DM pallium of mutant embryos when compared to controls (Figure 2Aii,Bii,Cii,Dii, E–F) in addition to a loss in the ventromedial region where septum-derived CR cells also migrate [20]. No decrease was detected in the DL, L and piriform cortices (Figure 2Aiii,Aiv,Biii,Biv,Ciii,Civ,Diii,Div). On the contrary, quantifications revealed an increase in the number of Reln+ cells in the L pallium of E11.5 mutant embryos (Figure 2E), suggesting an increase in the production or the rostral migration of CR cells at the PSB. In caudal regions, where hem-derived CR cells constitute the main CR cells population, the numbers of Reln+ and p73+ cells were unchanged with respect to controls (Figure S2I–N). We conclude that Dbx1DTA;Emx1iresCre embryos present a specific partial loss of septum-derived CR cells in rostral DM territories of the developing pallium.
Interestingly, at E12.5, the number of Reln+ cells in the rostral (L1) D and DM pallium of Dbx1DTA;Emx1iresCre embryos was similar to controls (Figure 2G–H and S2O). In contrast, that of p73+ cells was still decreased (Figure 2I–J and S2O). At more caudal levels (L2), quantification of p73+ cells showed no differences with respect to control embryos whereas Reln+ cells were reduced in the DM pallium of mutant embryos (Figure 2K–N and S2P). These results indicate that CR cells from the PSB (Reln+/p73−) invade the mutant DM and D regions at rostral levels whereas young hem-derived CR cells (Reln−/p73+)[26],[27] migrate into DM regions at more caudal levels. This effect was already in progress at E12.0 (unpublished data). We conclude that a rapid compensatory migration of CR subtypes into the ablated rostral pallial territories occurs in less than one day following depletion of septum-derived CR cells (Figure 2O).
The cortical primordium has been shown to be committed to a regional identity between E10.5 and E12.5 [8],[9]. The establishment of four opposing gradients of expression of TFs in response to signals from patterning centers at early stages (rostrolateralhigh to mediocaudallow for Pax6; rostromedialhigh to caudolaterallow for Sp8; mediocaudalhigh to rostrolaterallow for Emx2 and caudolateralhigh to mediorostrallow for CoupTF1) is crucial for the formation of tangential subdivisions of specialized cortical areas in the postnatal animal [2]–[5]. This corresponds to the time of generation of CR subtypes (E10.5–E11.5) [12],[13],[20] and indeed all three sources of CR subpopulations correspond to major signaling centers in the developing pallium (hem, septum, PSB). Furthermore, CR subtypes invade specific pallial territories between E10.5 and E12.5 and their distribution (Figure 2O) strongly correlates with the gradients of expression of TFs in the neuroepithelium. Specifically the distribution of hem-, PSB-, and septum-derived CR cells resembles the gradients of expression of Emx2, Pax6, and Sp8, respectively. These observations prompted us to analyze whether early cortical patterning and the establishment of gradients of Pax6, Emx2, CoupTF1, and Sp8 expression were defective in the neuroepithelium of mutant animals at E11.5 and E12.5.
Initial analysis of Pax6 expression in E11.5 control embryos using whole mounts in situ hybridization, showed this was low in the medial and high in the lateral pallium, respectively. In Dbx1DTA;Emx1iresCre embryos, Pax6 expression was reduced in rostromedial and increased in lateral territories when compared to control embryos (Figure 3A,B). This reduction was confirmed using immunohistochemical localization of Pax6 protein at rostral L1 levels (Figure 3G,I,J,L,M,O,P,R) that also showed a decrease in D regions. Notably, by E12.5 an increase of Pax6 expression was observed in both rostromedial (DM) and dorsal (D) regions (Figure S3K,M,N,P) in mutant embryos, suggesting a lateral to medial shift in its expression following an initial downregulation within the medial and dorsal pallium. In control embryos, Emx2 is expressed according to a caudalhigh to rostrallow gradient at medial levels. In contrast, mutant embryos showed a rostral expansion in Emx2 expression both at E11.5 and E12.5, as shown by in situ hybridization (Figure 3C,D and S3A–H), suggesting a caudal to rostral shift in its expression. Sp8 is expressed with a rostromedialhigh to caudolaterallow gradient in controls. In E11.5 mutant embryos, Sp8 expression was found to be downregulated in the most rostromedial and dorsal pallium using whole mounts in situ hybridization as well as immunohistochemistry (Figure 3E,F,H,I,KL,N,O,Q,R). At E12.5 Sp8 expression was strongly upregulated in rostral D pallial territories (Figure S3L,M,O,P) and this extended to DL regions. CoupTF1 is expressed at low levels in the rostromedial and high in the caudolateral pallium. Interestingly, the rostral domain of low CoupTF1 expression appeared larger in mutant embryos at E12.5, but not at E11.5, (Figure S3I,J and unpublished data), suggesting a medial to lateral shift in its expression. Taken together, these results show that TFs expression gradients are displaced upon CR cells ablation and their redistribution. To further characterize the regionalization defects in septum CR cells depleted embryos, we analyzed four additional genes whose expression is also restricted to specific pallial territories, namely Erm and Pea3 (involved in Fgfs signaling) and Wnt7b and Wnt8b. In E12.5 mutant embryos, Erm and Pea3 expression was decreased in DM/D regions whereas that of Wnt7b in the VZ was increased compared to control embryos (Figure 4A–F). Interestingly, the expression of Wnt7b appeared decreased in the MZ. Remarkably, the Wnt8b expression domain in DM pallial regions was expanded rostrally and ventrally both at E11.5 and E12.5 (Figure 4G–L). These results strongly support the conclusion that regionalization is affected in mutant animals and show that Wnt7b and Wnt8b are strongly upregulated in cortical progenitors upon septum-CR cells ablation and redistribution.
Cortical regionalization is known to be controlled by secreted molecules expressed at signaling centers [2]–. However, expression of Fgf8 and Fgf17 in the septum/commissural plate and surrounding regions, of Wnt3a and Wnt5a at the hem, and of Fgf15 and Shh in the subpallial rostral and caudal septum/preoptic area, respectively, were unaltered in Dbx1DTA;Emx1iresCre embryos along both the RC and DV axis between E11.0 and E12.5 (Figure 4M–T, Figure S4A–B and unpublished data). We also did not observe changes in the expression domains of Msx1 and BMP4 at the roof plate (Figure S4C–D and unpublished data). Furthermore, dorsoventral patterning at the septum was similar to controls, as shown by the position of the Ngn2 and Mash1 ventral and dorsal limits of expression (Figure S4G–N), respectively, and the dorsal limit of Fgf15 (Figure 4Q–R). Lastly, no changes in the ventral limit of expression of Gli3 were observed (Figure S4E–F) as well as in the expression of genes at the PSB (Sfrp2 and Tgfα) were observed (unpublished data). Together, these results show that septum-derived CR cells loss at E11.5 and the subsequent compensation by hem- and PSB-derived CR subtypes affect regionalization of the rostromedial and dorsal neuroepithelium without altering gene expression at signaling centers. The downregulation of Sp8/Pax6 at E11.5 and of Erm/Pea3 at E12.5 in the dorsomedial pallium correlate with loss of septum-derived CR cells and suggests a persistent loss of Fgf signaling. At E12.5 the increase in Emx2/Wnt7b/Wnt8b expression in DM territories and the upregulation of Sp8/Pax6 in DM and D regions correlate with compensation by hem- and PSB-derived CR cells.
Changes in patterning of the early neuroepithelium prompted us to analyze neurogenesis in mutant animals. At E11.5, Tbr1 labels preplate postmitotic glutamatergic neurons, including CR cells [31]. In mutant embryos Tbr1+ cells were detected in the correct location, but we observed a 30% decrease in their number in DM and D territories of the rostral pallium (Figure 5A–C), corresponding to Tbr1+ septum-derived CR cells loss. At E12.5, when compensation by other CR cells subtypes had already occurred, Tbr1+ cell number was decreased to 50% in the DM pallium, but no changes were observed in other pallial regions (Figure 5J–L and unpublished data). Decrease of Wnt7b expression in the MZ was also observed at E12.5 correlating with reduced differentiation (Figure 4E,F). Similar numbers of Tbr1+ neurons in rostral DM regions of control and mutant embryos was observed by E13.0 (unpublished data). Quantifications of the number of mitotic cells using immunostaining for PH3 and BrdU following a one-hour pulse at E11.5 revealed that proliferation was decreased in the region depleted in CR cells (Figure 5D–I). At E12.5, the number of mitotic cells in DM was similar in control and mutant embryos, indicating that the number of VZ progenitors undergoing mitosis is decreased temporarily at E11.5 (Figure 5M,N,Q). We also detected an increased number of PH3+ cells at E12.5 in D regions, where Reln+ cells from the PSB had repopulated the preplate (Figure 5O–Q). Thus, septum-derived CR cells loss results in a transient decrease of proliferation at E11.5 in the DM pallium. Repopulation of D regions by PSB-derived CR subtypes correlates with an increase in mitosis of cortical progenitors.
To further dissect the properties of the early neuroepithelium upon depletion of CR cells in the preplate, we analyzed the expression of Tbr2, a marker of early postmitotic cells in the MZ and of basal/intermediate progenitors [32]. Tbr2+ cell number in the preplate at E11.5 was similar in control and mutant embryos (Figure 5W), but decreased at E12.5 in DM regions in mutant embryos (Figure 5R–T). Notably, at E11.5, but not at E12.5, we observed Tbr2+ cells ectopically positioned at the apical side of the neuroepithelium flanking the ventricle in the mutant DM region (Figure 5U,V,X). Together, these studies show that the distribution of CR subtypes in the MZ influences the proliferation and differentiation of progenitor cells within the VZ.
Changes in TFs gradients prompted us to investigate whether arealization was affected in ablated animals. In control brains, Cdh8 is highly expressed in layers II/III of the frontal/motor cortex and visual areas [33],[34]. RORβ is restricted to layer IV in the somatosensory and visual areas [35]. On sagittal sections of P8 control brains at lateral levels, RORβ was expressed in the most rostral territories, where Cdh8 expression in superficial layers was not detected (Figure 6A,C). At corresponding lateral levels in Dbx1DTA;Emx1iresCre mutant brains a large domain of Cdh8 expression in superficial layers was observed in rostral territories (Figure 6A–B). Serial sections hybridized with RORβ showed a lack of expression in the rostral region expressing Cdh8 (Figure 6C,D) indicating that, at this level, somatosensory area is replaced by frontal/motor area. The size of the caudal domain of Cdh8 expression, corresponding to the visual area, appeared normal at this level. Notably, a small caudal domain, where RORβ is normally absent in control brains, likely corresponding to the lateral parietal association cortex, appeared enlarged in mutant brains (Figure 6C,D). Analysis of coronal sections at rostral levels showed a lateral shift of motor areas at the expense of somatosensory areas (Figure 6E–H). However, at more caudal levels the size of the RORβ expression domain appeared normal. RORβ expression analysis together with SERT immunostaining which identifies the barrelfield showed that primary somatosensory territories did not appear to change size or identity, but rather that their positioning was displaced caudally and laterally (unpublished data). Sagittal sections at medial levels also showed that the rostral domain of Cdh8 expression was reduced, whereas the caudal domain, corresponding to retrosplenial cortex, was rostrally displaced in mutant animals (Figure S5S,T). Lastly, the characteristic high expression of Cdh8 within lower layers of retrosplenial territories appeared to be shifted more rostrally, at the expense of cingulum regions (Figure S5S,T).
Quantifications of relative cortical area sizes on whole mount P0 brains, using Cdh8 (Figure S5A–D) and Lmo4 (unpublished data) as markers, showed a small but significant increase in the frontal/motor cortex with no differences in visual regions or overall neocortical size. Finally, although CR cells have been shown to play a role in cortical layers formation, lamination in each area appeared to be fairly normal (Figure 6 and Figure S5I–N), as reported upon hem-derived CR cells ablation [22].
Together, these results show that deletion of septum CR cells results in the rostral displacement of the retrosplenial cortex, an increase in the size of motor area and a medial-to-lateral and rostral-to-caudal shift in the position of the somatosensory area (Figure 6U). To further confirm that the observed defects were due to CR cells ablation, we used the deltaNp73 mouse line, which specifically labels and targets Cre recombination in CR subtypes [24]. Dbx1DTA;deltaNp73 embryos at E11.5 showed a specific decrease of septum-derived CR neurons (Reln, p73, TUNEL staining, Figure S5O–R and unpublished data) together with changes in cortical regionalization (Pax6, Sp8, Emx2), similar to ablation using Emx1iresCre animals (unpublished data). Analysis of Cdh8, RORβ and Lmo4 expression in P0 (Figure S5E–H and unpublished data) and P8 Dbx1DTA;deltaNp73 animals using Lmo4, Cdh8, RORβ, and Bhlhb5 confirmed an increase in the size of motor area and a caudolateral shift in the positioning of the somatosensory area (Figure 6I–T). Thus, we conclude that CR neurons mediate the regionalization/arealization defects observed in mutant animals.
To gain insights into the mechanisms by which CR neurons affect cortical regionalization, we used microarray analysis of purified septum Dbx1-derived CR cells to identify candidate secreted signaling molecules. Dorsomedial (DM) and dorsolateral (DL) pallial regions at rostral L1/L2 levels were dissected from E12.5 Dbx1CRE;ROSA26YFP embryos. Our data showed that most YFP+ cells in the DM part (≥95%) were Reln+/p73+ septum-derived CR cells whereas the DL one was enriched in Reln+ PSB-derived CR cells (∼55–65%) (see Materials and Methods, Figure S6A–C, [20] and unpublished data). YFP+ Dbx1-derived cells were FACS-sorted (Figure S6D–I) and RNA expression profile was analyzed using Affymetrix whole mouse transcript microarrays.
The identity of purified CR cells was confirmed by the presence of Reln and p73 in both DM and DL sorted cells together with the expression of other preplate markers (Figure 7A). We found that multiple secreted molecules were differentially expressed in DM and DL YFP+ cells (Figure 7A). Notably, we detected ∼10-fold higher levels of Fgf15, Fgf17 and Fgf18 in DM with respect to DL cells (Figure 7A and Table S1). In particular, Fgf15 and Fgf17 were expressed at high levels in the DM population, which contrasted with their relatively low levels of expression of Fgf18 and Fgf8 (Figure 7A and Table S1). In addition, although the total detected levels of expression were quite low, the expression of several other genes that encode secreted signaling factors (Wnt5a, Wnt5b, Wnt8b, Tgfβ2, Dkk2, Fstl1, Slit2, and Igf2) was comparatively ≥3–5 fold higher in DM CR cells (Figure 7A and Table S1). Further analysis of the expression of six genes (Fgf15, Fgf17, Fgf18, Fgf8, Wnt3a, and Wnt7b) using qPCR in YFP+ sorted DM and DL cells confirmed that Fgf15 and Fgf17 were expressed at higher levels in DM versus DL cells (Figure 7B and Table S2). The expression of Fgf18 was more variable and did not show a significant difference between the two cell types, likely reflecting the low levels of expression revealed by microarray analysis. qPCR analysis also confirmed the absence of Fgf8 and Wnt3a in both DM and DL cells and a higher expression of Wnt7b in DL cells.
These results show that septum-derived CR neurons express a highly specific repertoire of signaling factors and, in particular, high levels of Fgf15 and Fgf17, which have well-established roles in cortical patterning. Together, our data suggest that secretion of signaling molecules might be one of the mechanisms by which CR neurons contribute to cortical patterning, thereby enabling a refined interplay of multiple signaling pathways which might be crucial to this end.
In this study, we show that specific combinations of CR subtypes dynamically populate distinct regions of the developing pallium. Genetic ablation of septum Dbx1-derived CR cells and redistribution of PSB- and hem-derived CR subtypes between E11.0 and E12.5 in cortical territories result in changes in early patterning events and progenitor cell division and differentiation at long distance from CR cells generation sites. These early regionalization defects correlate with changes in the size and positioning of cortical areas at postnatal stages, without affecting signaling centers. Moreover, we show that septum-derived CR cells express a specific combination of secreted factors. Together, our results show that the distribution of CR subtypes in the preplate/MZ controls VZ progenitor properties and strongly point to a novel role of CR cells subtypes as mediators of early cortical patterning.
The molecular identity of CR neurons has been subject to debate and three main markers have been considered to be expressed in CR cells: Reln, p73, and Calretinin [12],[13],[20],[26],[27]. Together with previous reports [26],[27], our data show that p73 is expressed in CR cells generated at the hem- and the septum- but not in PSB-derived CR neurons. Reln appears to be expressed by all CR subtypes although the onset of its expression might be slightly delayed in hem- and septum-, with respect to PSB-derived cells. By studying the time course of the expression of these markers, we were able to map the distribution of the distinct subtypes and to show that upon elimination of septum Dbx1-derived CR cells, rostral dorsal territories are repopulated by PSB-derived CR cells and medial regions by hem-derived CR cells. The dynamic redistribution of CR subtypes upon ablation occurs very rapidly within a 24-hour period and is mostly accomplished by E12.5. Our results strongly point to the existence of a crosstalk between CR cells involved in regulating invasion of cortical territories at early stages of development, which are in agreement with previous reports suggesting that contact-inhibitory interactions between CR cells might control their dispersion throughout the surface of the cortex [20],[36]. Our data suggest that the sites of generation, the birthdates and the onset/speed of migration of CR subtypes are crucial for their kinetics of arrival and, thus, their distribution in pallial regions, and that these form a precise molecular map at E12.5. Small variations of these parameters might therefore have profound consequences on the construction of this map and may possibly occur among individuals. Invasion of neocortical territories by CR cells derived from other sources has also been shown in hem-ablated mouse mutants [22]. However, the suggested progressive increase of hem-derived CR cells in prospective neocortical regions at later stages of development, together with an almost complete loss of CR neurons in the neocortex of these mutants, might indicate that the distribution of CR neurons from midcorticogenesis is controlled by additional mechanisms, including selective survival of CR subtypes.
Multiple signaling centers or “organizers” have been shown to be involved in the induction and patterning of early telencephalic territories. In the cortical primordium, signaling molecules are thought to control the graded expression of TFs, among which Emx2, Pax6, CoupTF1, and Sp8 are involved in early cortical regionalization and arealization [2]–[5],[37]. We have shown that loss of CR cells correlates with opposite changes in Sp8/Pax6 and Emx2/Wnt8b expression in DM/D pallium at E11.5, whereas compensation by hem- and PSB-derived CR cells correlates with those in Emx2/Wnt7b/Wnt8b and Pax6/CoupTF1 at E12.5, respectively. These early regionalization changes parallel defects observed in arealization in postnatal animals. These results strongly support the notion that a fine regulation of the levels of expression of TFs, as well as that of players in the Wnt and Fgf signaling cascades, in each territory controls cortical areas positioning and size according to the “cooperative concentration model” [38]. Our data suggest that the regions of intersection of regionalization gradients are highly sensitive to changes and are crucial for setting up the borders of cortical areas. To complement “loss-of-function” (ablation of septum CR cells) and “gain-of-function” (repopulation by other CR subtypes) in vivo, we have attempted to perform grafting experiments in vitro using FACS sorted septum-derived CR cells. However, technical limitations, related to the number of sorted cells together with the culture conditions which do not preserve regionalization gradients of TFs or neurogenesis as in vivo, render such paradigm extremely challenging, if possible.
Our results are consistent with loss and gain-of-function experiments in mice. An anterior expansion of Wnt8b expression occurs upon a decrease in Fgf8 signaling [39] as well as in Pax6 mutants [40], whereas a reduction is detected in Emx2 mutants [40]. Defects of frontal cortical regions and a caudalization of medial cortical territories when septum-derived CR cells are ablated recapitulate some of the defects observed in Fgf17 and Fgf8 mutants and is consistent with an antagonistic regulation of Erm, Pea3, and Sp8 by Fgfs and Emx2/Wnt signaling [41]. Since Sp8, Erm, and Pea3 are genes induced by Fgf signaling [39],[42],[43] and Wnts are involved in graded expression of Emx2 [44], our results are consistent with septum CR cells being involved in maintaining Pea3 and Sp8 and, thus, mediating Fgf signaling. The expression of multiple Fgfs in purified septum-derived CR cells also parallels the changes in the regional-specific expression of downstream targets of the Fgf cascade in mutant animals. Furthermore, an expanded Pax6 expression at E12.5 at rostral dorsal levels correlates with an increase in motor area size and, together with a decrease in CoupTF1 expression, in a shift in the positioning of motor and somatosensory areas. Conversely, a loss of Dbx1 expression at the PSB, as observed in NesCre;Dbx1DTA (unpublished data) or Pax6 mutants [45], correlates with a decrease in the size of the motor cortex in these mutants. Together, these data suggest that PSB-derived CR cells are involved in mediating anti-hem signaling.
In addition, except for few studies in which Reln or Calretinin have been globally analyzed, such as in Emx2, Pax6, and CoupTF1 mutants [46]–[48], CR subtypes have not been analyzed in detail or at early enough stages leaving open the possibility that some of the effects observed in mutants for genes involved in regionalization/arealization might be mediated by differences in CR cells generation and/or their migration. Moreover, in p73 mutants differences in Calretinin expression at early stages were correlated with a dorsal shift of the entorhinal cortex and a reduced size of the occipital and posterior temporal areas [49], suggesting an involvement of CR cells in cortical arealization.
All three sites of CR subtypes generation coincide with patterning centers. Dbx1 expression domains at the septum and the PSB both reside in the immediate vicinity of, and possibly overlap with, regions highly enriched in Fgf and Wnt antagonists signaling. Inasmuch as no differences were observed in the domain of expression of Fgf8, Fgf17, Fgf15, Wnt3a, Msx1, and Shh as well as in DV patterning, it seems unlikely that CR cells ablation affects signaling centers themselves, and we rather propose that it modulates steps downstream from them. This is consistent with the timing of birth and ablation of CR cells, which occur at E10.5–E11.0, and, thus, later than the period when gene expression at signaling centers is not yet fixed, as in the case of the increase of Fgf8 in Emx2 mutants at E9.0 [42]. Moreover, it was recently demonstrated, using gain-of-function experiments, that Fgf8 induces the generation of rostral CR subtypes and that Fgf8, Pax6 and Emx2 loss-of-function mutants present defects in CR subtypes specification [29]. In the present manuscript we show that additional sources of Fgf15 and Fgf17 from migrating septum-derived CR cells exist in the developing pallium. Fgf8 has been shown to lie upstream of other Fgfs, notably Fgf17, and Fgf15 to be a modulator of Fgf8 signaling [41],[50]. Together, these data suggest that the role of Fgf8 in cortical regionalization and arealization might be mediated, in addition to passive diffusion, by secretion of morphogens by migrating Cajal-Retzius neurons. A fine tuning of the concentration of these factors is likely to tightly balance proliferation and differentiation of cortical progenitors. These results also unravel a mechanism by which Fgf8 might affect regionalization, which up to now has possibly been underestimated.
A rostro-lateralhigh to caudo-mediallow gradient of neurogenesis has been shown to distinguish pallial territories during development and to correlate with differences in neuronal numbers and lamination in cortical areas [51],[52]. How the molecular mechanisms controlling neurogenesis and lamination are linked to those of regionalization and arealization is still an open question. We show that septum-derived CR cells loss at E11.5 results in a decrease in proliferation (PH3+) and an increase of apical Tbr2+ cells, suggesting the precocious generation of postmitotic preplate neurons (Tuj1+, unpublished data) at the expenses of neuroepithelial self-renewing cell divisions [53]. By E12.5, a recovery in rostral DM proliferation and an increase in mitosis in D regions correlate with compensation by hem-derived and PSB-derived CR cells, respectively. This transient and early effect (even before the generation of layer VI neurons) is consistent with no major differences in the number of neurons in deep cortical layers. Since lower frequencies of differentiative divisions have been reported during early corticogenesis in reeler mutants [54] and we detect no differences in Tbr2 staining in these mutants (unpublished data), it is unlikely that Reln is responsible of the effects observed at E11.5. We find that ablation of the septum CR subtype does also not result in “inverted” cortical lamination, as would be expected from a loss in Reln signaling as in the reeler mutants. This is consistent with the absence of defects in layers formation in the neocortex of hem-ablated mutants [22] and strongly suggests that low Reln levels are sufficient to preserve cortical lamination [55]. Furthermore, defects in arealization have not been reported in reeler mutants [54], although the expression of molecular markers has not been so far analyzed in these mutants. Together these results strongly suggest that since all CR cells subtypes express Reln, and that hem- and PSB-derived CR cells rapidly repopulate the ablated regions, the role of CR cells in cortical arealization appears to be Reln-independent.
Passive diffusion occurs very efficiently at short time scales over a few dozens of cell diameters [56]. Various mechanisms can influence morphogen delivery in tissues surrounding a source, such as endocytosis and subsequent degradation [57], trapping in the extracellular matrix [58] or diffusion in the ventricular fluid [59],[60]. Nevertheless, as development proceeds and the cortex grows, this results in the increase in the distance between signaling centers. Additional mechanisms might be necessary to maintain and coordinate the growth and spatial patterning of the cortex and, thus, the robustness of morphogens signaling. Examples of “migrating” cells/structures affecting the development of distant territories are neural crest cells originating at the mid-forebrain junction and affecting craniofacial and anterior neural tube growth/survival in chick [61]–[63], and axonal projections affecting cell cycle progression of cortical progenitors [64]. CR cells are generated in regions highly enriched in signaling molecules and migrate in close contact with the early cerebral cortex neuroepithelium. Co-culture experiments with semipermeable membranes in the cerebellum provided evidence that neocortical CR cells release soluble signals other than Reln, thereby influencing the radial glia phenotype [18]. Even though we do not rule out additional mechanisms, such as CR cells carrying patterning molecules tethered to their cell surface or forming cell-cell contacts with radial glia basal attachments, our data strongly suggest that different CR subtypes act as mediators of cortical patterning by secreting a variety of ligands, including Fgfs and Wnts. Our data hint at CR cells participating in the fine tuning of multiple signaling pathways which is likely to underlie the regulation of cortical regionalization. Thus, we propose that these highly motile cells have a crucial role and serve as “mobile patterning units” at early stages of development.
All animals were handled in strict accordance with good animal practice as defined by the relevant national and/or local animal welfare bodies, and all mouse work was approved by the Veterinary Services of Paris (Authorization number: 75-1454).
In this study, we used a Dbx1nlsLacZ mouse line [20],[65] to trace Dbx1-derived cells. Dbx1nlsLacZ/+ embryos allow the transiently labelling of Dbx1-derived cells starting at their generation site and during the first phases of their tangential migration, due to the persistence of the β-galactosidase protein in the cells. In order to analyze the effect of eliminating Dbx1-derived CR cells, we inserted an IRES-loxP-stop-pGKneo-loxP-DTA (diphtheria toxin) cassette into the dbx1 locus by homologous recombination (Dbx1loxP-stop-loxP-DTA) [20]. A functional DTA is expressed exclusively upon Cre-mediated recombination. Mutant animals were crossed with a Emx1iresCre [30] mouse line which expresses the Cre recombinase in pallial progenitors. Dbx1loxP-stop-loxP-DTA and Emx1iresCre animals were used as controls for all experiments. ROSA26loxP-stop-loxP-YFP mice [66] crossed with Emx1iresCre and Dbx1CRE/+ [20] were used to permanently trace cells derived from Emx1- and Dbx1-expressing cells, respectively. Permanent tracing using Dbx1CRE;ROSA26YFP embryos is very similar to that using Dbx1CRE; βactin:lacZ and Dbx1CRE;TAUGFP at E12.5, as previously reported [20]. We also used deltaNp73 animals which have been engineered to express the Cre recombinase in the p73 locus and, thus, in CR subpopulations [24]. Embryos and postnatal animals were genotyped by PCR using primers specific for the different alleles. For BrdU experiments, E11.5 embryos were obtained from females injected intraperitoneally with a single dose of BrdU (50 mg/kg) one hour prior to collection.
For staging of embryos, midday of the vaginal plug was considered as embryonic day 0.5 (E0.5). Embryos for immunohistochemistry were fixed by immersion in 4% PFA, 0.1 M phosphate buffer (PB) pH 7.2 for 2 h at 4°C and rinsed in PBS for 2 h. Postnatal animals were anesthetized and perfused with 4% PFA for 10 min. Brains were cryoprotected overnight in 30% sucrose, 0.1 M PB and embedded in O.C.T. compound (Sakura). Embedded tissues were sectioned on a cryostat with a 14 µm step for embryonic stages and 50 µm for postnatal brains. In situ hybridization on sections and whole-mount preparations was performed as previously described [65]. In situ hybridization probes used in this study were mouse Bhlhb5, Cdh8, CoupTF1, Emx2, Erm, Fgf8, Fgf15, Fgf17, Gli3, Lmo4, Msx1, Pax6, p73, Pea3, Reln, RORβ, Shh, Sp8, Wnt3a, Wnt7b, and Wnt8b. For BrdU experiments, sections were incubated for 10 min in 4% PFA, 0.1M PB, rinsed 3 times in PBS and permeabilized with 4N HCl for 5 min. Immunohistochemistry on sections was performed as previously described [65]. Primary antibodies were rabbit anti-β-galactosidase (Rockland; 1∶1000), G10 mouse anti-Reelin (Calbiochem; 1∶1000), goat anti-p73 (Santa Cruz; 1∶200), rabbit anti-Tbr1 (Chemicon; 1∶1000), rabbit anti-Tbr2 (Chemicon; 1∶2000), rabbit anti-PH3 (Upsdate; 1∶500), rabbit anti-GFP (Molecular Probes; 1∶1000), rabbit anti-Dbx1 (gift of S. Morton and T.M. Jessell, 1∶10000), chicken anti-β-galactosidase (AbCam; 1∶2000), mouse anti-Mash1 (BD Pharmingen; 1∶100), mouse anti-Ngn2 (gift of D.J. Anderson; 1∶10), goat anti-Sp8 (Santa Cruz; 1∶8000), mouse anti-Pax6 (DSHB; 1∶50), and rat anti-BrdU (Accurate Chemical; 1∶400). All fluorescent secondary antibodies were purchased from Jackson ImmunoResearch. Tbr1 antibodies were also detected with biotinylated secondary antibodies using the Elite ABC kit (Vector). TUNEL staining was performed according to the manufacturer's protocol (Roche). The triple immunohistochemistry using rabbit anti-Tbr1 and rabbit anti-Dbx1 in Figure 1 was performed with Zenon Alexa Fluor 647 Rabbit IgG according to the manufacturer′s protocol (Invitrogen).
Whole mount brain pictures were acquired using a digital camera (Zeiss Axiocam HRc) coupled to a binocular lens (Leica MZFLIII), brightfield pictures of telencephalon sections using a colour camera (Zeiss Axiocam HRc) coupled to a Zeiss Axiovert 200 microscope, and immunofluorescence pictures using an inverted confocal microscope (Leica TCS SP5 AOBS Tandem resonant Scanner). Most images in brightfield are composites and were acquired using an AxioVision 4.6 software, options Mosaix and Tiling, which automatically acquires multiple images on the same specimen and reconstructs the final image.
For all experiments, results have been obtained from at least three pairs of control and mutant littermates. Quantification of cell numbers was carried out at several levels along the rostrocaudal axis, namely L1, a rostral level where Dbx1 protein and mRNA were not detected; L2, a level where Dbx1 protein and mRNA are detectable and L3, a caudal level (at the choroid plexus level). On each section, the number of cells was counted in boxes that were placed over the DM, D, DL and L pallium in region-matched control and mutant sections. First, a box for the DM region was defined as the dorsal half of the medial wall in between the dorsal and the ventral morphological hinges. The same plial length was then used to define matched size boxes for the three other regions. All measurements realized in these boxes were normalized in number of cells per 100 µm (measured at the pial surface for Reln, p73, Tbr1 and Tbr2 and at the ventricular surface for PH3 experiments). Lengths and sizes measurements were done using Image J Software. For all quantifications, normal distribution was confirmed and unpaired, two-tailed t test on group means were performed for statistical analysis, using Microsoft Excel software.
E12.5 Dbx1CRE;ROSA26YFP embryos were selected using a fluorescent stereomicroscope. DM and DL cortical regions at L1 and L2 levels were dissected as shown in Figure S6 by separating them at the dorsal hinge. The percentage of septum and PSB Dbx1-derived CR cells in each sample was estimated using immunohistochemistry and in situ hybridization for Reln and p73 on matched sections, as for Figure 1 and Figure 2, and Calretinin [20]. Explants were kept in cold Hank's (Invitrogen) and treated with 0.25% trypsin (Invitrogen) at 37°C for 5 min. After digestion, 0.1% FBS serum (Invitrogen) was added and cell suspension was obtained by mechanical dissociation. Dissociated cells were filtered through a 50 µm nylon mesh filter (celltrics Partec) and propidium iodide (PI) (0.1 µg/mL, final concentration) was added to cells immediately prior to analyzing and sorting. Analysis and cell sorting were performed using an Influx 500 cell sorter (Cytopeia, BD Biosciences since 2008, San Jose, CA, USA). Yellow fluorescent protein (YFP) and PI were excited with solid-state laser 488 nm, 200 mw (Coherent sapphire) and their emission signals were detected using a 528/38 nm band pass (BP) filter and a 610/20 BP, respectively. Fluorescence data were displayed on four-decade log scales. Sorts were performed at low pressure (15 PSI) with a 100 micron nozzle. Positive and negative YFP cell populations were collected simultaneously from the same sample, excluding dead cells by gating on negative red fluorescence (PI−) regions. YFP− embryos were used as control for fluorescence. The purity of the YFP+ sorted cells using the established windowing level was confirmed by analysis under a fluorescent microscope and estimated to be above 98%. An average of 6 090 YFP+ cells for DL and 22 150 for DM were obtained from 10 embryos dissected from 3 litters.
Each experimental condition was tested in duplicates. Total RNA was extracted from YFP+ cells using the RNeasy mini Kit (Qiagen) following the manufacturer's protocol. Biotinylated cRNAs were prepared from 3 to 5 ng of total RNA using the GeneChip Expression 3′ Amplification Two-Cycle Target Labeling and Control Reagents, according to the manufacturer's instructions (Affymetrix, Santa Clara, CA, USA). Following fragmentation, cRNAs were hybridized for 16 hours at 45°C on GeneChip Mouse Genome 430 2.0 arrays, interrogating over 39 000 transcripts. Each microarray was then washed and stained using the EukGE-WS2v5_450 protocol on a GeneChip fluidics station 450 and further scanned with a GeneChip Scanner 3000 7G. Image processing and analyses were performed using GeneChip Operating Software (GCOS) version 1.4. Absolute and comparison analyses between experimental conditions were conducted using the statistics-based Affymetrix algorithms MAS-5.0 [67],[68] with default settings and global scaling as normalization method. The trimmed mean target intensity of each chip was arbitrarily set to 100 and were selected the genes which showed an “increase” or “marginal increase” difference as the result of this statistical analysis. Additional data analysis was performed using MatLab (The MathWorks, USA) and Mev v4.4 (TM4 microarrays software suite) [69].
20 ng of RNA extracted from YFP+ cells were used for cDNA synthesized with the SuperScript VILO cDNA Synthesis Kit (Invitrogen), following the manufacturer's instructions. Real-time PCR was carried out on a Roche LightCycler according to the manufacturer's instructions for the SYBRGreen detection kit. Primers were designed using PrimerBank [70] and Primer3 [71]. The primers were verified for specificity with Primer-Blast from NCBI. Expression of each gene was calculated relative to that of the mRNA for the ribosomal protein rpS17 [72] in the same sample for the DM and DL cell types and in three independent experiments. Relative quantifications of gene expression were calculated as described by Livak and Schmittgen [73]. The PCR efficiency for each primer pair was estimated with the LightCycler software using a calibration dilution curve for each primer set.
|
10.1371/journal.ppat.1005656 | Chimeric Mice with Competent Hematopoietic Immunity Reproduce Key Features of Severe Lassa Fever | Lassa fever (LASF) is a highly severe viral syndrome endemic to West African countries. Despite the annual high morbidity and mortality caused by LASF, very little is known about the pathophysiology of the disease. Basic research on LASF has been precluded due to the lack of relevant small animal models that reproduce the human disease. Immunocompetent laboratory mice are resistant to infection with Lassa virus (LASV) and, to date, only immunodeficient mice, or mice expressing human HLA, have shown some degree of susceptibility to experimental infection. Here, transplantation of wild-type bone marrow cells into irradiated type I interferon receptor knockout mice (IFNAR-/-) was used to generate chimeric mice that reproduced important features of severe LASF in humans. This included high lethality, liver damage, vascular leakage and systemic virus dissemination. In addition, this model indicated that T cell-mediated immunopathology was an important component of LASF pathogenesis that was directly correlated with vascular leakage. Our strategy allows easy generation of a suitable small animal model to test new vaccines and antivirals and to dissect the basic components of LASF pathophysiology.
| Lassa fever is an arenaviral hemorrhagic fever that causes high morbidity and mortality in West Africa. Unfortunately, the lack of immunocompetent small animal models of disease has precluded understanding of the basic pathophysiology mechanisms of Lassa fever. Here we show how transplantation of a wild-type hematopoietic system into type I interferon deficient recipient mice results in a lethal mouse model of disease. Importantly, experimental infection of these chimeric mice reproduced key features of Lassa fever in humans and underscored the important role of T cells on Lassa fever-associated immunopathology.
| Lassa fever (LASF) is a severe zoonotic viral infection caused by Lassa virus (LASV), an Old World arenavirus whose natural reservoir is the multimammate rat Mastomys natalensis. In endemic West-African countries such as Sierra Leone, Nigeria, Guinea and Liberia, LASF causes yearly epidemics responsible for three hundred thousand cases and more than five thousand deaths [1,2]. Clinically, LASF ranges from mild disease with non-specific signs such as malaise, asthenia, fever and myalgia to severe hemorrhagic fever, usually with the involvement of liver and kidney failure, edema of the face and neck, hypovolemia, hypoxemia, and in some cases mucosal bleeding [3]. However, the pathophysiological mechanisms responsible for these very different disease manifestations are poorly understood.
There is indirect evidence that the host immune response plays an important role in the severity of LASF. Two proteins encoded by the virus, the nucleoprotein (NP) and the Z protein, have been shown to disrupt host innnate immunity by several mechanisms leading to inhibition of the type I interferon (IFN-I) antiviral response [4–6]. The ability of LASV to inhibit IFN-I has also been linked to the capacity of the virus, at least in vitro, to prevent activation of monocyte-derived dendritic cells (DCs) [6,7]. While these findings provide proof of LASV antagonism in vitro, there is no evidence that virus-mediated immunosupression mediates LASV immunopathology in vivo. On the contrary, accumulating research suggests exacerbated host immune responses as a key component of LASF [3,8,9].
In humans recovering from LASF very little humoral response is observed. The neutralizing antibody titers are low and only detectable months after recovery [10,11]. Thus, it has been long thought that recovery from LASF in humans is associated with effective T cell responses [9,12]. This hypothesis has been further strengthened by experimental LASV infections in non-human primates, which indicated that recovery from LASF was associated with early detection of circulating antigen-experienced T cells [13]. In addition, a wealth of data from inbred mice infected with the prototypic old-world arenavirus lymphocytic choriomeningitis virus (LCMV), have demonstrated an important role of CD8 T cells on early control of LCMV infection in this model system [14]. Paradoxically, poor early control of LASV replication results in subsequent T cell-mediated immunopathology, a phenomenon that, at least in LCMV models, has been related to accumulation of CD8 T cells refractory to immune regulation, which in turn, promote vascular permeability [15]. In addition, mice expressing human HLA-A2 are susceptible to LASV due to poor control of early viral replication and late T cell-mediated immunopathology [8]. These previous findings raised the question of whether this deleterious effect was restricted to a particular HLA or was a pathophysiological mechanism of LASF.
A major hurdle to provide answers to these questions and investigate this general area, is that basic reseach into LASV immunology and pathogenesis has been greatly impaired by the lack of small immunocompetent animal models of disease. This is a similar case with other hemorrhagic fevers such as those caused by filoviruses and flaviviruses. With the exception of CBA mice inoculated intracranially [16], inbred laboratory mice are highly resistant to experimental infection with LASV, and only knockout mice with immune-associated deficiencies such as IFN-I receptor knockout (IFNAR-/-), and Signal Transducer and Activation of Transcription (STAT)-1 knockout mice, have been shown to be susceptible to LASV [17,18]. Despite the applicability of these models as in vivo platforms to test antiviral treatments, immunodeficient mice cannot be used to dissect the mechanisms of LASV immunity. In this study, this research question was addressed by developing a mouse model that was susceptible to LASV with intact hematopoietic immunity. Transplantation of wt bone marrow progenitor cells into irradiated IFNAR-/- mice, rendered them susceptible to LASV. The infection was highly lethal and associated with viremia, liver damage and edema. In addition, immunopathology was a key component of LASF pathogenesis, was dependent on CD8 T cells and was directly correlated with vascular leakage. Thus, bone marrow transplantation allowed the establishment of an in vivo model retaining competent hematopoietic-driven immunity and which reproduced key features of LASF pathophysiology.
This study was carried out in strict accordance with the recommendations of the German Society for Laboratory Animal Science and under the supervision of a veterinarian. The protocol was approved by the Committee on the Ethics of Animal Experiments of the City of Hamburg (permit no. 125/12). All efforts were made to minimize the number of animals used for experiments and to alleviate suffering of animals during experimental procedures. All staff carrying out animal experiments passed an education and training program according to category B or C of the Federation of European Laboratory Animal Science Associations. The animal experiments in this study are reported according to the ARRIVE guidelines (Kilkenny C, Browne WJ, Cuthill IC, Emerson M, Altman DG, Improving bioscience research reporting: the ARRIVE guidelines for reporting animal research. PLoS Biol. 2010;8: e1000412) A total of 145 mice were used for this study and all mice were included in the analysis.
LASV strains AV, Ba289, Nig04-10, Nig-CSF and Morogoro virus (MORV) strain 3017/2004 have been isolated in our laboratory and have been passaged ≤ 3 times before their use in this study. LASV strains Ba366 and Josiah, Mopeia virus strain AN21366 (MOPV) and Mobala virus strain 3099 (MOBV) have been obtained from another lab with an unknown passage history and have been passaged in our laboratory ≤ 3 times before use in this study. All virus stocks have been grown on Vero E6 cells, quantified by immunofocus assay and stored at -80°C until use. All experimental infections described in this study were performed within the biosafety level 4 (BSL4) facility at the Bernhard Nocht Institute for Tropical Medicine in Hamburg in accordance with institutional safety guidelines. Personnel wore appropriate protective equipment (biosafety suits).
Murine bone marrow derived macrophages and monocytes were obtained from wild type C57BL/6 mice. Eight to twelve week-old female mice were euthanized and bone marrow from tibiae and femurs was harvested as described elsewhere. Red blood cells were lysed (BD Pharm Lysing Buffer) and cells were seeded with a density of 106 cells/ml in 6-well plates. For differentiation into dendritic cells, cell culture medium (RPMI with 10% FCS) was supplemented with 50 ng/ml of GM-CSF (PreProtech) or with GM-CSF conditioned media from X63-GM-CSF cells. For differentiation of bone marrow progenitors into macrophages, the medium was supplemented with M-CSF conditioned media from L929 cells. Cells were infected with a multiplicity of infection (MOI) of 0.01 with different arenaviruses as indicated, and daily samples were taken for up to 7 days post-infection. Infectious viral particles were quantified using immunofocus assay as previously described [19].
IFNAR-/- mice (C57Bl/6 background) were obtained from the Friedrich Loeffler Institute, Isle of Riems, Germany, and bred in the Specific Pathogen Free animal facility of the Bernhard Nocht Institute for Tropical Medicine. C57BL/6J and CD45.1+ congenic B6 mice, CD11b-DTR (B6.FVB-Tg(ITGAM-DTR/EGFP)34Lan/J), and CD11c-DTR (B6.FVB-Tg(ITGAX-DTR/EGFP)57Lan/J) mice were purchased from Jackson Laboratories and bred at the Heinrich Pette Institute animal facility. Bone marrow chimeric mice were generated at the Heinrich Pette Institute animal facility. Four to eight week old female mice were irradiated twice (550 rad, 4 hr apart by a Caesium source) and reconstituted with 2x106 bone marrow cells from donor mice as previously described [20]. Split irradiation reduced the number of animals succumbing to irradiation to under 3%. Three to five mice per group were kept together in individually ventilated cages (IVC) and had ad libitum access to food and water. Engraftment in peripheral blood was evaluated 4 weeks after reconstitution by flow cytometry and the experiments were performed 4–5 weeks after transplantation. Chimeric mice showing reconstitution of donor hematopoietic cells above 85% were selected for the experiments. More than 90% of chimeric mice generated had this level of donor engraftment or higher.
Five to thirteen animals per group were infected intraperitoneally (i.p.) with 1000 focus forming units (FFU) of LASV or MORV in 100 μl DMEM containing 2% FCS. Infected mice were monitored daily for signs of disease and body weight and rectal body temperature (Thermometer BIO-TK8851 with BIO-BRET-3 rectal probe) were measured daily. Animals with severe signs of disease such as seizures, bleeding, abdominal distention, diarrhea, agony, temperature < 28°C or weight loss > 20% were euthanized as per our approved protocol guidelines. For evaluation of clinical chemistry and viremia, 30–50 μl of blood was drawn by tail vein puncture every 3–7 days over a period of 21 days (≤ 6 blood samples). When criteria for euthanasia were fulfilled or at the end of the experiment, animals were euthanized with an isoflurane overdose followed by cervical dislocation. Organs were collected from 1–3 randomly chosen mice per experimental group of 5 or more mice at day 7 post infection (p. i.) and from all mice that were euthanized between days 7–9 p. i. Organs were evaluated for the presence of infectious virus particles.
CD4 and/or CD8 T cells were depleted by i.p. administration of monoclonal antibodies YTS191 (anti-CD4, BioXcell) and YTS169 (anti-CD8, BioXcell) on days -3 and -1 of infection. Isotype depletion control was done using the antibody LFT-2 (BioXcell). On each day, 300 μg of the respective antibody was administered. The efficiency of the depletion was verified on the day of infection (day 0) by flow cytometry and was > 98%. Specific depletion of CD11b+ and CD11c+ cells in DTR chimeric mice was accomplished by i.p injection of 0.2 μg of diphtheria toxin (DT) on days -1, 0, 1 and 2 of infection.
Organ samples were homogenized in 1 ml of DMEM with 2% FCS using Lysing Matrix D (MP Biomedical) in a beat mill. Infectious virus particles in cell culture supernatants, blood and organ samples were determined by immune focus assay using a monoclonal anti-LASV nucleoprotein antibody (2F1) for detection of infected foci as previously described [17].
For quantification of serum aminotransferases, serum samples were diluted 1:10 or higher in 0.9% NaCl and determined by using commercially available assays (Reflotron, Roche diagnostics). The parameters were measured for individual mice. The limit of detection for aspartate aminotransferase (AST) at 25°C was 2,25 U/l in undiluted samples. The normal range for mice was determined in 20 uninfected mice and was 40–60 U/l.
For flow cytometry experiments mice were euthanized on days 0, 4 or 7 p. i. and spleens and lungs were collected for analysis. Single cell suspensions were prepared by cutting tissues into small fragments followed by enzymatic digestion for 45 min at 37°C with Collagenase D (2mg/ml) (Roche) in RPMI-1640 medium. Tissue fragments were further disrupted by passage through a 70-mm nylon cell strainer (BD Biosciences). Red blood cells were lysed with BD Pharm Lysing Buffer (BD Biosciences). Fc receptors were blocked with CD16/CD32 Fc Block antibody (BD Biosciences) followed by staining with fluorochrome-conjugated antibodies. A FACS LSR II or LSR Fortessa instrument (BD Biosciences) was used for flow cytometry acquisition. Analysis of data was done with FlowJo Software (Treestar).
Mouse tissues were fixed in 4% formalin/PBS and were embedded in paraffin. Sections (2μm) were stained with hematoxylin-eosin (H/E) or processed for immunohistochemistry as follows: After dewaxing and inactivation of endogenous peroxidases (PBS/3% hydrogen peroxide), antibody specific antigen retrieval was performed. Sections were blocked (PBS/10% FCS) and afterwards incubated with primary antibodies for rat anti-mouse CD3 (T-cells; Serotec), rabbit anti-Iba1 (monocytes/macrophages; Wako Pure Chemical Industries), B220 (B-cells; eBioscience), iNOS (inducible nitric oxide synthase; Abcam), Ki67 (cell proliferation; Abcam), cleaved caspase 3 (apoptosis marker; R&D systems) or Ly6G (neutrophils; BD Bioscience) Bound primary antibodies were detected with anti-mouse, anti-rabbit or anti-rat Histofine Simlpe Stain MAX PO immune-enzxme polymere (Nichirei Biosciences) and stained with 3,3′-Diaminobenzidine (DAB) substrate using the ultraView Universal DAB Detection Kit Ventana). Tissues were counterstained with hematoxylin. For fluorescence double labeling, primary antibodies were visualized using species-specific Cy3- or Cy2-conjugated secondary antibodies (all from Jackson ImmunoResearch Laboratories Inc.) with DAPI (Sigma-Aldrich) as nuclear staining.
Mice were injected intravenously with 100 μl of Evans Blue (2% in NaCl). Thirty minutes post-injection mice were euthanized and perfused transcardially with 30 ml of NaCl (0.9%) containing 2 U/l heparine. Lung and liver were collected and homogenized in formamide. Evans blue was extracted in 3 ml formamide by overnight incubation at room temperature. Samples were inactivated by adding 4% formaldehyde and Evans blue absorption was measured at 650 nm.
Analysis of cytokine concentrations was done using Milliplex MAP Mouse CD8 T Cell Magnetic Bead Panel Premix 15 Plex (MCD8MAG48K-PX15) according to the manufacturer’s instructions. A Luminex 200 system (Millipore) was used for data acquisition.
Statistical analyses were done using Graphpad Prism 6 software. Differences in survival rate were analyzed using Fisher’s exact test. Differences in weight, temperature, AST activity or virus titer in blood on individual days were analyzed using Mann-Whitney non-parametric test. Differences in plasma cytokine levels among multiple groups were evaluated via non-parametric Kruskal-Wallis test followed by Dunn’s post-test. Differences in virus titers in organs were analyzed using multiple t tests. Differences in the cellularity of immune cells over time were analyzed using a Two-Way ANOVA followed by a Bonferroni’s post-test. Statistical significance was depicted as follows:
While immunocompetent C57BL/6 mice are resistant to LASV, IFNAR-/- mice of the same genetic background show LASV-associated morbidity, with transient weight loss and viremia [17]. To evaluate the effect of competent hematopoietic immunity on LASF pathogenesis, we generated mice in which IFN-I signaling deficiency was restricted to the radio-resistant cell compartment, namely, mostly cells of stromal origin. IFNAR-/- mice were irradiated and transplanted with bone marrow progenitor cells isolated from wt C57BL/6 donor mice (Fig 1A). To ensure the greatest number of donor-derived circulating T cells, recipient mice were irradiated with split maximum radiation doses. Experimental infections were performed 4–5 weeks after transplantation in mice with more than 85% of hematopoietic donor cell engraftment (S1A Fig). Under these conditions, it is estimated that less than 1% of circulating T lymphocytes are host-derived [21]. In these chimeric IFNAR-/- B6 mice, LASV infection was 100% lethal, and displayed indicators of severe disease such as high serum levels of aspartate aminotransferase (AST), viremia, loss of core temperature and rapid weight loss (Fig 1B).
Next, the effect of LASV infection was tested in the reverse chimeric mouse model, in which B6 mice were transplanted with IFNAR-/- bone marrow donor cells. Infection of chimeric B6 IFNAR-/- mice with LASV resulted in 75% lethality which was marked by significant weight loss in those animals who died, as well as high viremia, elevated levels of AST, viral replication in peripheral organs, and significant decrease of body temperature. Surviving mice were able to clear virus from the blood, gain weight and recover homeostatic control of the body temperature (Fig 1C).
To rule out effects associated with the generation of chimeric mice, the effect of LASV infection was assessed in wt mice transplanted with syngenic wt bone marrow (B6 B6) as well as IFNAR-/- mice transplanted with IFNAR-/- bone marrow (IFNAR-/- IFNAR-/-). All B6 B6 chimeric mice survived infection showing no signs of morbidity despite low levels of viral replication in peripheral organs (Fig 1E). However, the IFNAR-/- IFNAR-/- chimeras showed elevated viremia and high virus replication in all organs tested (Fig 1D). Strikingly, despite ongoing systemic infection and failure of clearing virus from the blood, IFNAR-/- IFNAR-/- chimeras showed only mild signs of disease with moderated weight loss and 80% survival. These results mimicked those described in knockout IFNAR-/- mice, indicating that chimeric mice infected with LASV reproduced the disease observed in non-transplanted mice with the same genotype. Of note, all chimeric mice lacking IFN-I signaling in any compartment showed similar levels of viremia and virus replication in peripheral organs (Fig 1B, 1C and 1D), which indicated that virus replication alone was not responsible for LASV-associated morbidity. Non-infected IFNAR-/- B6 and B6 IFNAR-/- control chimeras did not show any pathological features ruling out effects of transplantation in the observed phenotype (S1B Fig). Moreover, these findings could be reproduced with representative LASV strains from lineages II, III and IV, which ruled out effects associated with a particular LASV strain (S2 Fig).
LASV susceptibility in our model was not restricted to IFNAR-/- B6 chimeras since B6 IFNAR-/- mice were also highly susceptible to LASV infection with 75% lethality. We hypothesized that IFN-I competence in either the radio-resistant or the hematopoietic compartment was sufficient to provide pro-inflammatory cues that would result in disease-associated immunopathology. To test this hypothesis, the levels of serum pro-inflammatory cytokines were measured in all four chimeric mouse models at the peak of LASV infection. Significantly higher levels of TNF-α and IFN-γ were observed in both IFNAR-/- B6 and B6 IFNAR-/- chimeras in comparison with control chimeras (Fig 2A). Interestingly, T cell cytotoxicity markers such as FAS, FAS-L and granzyme B, as well as pro-inflammatory chemokines such as MIP-1β, were significantly upregulated in IFNAR-/- B6 chimeras in agreement with a more severe disease in this model and suggesting a putative role of T cells on LASV immunopathology. Further supporting immunopathology as a key component of disease, histopathological analysis of liver sections in all chimeric models showed evidence of lymphocytic infiltrates clustering in the liver of IFNAR-/- B6 and B6 IFNAR-/- mice, but not in control chimeras (Fig 2B). Taken together, our results indicated that chimeric mice with competent IFN-I signaling in either the radio-resistant or the hematopoietic compartment, reproduced main features of severe LASF and suggested an important involvement of host-mediated immunity in LASF pathogenesis.
Within the Arenaviridae family of viruses some members cause severe hemorrhagic fever in humans (e. g. LASV, Machupo, Junin, Lujo virus), while others are seemingly low- or non-pathogenic in humans (e. g. Mopeia, Mobala, Morogoro virus). To test whether our mouse model reflected the virulence associated to human pathogenic arenaviruses, the morbidity and mortality of LASV in the chimeric mice was compared with that of Morogoro virus (MORV), a Mastomys natalensis-borne non-pathogenic African arenavirus closely related to LASV [22]. In IFNAR-/- B6 mice MORV infection caused only mild weight loss and moderated increase of serum AST, with all mice fully recovering from infection (Fig 3A). Similar results were obtained in the IFNAR-/- IFNAR-/- chimeras with survival of 100% of infected mice (Fig 3C). In both B6 B6 and B6 IFNAR-/- chimeric mice, no signs of any morbidity after MORV infection were detected (Fig 3B and 3D). These results indicated that MORV was not pathogenic for any of the tested mouse models, and was effectively cleared from infected mice even in the total absence of IFN-I signaling. Furthermore, these results also indicated that the mortality associated to chimeric mice infected with LASV, reflected pathophysiological mechanisms associated to LASF and not other arenavirus-induced disease. This was in agreement with the differences between pathogenic and non-pathogenic arenaviruses in humans.
Since the chimeric IFNAR-/- B6 mouse model showed a high degree of susceptibility to LASV but was entirely resistant to MORV, we reasoned that this model was suitable to dissect pathophysiological mechanisms associated to LASF. To test this hypothesis IFNAR-/- B6 mice were infected with LASV or MORV and evaluated for tissue-specific pathological findings. Infected mice as well as mock-infected mice were euthanized at day 7 post-infection which coincided with the peak of LASV viremia (Fig 1B). Due to the important involvement of kidney and liver failure in human LASF [3], infected kidney and liver sections were compared with those from mock-infected mice. Immunostaining using anti-Iba1 antibody, a marker of activated monocytes and macrophages [8], showed infiltration of these inflammatory leukocytes in mice infected with either virus (Fig 4A and S3 Fig). Further analysis of liver sections revealed additional similarities between the pathology features caused by both viruses at the peak of viremia, with upregulation of T cell lymphocyte numbers, infiltration of granulocytes, moderated upregulation of inducible nitric oxide synthase (iNOS) and low overall proliferation (Fig 4A). Thus, it seemed that both viruses caused similar degree of inflammation in liver and kidneys, and suggested that more subtle differences in the host immune response were responsible for the radical differences observed in virus-associated morbidity and mortality.
Previous studies have shown that alterations of the lymphoid and non-lymphoid tissue cellularity due to infiltrating antigen-presenting cells and CD8 T cells, are related with the efficacy of the host antiviral response to a variety of viruses [23,24]. Thus, we next evaluated the cellularity of different leukocyte populations over the course of LASV and MORV infection in both lymphoid tissues (spleen) and peripheral tissues (lung). At late time points after infection (day 7), LASV-infected mice showed a significant increase in the frequency of CD8 T but not CD4 T cells compared to MORV-infected mice in both spleen and lungs (Fig 4B). In addition, a significant increase in the frequency of granulocytes, and inflammatory Ly6Chi monocytes was observed (Fig 4C and S4 Fig). These changes in cell frequencies were related with significantly higher virus titers in LASV-infected mice compared with their MORV-infected counterparts (Fig 4D). These results pointed out at CD8 T cells and inflammatory myeloid cells as putative mediators of LASV immunopathology and supported the idea that infiltrating inflammatory cells may support LASV replication.
A significant accumulation of Ly6Chi inflammatory monocytes was observed over the course of LASV infection (S4 Fig). This is a myeloid population with an important immunopathologic role in other viral infections such as pulmonary influenza [24,25]. Thus, we reasoned that specific depletion of monocytes could serve to alleviate LASV immunopathology. To test this, chimeric mice were generated in which irradiated IFNAR-/- recipients were transplanted with bone marrow progenitor cells from CD11b-diphtheria toxin (DT) receptor (DTR) transgenic donor mice, or CD11c-DTR donor mice. In these mice, administration of DT kills CD11b-expressing monocytes and macrophages, or CD11c-expressing dendritic cells respectively, thus allowing evaluation of the specific functions of these myeloid populations [26,27]. In chimeric IFNAR-/- CD11b-DTR and IFNAR-/- CD11c-DTR mice more than 95% of monocytes and dendritic cells were from donor origin and were depleted by intraperitoneal administration of DT (S5 Fig). Perhaps surprisingly, depletion of CD11b- and CD11c-expressing cells did not prevent death from LASV infection, nor did it prevent viremia and virus replication in peripheral organs (Fig 5A). These findings indicated that despite the infiltration of inflammatory myeloid cells observed in the organs of LASV-infected mice, these cells were not responsible for virus-associated immunopathology. However a reduction in the levels of cell damage (as indicated by reduced levels of serum AST) was observed and also viremia in CD11b-depleted mice compared with IFNAR-/- B6 chimeras (Figs 5A and 1). These results suggested that, at least to some extent, CD11b+ cells could play a role supporting viral replication and dissemination. To test this hypothesis, monocyte-derived dendritic cells (moDCs) were differentiated in vitro via incubation of bone marrow progenitor cells obtained from C57BL/6 mice with granulocyte-macrophage colony-stimulating factor (GM-CSF). In vitro-differentiated mo-DCs were productively infected with LASV strains belonging to various lineages, but were entirely refractory to MORV infection as well as infection with other non-pathogenic arenaviruses such as Mobala virus (MOBV) and Mopeia virus (MOPV) (Fig 5B). Together, our findings indicated that productive infection of inflammatory myeloid cells may play a role on systemic LASV dissemination or virus replication in peripheral tissues (see Fig 4D), but that the infiltration of these cells in peripheral tissues does not influence LASV immunopathology.
Next the role of T cells on LASV pathogenesis was examined. IFNAR-/- B6 chimeric mice were depleted of CD4 and CD8 T cells via i.p. injection of monoclonal antibodies (clones YTS191 and YTS169 respectively) starting three days before infection with LASV (S5B Fig). As controls, mice were injected with isotype antibodies or left untreated. As shown in Fig 6A, isotype and non-treated mice displayed features of severe LASF with high viremia, elevated AST, virus replication in peripheral organs and death within ten days post-infection. This was also the case for mice depleted of CD4 T cells, which indicated that these cells alone were not responsible for immunopathology (Fig 6B). Strikingly, depletion of CD8 T cells rescued 87,5% of mice from LASV-induced death, despite the fact that elimination of these cells resulted in persistent viremia (Fig 6C). Mouse survival was also associated with a marked reduction of cell damage, as indicated by low levels of serum AST. Finally, simultaneous depletion of CD4 and CD8 T cells resulted in 100% survival of chimeric mice infected with LASV (Fig 6D), which demonstrated a chief role of CD8 T cells on LASV-induced immunopathology, and also a putative contribution of CD4 T cells, albeit at lower levels. Of note, neither depletion of CD4 T cells nor CD8 T cells resulted in changes on the levels of virus replication in peripheral organs compared to non-depleted controls (Fig 6B–6D). These results indicated a key role of T cell-mediated immunopathology on LASF severity in our model, and a poor capacity of T cells to contribute to viral clearance.
A common feature of viral hemorrhagic fevers including LASF is vascular leakage which in the case of LASF has been associated to edema and hematological disfunction [3,15,28]. CD8 T cells have been strongly correlated with vascular leakage in other acute infection models such as cerebral malaria [29,30] and LCMV, in the latter, due to direct killing of infected endothelial cells [31]. Thus, we reasoned that CD8 T cell-mediated vascular leakage could explain mechanistically their association with disease severity in our model. To test this hypothesis we administered Evans Blue (EB), a dye with high affinity to serum albumin, to IFNAR-/- B6 mice at the peak of infection with LASV or MORV. Photometrical quantification of EB in the lung and liver of infected mice indicated significantly higher vascular leakage in LASV-infected mice compared with MORV-infected mice (Fig 7A). Strikingly, vascular leakage in peripheral organs of LASV-infected mice was entirely prevented via depletion of CD8 T cells (Fig 7A). In agreement with these findings being correlated with CD8 T cell-mediated cytotoxicity, the serum levels of FAS and FAS-L were significantly higher in LASV-infected mice compared to MORV-infected mice, and were significantly reduced after CD8 T cell depletion (Fig 7B). These results demonstrated that LASV lethality was associated to CD8 T cell mediated immunopathology which was mechanistically correlated with increased vascular permeability.
Previous studies have described several animal models of LASF including immunodeficient mice, guinea pigs and non-human primates (NHPs) [11,32]. Of these models, only NHPs displayed immunocompetence as well as human features of LASF (i. e. liver tropism) [11,13]. However, due to the limited availability, lack of reagents, and high costs associated to NHP research, this model is not optimal for the dissection of the physiological mechanisms involved in LASF immunity and pathogenesis.
Here we present a chimeric mouse model generated by bone marrow transplantation, in which restriction of IFN-I deficiency to either the radio-resistant or the hematopoietic compartment resulted in high susceptibility to LASV. These results are in agreement with the role of IFN-I-induced antiviral state in restricting early LASV infection [18,33], and suggest that viral replication in discrete IFN-I-deficient cell compartments is sufficient to establish disease. Moreover, our study also suggested that not only dendritic cells and monocyte-derived macrophages of hematopoietic origin, but also other radio-resistant or stromal cell types may be susceptible to LASV infection in vivo. It is worth noting that in our chimeric mouse model, putative LASV target cells such as tissue-resident macrophages, alveolar macrophages, Kupffer cells and Langerhans cells, [8,9] are host-derived due to the fact that these cells are originated from yolk sac, and not hematopoietic progenitors [34]. The importance of early replication of LASV in susceptible cells for the establishment of infection, is also supported by the finding that LASV but not non-pathogenic arenaviruses such as MORV, MOBV and MOPV, was able to replicate in mouse monocyte-derived dendritic cells in vitro.
Interestingly, despite the fact that LASV but not MORV was lethal for chimeric mice, histopathological studies revealed moderated liver damage–a prominent feature of LASV infection in NHPs and in humans [35]–in either infection. A possible explanation for this observation is that lack of IFN-I signaling in hematopoietic or radio-resistant cells may reduce IFN-I-mediated oxidative damage of hepatocytes [36]. This is also consistent with the finding that STAT-1 mice infected with LASV also showed lower levels of circulating serum aminotransferases compared to NHPs and humans [18]. As previously reported in other animal models as well as in humans, LASV infection resulted in moderated inflammation in several peripheral organs, with infiltration of inflammatory monocytes/macrophages, T cells and granulocytes. Similar findings were found in chimeric mice infected with MORV. However, specific depletion of antigen-presenting cells such as monocytes and dendritic cells in LASV-infected mice, only caused moderated reduction of viremia, and did not alleviate LASV-induced disease. Conversely, depletion of CD8 T cells was sufficient to alleviate LASV-induced morbidity and mortality in IFNAR-/- B6 mice. The importance of T cells as drivers of LASV-mediated immunopathology in our model does not reconcile easily with the finding that T cell activation is necessary for early control of LASV replication in NHPs [13]. However, our previous studies utilizing mice expressing human HLA also demonstrated that T cells were required for early viral clearance but if they failed to do so–due for example to high viral loads–they worsened the ensuing disease process [8,13]. Similar paradoxical findings indicating both detrimental and protective effects of T cells have been found in many models of viral infection including influenza virus [37], dengue virus [38], LCMV [31] and hantavirus [39]. These findings suggest that defects in T cell homeostasis mechanisms modulating the balance between protective and immunopathogenic T cell responses may play an important role in many viral infections. Thus, perhaps, the differences found in our model and NHPs only reflect partial observations of a complex physiological process. Further experiments utilizing different doses and routes of infection in our model as well as T cell depletion assays in NHPs may help to further evaluate similarities between our model and the NHP model. Of note, simultaneous depletion of CD8 and CD4 T cells further alleviated LASV-induced disease, a finding that may reflect CD4-mediated cytotoxicity as described for other acute viral infections [40], or perhaps the important role of Foxp3+ regulatory CD4 T cells (Tregs) at preventing arenavirus-mediated immunopathology [41].
In our model, lethality was not restricted to IFNAR-/- B6 chimeras, and B6 IFNAR-/- mice were also highly susceptible to LASV infection with 75% lethality. We showed that IFN-I competence in either compartment was correlated with increased levels of serum pro-inflammatory mediators in agreement with IFN-I as a key inducer of antiviral pro-inflammatory signals [15,42,43]. A number of previous reports have identified innate or bystander CD8 T cell activation via cytokine-mediated, TCR-independent stimulation, which results in rapid production of IFN-γ and TNF-α as well as cytotoxicity [44–46]. Indeed, we observed significantly higher levels of serum IFN-γ and TNF-α at the peak of LASV infection in IFNAR-/- B6 and B6 IFNAR-/- mice compared to control chimeras. Although preliminary, these results strongly suggest that cytokine-mediated CD8 T cell activation may be an important mechanism of LASF pathogenesis and warrants further investigations. Additionally, the levels of FAS, FAS-L and granzyme B were significantly upregulated in IFNAR-/- B6 mice which suggests that T cell-mediated cytotoxicity may account, at least to some extent, for the increased lethality in this model.
Previously, mice lacking functional STAT-1 were shown to reproduce features of human LASF also present in our model such as systemic virus dissemination, elevated serum aminotransferases and hypothermia during the late stages of disease [18]. Despite lacking functional IFN-I signaling, these mice were highly susceptible to LASV and developed severe disease and post-recovery sequelae such as hearing loss which is commonly found in LASF survivors [47]. Interestingly, the severity of infection was suggested to be dependent in this model at least partially to loss of STAT-1-dependent regulation of T cell homeostasis [48,49] which resulted in T cell infiltration and damage to the cochlear nerve [47]. These findings further strengthen the notion that T cell-mediated immunopathology is an important pathophysiological feature of LASF.
As CD8 T cell-mediated cytotoxicity has been directly linked to vascular leakage [31], we investigated the correlation between CD8 T cell immunity and edema, a prominent feature of human LASF and other arenaviral hemorrhagic fevers [3,50,51]. We demonstrated that LASV but not MORV-infected mice suffered vascular leakage leading to edema in lungs and liver, and that this phenotype was entirely alleviated by CD8 T cell depletion. These results strongly suggest association between CD8 T cell immunopathology, vascular leakage and death, and reflect important differences in our model between infection with non-pathogenic and pathogenic arenaviruses.
Our system provides a platform for testing arenavirus pathogenicity and strongly suggests that T cell-mediated immunopathology plays an important role in LASF pathophysiology. However, whether or not our findings reflect human disease remains to be determined. Unfortunately, to date there is little insight into the pathophysiology mechanism of human LASF which requires adequate clinical studies in the field. Elucidation of the role of T cells in disease severity in humans could provide a rationale for immunotherapeutic interventions against LASF.
|
10.1371/journal.ppat.1006073 | When Viruses Don’t Go Viral: The Importance of Host Phylogeographic Structure in the Spatial Spread of Arenaviruses | Many emerging infections are RNA virus spillovers from animal reservoirs. Reservoir identification is necessary for predicting the geographic extent of infection risk, but rarely are taxonomic levels below the animal species considered as reservoir, and only key circumstances in nature and methodology allow intrinsic virus-host associations to be distinguished from simple geographic (co-)isolation. We sampled and genetically characterized in detail a contact zone of two subtaxa of the rodent Mastomys natalensis in Tanzania. We find two distinct arenaviruses, Gairo and Morogoro virus, each spatially confined to a single M. natalensis subtaxon, only co-occurring at the contact zone’s centre. Inter-subtaxon hybridization at this centre and a continuum of quality habitat for M. natalensis show that both viruses have the ecological opportunity to spread into the other substaxon’s range, but do not, strongly suggesting host-intrinsic barriers. Such barriers could explain why human cases of another M. natalensis-borne arenavirus, Lassa virus, are limited to West Africa.
| Reservoirs of zoonotic viruses are usually equated with a particular wildlife species. It is rarely assessed whether genetic groups below the species level may instead represent the actual reservoir, though this would have major implications on estimations of the zoonosis’ spatial distribution. Here we investigate whether geographically and genetically distinct subtaxa of the widespread African rodent Mastomys natalensis carry distinct arenaviruses, by sampling in detail across a contact zone of two of these subtaxa. Ongoing hybridization shows that individuals of the subtaxa are in direct physical contact, in principle allowing viral exchange, yet neither of the two arenaviruses -Gairo and Morogoro virus- were found to have crossed the zone. Such intraspecific genetic barriers to arenavirus spatial spread have important implications for our understanding of the related Lassa arenavirus, a pathogen potentially lethal to humans of which Mastomys natalensis is also the main reservoir. Although Lassa virus appears to infect several secondary hosts, its distribution is restricted to West Africa and matches that of another M. natalensis subtaxon. Our data thus indicates that it is because of M. natalensis intraspecific distinctions that the human Lassa fever endemic area has not expanded to the rest of sub-Saharan Africa.
| Most emerging RNA virus infections originate from wild animals [1]. Fortunately, outbreaks of many such infections are geographically restricted, e.g. MERS coronavirus in the Middle East, Marburg filovirus in central and southern Africa, and Nipah henipavirus in south-east Asia. This restriction is most likely due to dependence on particular host reservoir species (single or multiple) for persistence in nature–those hosts themselves having restricted distributions. Identification of the reservoir host and its geographic range are therefore essential for informed public health responses [2], dramatically illustrated by the 2014 outbreak of Zaire-Ebola virus that unexpectedly emerged in West Africa [3].
Repeated detection of a particular virus in a particular animal species, and not in other species in sympatry, usually implicates that species as the main reservoir. However, species with a wide geographic range are often (cryptically) genetically subdivided into subtaxa, yet it is rarely assessed whether a local intraspecific taxon may represent the reservoir instead of the entire species. Such associations between intraspecific animal taxa and particular viral taxa could explain why the distribution of some viruses appears smaller than the range of the reservoir species, for example in the case of distinct hantaviruses of the widespread rodent species Peromyscus leucopus [4, 5], P. maniculatus [4, 6, 7] and Oligoryzomys flavescens [8], and the Simian Immunodeficiency viruses (SIV) and Simian Foamy viruses (SFV) of chimpanzees (Pan troglodytes) [9–12]. However, it is difficult to corroborate these associations. Experimental infections require housing individuals of the particular subtaxa in biosafety laboratories, and will in any case reflect capacity of the virus to infect the host, rather than whether the host has a reservoir status in nature: whether an infection may be persistently transmitted in a particular host population depends on more factors than the ability to propagate in a host body after manual inoculation. For example, the route and timing of viral shedding in concordance with the host’s population dynamics may be important determinants of the infection’s invasion and persistence probabilities in a host population [13, 14]. Inferring subtaxon-virus associations from observations in nature may be more appropriate, but when spatial gaps or coinciding geographic barriers occur between sampling points of the distinct subtaxa, host-extrinsic factors such as isolation-by-distance or host movement barriers are indistinguishable as explanations of the spatial separation of viruses. Therefore, the association between virus and host taxa must be evaluated in areas where distinct host (sub)taxa carrying distinct viral taxa are in direct physical contact.
This situation can be found in secondary contact zones. These are formed when vicariant subtaxa that had allopatrically diverged in the past, re-expand into secondary contact during favourable environmental conditions [15]. Commonly, this contact results in the production of fertile hybrids across a delineated and stable hybrid zone [16]. These limited zones are often maintained by a balance between dispersal and (endogenous) selection against hybrids, so that distinctive genepools co-exist in the face of gene flow [17]. For example, using fine scale sampling across the European house mouse hybrid zone, it was demonstrated that strains of Murine cytomegalovirus and of the protozoan Cryptosporidium tyzzeri are each associated with a distinct Mus musculus subspecies [18, 19]. For RNA viruses, it might be expected such secondary contact zones provide the optimal ecological conditions for an evolutionary host shift to the closely related taxon across the zone, as even for host-specific RNA viruses their high mutation rates might ensure rapid adaptation to the exposed novel host. However, this has not yet been evaluated in nature. Host shift potential of RNA viruses has previously mainly been studied by comparing genealogical histories of virus phylogenetic groups and of their corresponding hosts [20–26]. This has allowed identification of e.g. phylogenetic distance between host taxa as an important constraint for a host shift; yet it remains unclear whether such a constraint may last when closely related subtaxa carrying different RNA viruses physically meet, for example at a secondary contact zone. Here, we characterize a secondary contact zone of subtaxa of the African rodent Mastomys natalensis to better understand host-imposed constraints to the distribution patterns of the rodent’s arenaviruses.
Arenaviruses are bi-segmented RNA viruses and those of the genus Mammarenavirus are typically hosted by rodents. Only a few can successfully infect humans, and while human-to-human transmission is possible, it has so far never resulted in a sustained epidemic [27, 28]. Lassa mammarenavirus (LASV) may cause a severe haemorrhagic fever in humans, and with about 200,000 cases and 3,000 deaths annually [29] it has a major public health impact [30]. LASV’s main natural reservoir host species is the Natal multimammate mouse Mastomys natalensis [31–33], although recently LASV and LASV-related strains have also been detected in other rodent species [34]. While this common rodent occurs throughout most of sub-Saharan Africa, LASV and Lassa fever in humans is restricted to West Africa and has never been detected east of Nigeria (Fig 1). Instead, five other arenaviruses have so far been detected from M. natalensis in various other regions (Fig 1): Mopeia virus (MOPV) in Mozambique [35], Morogoro virus (MORV–a strain of MOPV) in Tanzania [36], Luna virus (LUV) in Zambia [37], Gairo virus (GAIV) in Tanzania [38] and recently an unnamed Mobala-like virus in east-Nigeria [39]. These have never been detected in humans.
Mammarenaviruses are in general considered rodent-host specific. The majority has only been detected in a single rodent species, and on several occasions distinct arenaviruses have been found in distinct rodent species captured at the same sites [45–49]. On the other hand, five arenaviruses have been detected in multiple, yet closely related, rodent species [34, 50–53], suggesting that the level of host specificity may vary between arenaviruses.
M. natalensis is one of the most widespread and common mammals in sub-Saharan Africa; it occurs in all terrestrial habitats apart from dense forests and (semi-) deserts [54, 55]. Its mitochondrial DNA can be divided into six matrilineages that differ up to 3.8% at the cytochrome b gene and that are each geographically confined to distinct regions [43]. When superimposing the distributions of these matrilineages and M. natalensis-borne arenaviruses, each arenavirus seems to be restricted to the range of a single matrilineage, the Lassa fever endemic area roughly matching the range of M. natalensis matrilineage A-I (Fig 1). This pattern suggests that host intraspecific structure, as approximated by the matrilineal pattern, may constrain the geographic ranges of arenaviruses including LASV. Recently, Olayemi et al. (2016) concluded this is not the case, as they detected LASV in three M. natalensis individuals carrying the A-II mitochondrial lineage in eastern Nigeria. However, no nuclear markers were typed, and the A-II mitochondrial lineage was observed in less than 25% of the animals in the two localities where these LASV positive animals were detected, located at the edge of A-I matrilineage distribution. Moreover, a different and new arenavirus was found in an eastern locality across the river Niger where the A-II mitochondrial lineage was found in all individuals. Therefore, it remains unclear whether the observed pattern was due to LASV dispersal into the range of the host taxon associated with the A-II matrilineage, or due to A-II mitochondrial introgression into the range of the host taxon carrying LASV. Indeed, introgression of mitochondrial lineages into other taxa is a common phenomenon, therefore the spatial distribution of mitochondrial lineages often does not closely match that of the (multi-locus inferred) taxa themselves [56].
In central Tanzania, M. natalensis–borne GAIV and MORV are known to occur in close proximity [36, 38], and the subtaxa potentially represented by M. natalensis matrilineages B-IV and B-V are estimated to be in secondary contact in that region [43]. We previously found within the B-V matrilineage range that nuclear markers are further substructured in relation to an urban-rural contrast, and that the region varies in landscape features that likely translate into spatially varying M. natalensis densities [57]. Such areas of low host densities and/or inter-host contacts might in themselves present a sufficient barrier for virus transmission [58], and should therefore be taken into account when distinguishing host-intrinsic and host-extrinsic factors of viral distribution patterns.
In this study, we sampled M. natalensis at a fine scale across the spatial transition zone between GAIV and MORV, multilocus genotyped hosts and their arenaviruses, and estimated landscape connectivity between localities. In the context of this natural laboratory, we are able to discern the contributions of host genetic structure and landscape features on the spatial distribution of arenaviruses.
We sampled small mammals at twelve localities along the road between Dar es Salaam and Dodoma, spaced approximately 20 km apart and spanning the distribution range boundary of the B-IV and B-V M. natalensis matrilineages [43] and Gairo and Morogoro arenaviruses (Fig 1). Part of the sampling overlaps with data presented in [38] and [57]; see Table 1. At each locality small mammals were captured in Sherman live traps baited with a mixture of peanut butter and maize flour. Traps were set in a 1 ha square grid of 10x10 traps in fallow lands. At each locality minimum two grids were constructed minimum 500 m, maximum 2.5 km apart. If trapping success was low after two nights, we set additional grids. For each sampled grid maximum 20 M. natalensis individuals were euthanized by Isoflurane inhalation.
Blood was drawn either from the retro-orbital sinus or the punctured heart with a capillary tube and preserved on pre-punched filter papers (Serobuvard, LDA 22, 106 Zoopole, France), and organ samples were preserved in RNAlater and ethanol. RNAlater samples were kept at 4°C for maximum six weeks prior to storage at -80°C. When more than 20 M. natalensis were captured in a grid, supernumerary animals were sedated through Isoflurane inhalation, blood and toe-clips sampled on filter paper and in ethanol, respectively, and each was released at point of capture. Molecular screening of arenaviruses was augmented with six dried-blood sample collections from previously published rodent-trapping work [57, 59–61] (see details in Table 1). M. natalensis genotyping for microsatellite and cytochrome b markers (see below) was augmented using three of these additional collections (see Table 1).
All animal work was approved by the University of Antwerp Ethical Committee for Animal Experimentation (2011–52), and followed regulations of the Research Policy of Sokoine University of Agriculture as stipulated in the “Code of Conduct for Research Ethics” (Revised version of 2012). Euthanasia of small mammals was performed using an overdose of Isoflurane or via cervical dislocation.
DNA was extracted from toe or liver samples using the DNeasy Blood & Tissue Kit (Qiagen). Fifteen microsatellite loci [62] were genotyped as described in [57]. However, only those samples for which more than 10 loci were successfully amplified were considered for downstream analyses. Parts of cytochrome b (on the maternally inherited mitochondrion) and smcy (on the paternally inherited Y chromosome) were amplified in PCRs and Sanger sequenced in one direction. See further PCR details in in S1 Text.
We analysed the population genetic structure of M. natalensis microsatellite genotypes using the Bayesian clustering algorithm implemented in the program STRUCTURE v2.3.2. [63, 64], using the same settings as described in [57]. In brief, genetic clusters are sought in which deviation from genetic disequilibria are minimised, with proportions of each microsatellite genotype assigned to each of K clusters. The analysis was replicated 25 times for each K value, allowing for admixture and using a prior on shared sampling location (at the locality-level). Modes in STRUCTURE outputs were distinguished using CLUMPAK; similar level-K replicates are placed in the same mode, within-mode cluster labels are standardized, and assignments across modes and K levels calculated [65]. Similarity between clusters at different K-levels and modes was assessed by eye and for visual clarity given the same colour.
Cytochrome b sequences were aligned and compared to published M. natalensis sequences by constructing a Maximum Likelihood phylogenetic tree in RAxML (GTR substitution model, gamma rate variation, 1000 bootstraps) [66]. Each sequence was then assigned to lineage B-IV or lineage B-V as described in [43]. Smcy sequences were aligned in Geneious 6.1 using the Geneious alignment algorithm with a 5.0/-9.203 match/mismatch cost model.
Arenavirus RNA was screened in RNA extracted from dried blood samples (pooled by two) using two independent one-step reverse transcription-PCRs (RT-PCRs) targeting the same 340 nucleotide (nt) portion of the RNA-dependent RNA polymerase gene (L segment), but with different primers with different target affinities (see details in S1 Text). For pools positive for this viral gene (and a subset of 347 negative samples), additional RNA was extracted from individual kidney biopsies preserved in RNAlater using the Nucleospin RNA II kit (Macherey-Nagel) when available. From these RNA extract parts of the GPC gene (979 nt or 234 nt) and NP gene (558 nt or 450 nt) were amplified. All amplicons were Sanger sequenced in both directions. See S1 Text for further details on these assays.
We aligned our L, NP and GPC sequences with arenavirus sequences available in GenBank (all from rodents, except Lujo virus) in Geneious 6.1 based on the translated amino acid sequences (Blosum62 cost matrix). We removed the short non-coding parts, and constructed the phylogenetic trees of each partial gene sequence in MrBayes, (GTR substitution model, gamma rate variation: 6 categories, rate parameters estimated separately for each codon). Since we were only interested in topology and not in dating nodes, we minimised parameters by using a uniformly distributed strict clock prior on branch lengths. We let 4 MCMC chains run for 1 million iterations after the standard deviation of the split frequency reached 0.01. The replicate analyses without assuming a clock model (unconstrained branch lengths using an exponential prior probability distribution) did not significantly differ in likelihood or topology (S3 Fig).
As a quantitative measure of the potential for M. natalensis to move between localities, i.e. a measure of environmental barriers, we estimated landscape resistance pairwise between localities of the transect (Fig 1, localities A to L, and excluding all other sampled localities), using the methodology of [57] but applied over a larger geographic extent. In brief, ten field experts in M. natalensis ecology translated landscape elements of Tanzania’s land cover layer of 1997 [67] to exponentially increasing categories of M. natalensis habitat quality. After integrating linear landscape elements into this habitat quality layer (rivers and roads of three different width categories), the resulting expert opinion layer was modelled using Circuitscape [68] as a conductive surface from which resistance values between pairs of polygons (minimum polygons drawn around sampling sites in each locality) are calculated, in analogy with circuit theory.
The centroids of each locality and the earth (great-circle) distance between them were calculated in the R package ‘fields’ [69]. We based the mean genetic distance between arenavirus samples of each locality on a concatenation of the three arenavirus gene sequences (to a total of 1848 nucleotides) and then calculated this mean distance between localities in MEGA 5.2 (Tamura 3-parameter model) [70].
The correlation between the genetic distance matrices and the landscape resistance distance matrix or earth distance matrix was calculated using simple Mantel tests in the R package ecodist [71] (1,000,000 permutations for bootstrapping). Partial Mantel tests of the same package were used to correlate the genetic distance matrices to the landscape resistance distance matrix while ‘partialling out’ the influence of earth distance between localities.
Sampling coordinates were transformed to a flat surface using gnomonic projection. The centroids of the twelve transect localities (Fig 1, A to L) were then orthogonally projected onto their regression line to form a one-dimensional transect. Narrow clines are robust to such mapping details [72]. For each locality, frequencies of the M. natalensis matri- (cytochrome b) and patri- (smcy flank) lineages and the average assignment to either of the microsatellite clusters in the integrated Q matrix of the K = 2 STRUCTURE scenario were tabulated. Numbers of MORV and GAIV in the total arenavirus infected (RT-PCR positive) animals were tabulated per locality. We fitted clines to these observations using the software Analyse [73].
To evaluate whether by chance we sampled more related animals in some localities than others, we calculated Li’s relationship coefficient r [74] between pairs of host genotypes within each locality in SPAGeDi [75]. See details in supplementary Methods in S1 Text.
Genetic sequences generated in this study are deposited in GenBank:
Original data deposited in the Dryad repository: http://dx.doi.org/10.5061/dryad.5n00k [76]:
Of 1,289 M. natalensis individuals sampled in 12 localities, a random subset was genotyped, as well as 39 additional M. natalensis individuals from one southern and two northern localities in Tanzania (“wide-scale localities”; Fig 1, Table 1). We identified two M. natalensis cytochrome b matrilineages (B-IV and B-V sensu Colangelo et al. (2013) [43]; S1 Fig), as well as a bi-allelic SNP in the Y chromosome smcy-gene intron, suggesting two distinct patrilineages (later referred as A and T lineages) (Table 1). When allowing for two microsatellite clusters (K = 2) in STRUCTURE, all replicate runs (25/25) converged to the same spatial pattern of genetic structure (Fig 2), strongly supporting a division of our sample into two replicable genetic disequilibrium-minimising groups. Outwith the transect, in northern localities Shinyanga and Itigi all M. natalensis carried the mitochondrial B-IV lineage and belonged to microsatellite cluster 1 (yellow), while in the southern locality Lihale all mice carried B-V mitochondrial lineage and belonged to microsatellite cluster 2 (blue) (Table 1, Fig 2).
Along the sampled transect the proportions of the two matrilineages, patrilineages and microsatellite cluster memberships changes sharply between localities B and E (Fig 3, Table 1). In locality C, Berega, both maternal and paternal lineages are present and autosomal hybrid genotypes dominate (Table 1, Fig 2), indicating ongoing hybridization between two M. natalensis subtaxa. The clines for all three M. natalensis genomic compartments (mitochondrion, Y chromosome and autosomes) have narrow confidence intervals and very similar estimated cline centre positions (near locality C) and cline widths (Fig 3). The consistency across genomic compartments and the relatively narrow estimated cline widths (20.0, 21.0 and 21.6 km, respectively–Fig 3) indicate a multilocus barrier to gene flow between the two M. natalensis subtaxa. Together with the observations from the wide-scale localities, we can conclude M. natalensis matrilineages B-IV and B-V correspond with genome-wide genetic structure in this region. We will therefore subsequently refer to these genome-wide clusters as M. natalensis subtaxa B-IV and B-V.
The B-IV mitochondrial lineage, but not the smcy A allele, is observed at low frequencies throughout the sampled transect localities in B-V’s range (localities E-L, up to 140 km from the estimated clines centres), indicating wide scale but low level mitochondrial introgression (Table 1). Low level introgression of both B-V mitochondrial lineage and smcy T allele was also observed in the sampled transect localities in the B-IV subtaxon range (localities A and B), but this range was only sampled up to 40 km from the estimated clines centers.
For STRUCTURE K>2, the two subtaxa’s microsatellite clusters are further hierarchically substructured with consistent spatial pattern (Fig 2). At K = 3 (major modes) or K = 4 (minor modes), animals from locality B (Chakwale) form a sub-cluster embedded within subtaxon B-IV. This substructure may be due to significantly higher levels of relatedness within this locality than in others (S1 Text and S2 Fig). At K = 3 (minor modes) or K = 4 (major modes), animals from locality H (Morogoro) also form a sub-cluster embedded within subtaxon B-V, consistent with previous inference over a subset of this dataset [57]. These animals do not show high relative relatedness patterns (S1 Text and S2 Fig). At K = 5 (minor modes) and K = 6 (major modes), a subset of the animals at the southern locality Lihale are consistently distinguished. These animals are significantly more related to each other than animals in other localities (S1 Text and S2 Fig).
A total of 53 arenavirus positive samples were found in 1,167 dried blood samples (DBS) from the transect-localities and in 392 blood samples from additional collections (Table 1). A further 6 out of 347 kidney samples tested (from individuals with negative DBS) were arenavirus RT-PCR positive. Phylogenetic reconstruction of parts of the L, GPC and NP genes showed positive samples contained the arenaviruses MORV and GAIV (Fig 4).
Bayesian phylogenetic trees based on the partial L (Fig 4A), GPC (Fig 4B) and NP gene (Fig 4C) showed four clades for MORV that were each specific to a single or set of adjacent localities (Fig 1). One clade (MORV-III) was however not well supported by the phylogenetic tree based on NP sequences (Fig 4C). MORV-III monophyly was also not supported by phylogenetic reconstruction without branch length constraints (S3 Fig). The remaining topology was similar under clock and non-clock models (S3 Fig).
Along the transect GAIV was only detected in majority B-IV host subtaxon localities and MORV only where the B-V host subtaxon was in the majority (Table 1, Figs 1, 3 and 4). The only exception was the hybrid-host-rich locality C, Berega, where both GAIV and MORV were detected at low prevalence (1.5% and 0.2%, respectively; Table 1). GAIV was also found in Mbulu and Shinyanga in north Tanzania, consistent with it being present across the range of the northern B-IV subtaxon (Fig 1).
The spatial frequency clines of the two arenaviruses thus coincide with their host’s genotypic clines, with estimated virus cline centre and width falling within the 95% confidence intervals of those of the host clines (Fig 3), although virus cline confidence intervals are much broader due the lower number of virus observations. The association between viral type and host taxa on either side of the cline centre is highly significant (χ2 = 42, df = 1, p = 9.1 x 10−11).
Both virus and host transition zones centre around locality C (Fig 3). Pairwise landscape resistance varies across the localities within the zone centre’s confidence intervals (B-C-D-E-F, 13.6, 12.2, 30.6, 7.9, Fig 3, S1 Table) with maximum between D-E. Across the transect as a whole, J-K-L comparisons have higher resistance estimates than D-E. Neither of these regions of high host-movement-resistance are likely to match the true centre of the GAIV-MORV transition zone nor appear to genetically structure the arenavirus associated with the B-V host subtaxon: MORV occurs on both sides of D-E and the clade MORV-II occurs in localities J as well as K. A partial mantel test correlating mean arenavirus (GAIV+MORV) genetic distance between pairwise localities to the landscape resistance, while partialing out the earth distance matrix, was not significant (R2 = -0.13, p = 0.95). Mantel tests (both simple and partial) lead to a high number of false-positive correlations [78], making this lack of correlation more striking. To summarise: away from the host contact zone landscape resistance is stronger than that within the zone, and even at its strongest, does not correlate with intraspecific arenavirus genetic substructure.
In contrast to the large-scale geographic association between arenavirus species and M. natalensis subtaxa, there was no geographic match between population genetic structure of M. natalensis within the B-V taxon (Fig 2) and the sublineages of MORV (Fig 1 and Fig 4). M. natalensis-B-V from locality H (Morogoro) show genetic separation from surrounding localities, but their MORV strains are of the same sublineage as those of adjacent localities G and I (Fig 1, Fig 2, Fig 4). M. natalensis from all other B-V localities belong to the same population genetic cluster, while carrying four distinct MORV sublineages (Fig 1, Fig 2 and Fig 4). A partial mantel test correlating mean MORV genetic distance between pairwise localities (excluding GAIV specific localities A and B) to the landscape resistance, while partialling out earth distance was significant (R2 = 0.02, p = 0.001) but weaker than a simple mantel test correlating mean genetic distance to earth distance (R2 = 0.38, p<0.001), indicating isolation-by-distance. GAIV’s range was not sampled broadly enough for equivalent analyses in relation to structure within M. natalensis-B-IV subtaxon.
Our results demonstrate that the Mastomys natalensis arenavirus system in central Tanzania does indeed form a natural laboratory suited for distinguishing host-intrinsic and environmental factors limiting the spread of arenaviruses. Multilocus genotyping shows a narrow contact zone between two subtaxa of M. natalensis. The ranges of these subtaxa are largely consistent with previously described matrilineages, though low frequency widespread mitochondrial introgression is also observed. The M. natalensis subtaxa each carry a different arenavirus right into their contact zone, where clustering of nuclear genotypes reveals ongoing hybridization. This shows that members of the subtaxa are in direct individual-to-individual contact, yet neither virus is found to have crossed the zone. Landscape analyses furthermore show the largest indicators of barriers to host movement or troughs in host density in our study area lie outside the zone and do not correlate with interspecific or even intraspecific viral genetic substructure. However, a number of alternative hypotheses for the observed M. natalensis B-IV—GAIV and B-V—MORV associations remain to be discussed.
1) The host taxa and their viruses could have only recently arrived at their current boundaries, the arenaviruses lacking the time to spread through new host taxon’s range. Using multiple fossil calibration points, the matrilineages of the subtaxa were previously estimated to have diverged allopatrically about 1.4 million years ago, during humid climate that could have encouraged dense rainforests (a M. natalensis barrier) to surround isolated patches of savannah (suitable M. natalensis habitat) [43]. Since then the African climate has experienced many fluctuations, but after the last African Humid Period between 14,800 and 5,500 years ago, during which large parts of East Africa would have been forested, climate has not been very different from its current state [79]. It therefore seems reasonable to suggest the M. natalensis subtaxa could have come into contact any time during at least the last 5,500 years, making it unlikely the time window in which the contact zone was formed coincided with our sampling.
2) Arenaviruses of M. natalensis appear to cause mainly acute infections with subsequently life long presence of antibodies [38, 80–82]; and antibodies are widely cross-reactive between several Old-World mammarenaviruses [83, 84], which could lead to cross-immunity [85]. Such cross-immunity on first inspection might appear to have a potential role blocking viral transmission across a contact zone of two host taxa with different arenaviruses. However, on closer inspection, this appears highly unlikely for our study system. Frequency of immune individuals on either side of the host contact is clearly below any threshold allowing viral spread and persistence. Complete cross-immunity would mean a spreading virus would encounter the same frequency of immune individuals as the virus already present–which by definition is below any threshold allowing viral spread and persistence. Cross-immunity is not, therefore, the factor blocking viral spread.
3) The available landscape layers for Tanzania might not cover all possible environmental barriers for M. natalensis-borne arenavirus dispersal. In particular the host/arenavirus clines roughly correlate with the position of an altitudinal cline (S4 Fig). Local precipitation patterns vary along the altitudinal cline (S4 Fig) and are determinants of onset and duration of the M. natalensis breeding period [86]. Nevertheless, the duration and peaks of rainfall still largely overlap between localities north (Mbulu), within (Berega), and south (Morogoro) of the contact zone (S4 Fig), and M. natalensis directly monitored in Berega and Morogoro indicate that breeding seasons are not timed very differently on either side of the contact zone: while in some years M. natalensis’ breeding onset can differ north-south by 3 months, in others it is well synchronized [60, 86]. While breeding asynchrony may, in the case of cross-immunity, cause a delay in virus transmission from a high-density population (where transmission rates could be increasing due to recruitment of susceptible juveniles) to a low-density population (which consists mostly of adults that are more likely to have cross-reactive antibodies [38, 80, 81]), this delay would only be temporary and would certainly not be able to explain a long-term geographic separation of the two arenaviruses.
Therefore, our findings strongly suggest spread of the two arenaviruses in question across the M. natalensis subtaxa contact zone is blocked by host-intrinsic rather than geographic/environmental factors. We thus directly assess how two arenaviruses fail to successfully emerge into closely related host taxa, despite optimal ecological exposure conditions, highlighting the strong limits to RNA virus adaptive flexibility that host-imposed constraints may enforce. These constraints might not be present at lower levels of host differentiation than the two subtaxa: we did not observe a spatial match between intraspecific genetic structure of MORV and population genetic structure within M. natalensis B-V, where an urban population has only recently differentiated from rural mice [57].
These insights highlight how our understanding of zoonotic virus epidemiology could strongly benefit from an investment in molecular taxonomic research of wildlife reservoir hosts, and that the intraspecific level deserves more attention. Small mammals such as rodents and bats have been indicated as a major source of viral zoonoses [87], but much of the diversity that these animals harbour is likely to be cryptic and understudied at the genetic or phylogeographic level. For example, only 66% of the 2277 rodent species currently recognized worldwide have more than three DNA sequences deposited in GenBank (accessed 30/01/2016). We therefore argue that thorough taxonomic and phylogenetic investigations of reservoir hosts should go hand in hand with zoonoses surveillance programs.
We show evidence suggesting host-intrinsic factors determine the spatial distributions of two arenaviruses. This paves the way for experimental studies investigating what such factors might be. Candidate factors should include a wide range of possibilities from direct host immune defense differences to indirect effects, for example differences in infection-mediated behavior that enhance viral transmission. Interactions with other infections/symbionts should not be ignored: the host taxa may have, for example, diverged not only in their genomes but also in their gut microbiomes. It should also be borne in mind that host-intrinsic effects may be associated with hybrids, which can show vigorous immune response due to heterosis [88]. The combination of mechanisms involved is unlikely to be simple. For example, the configuration of α-dystroglycan, the (known) main cell entry receptor of Old-World mammarenaviruses is invariant across several species of the Mastomys genus [89], ruling this out as a simple explanation. Replicating relevant aspects of the population-level process of viral transmission in the laboratory will therefore be challenging.
As mentioned in the introduction, the human Lassa fever endemic area roughly matches the distribution range of M. natalensis A-I matrilineage, but recently, Olayemi et al. (2016) found LASV in three M. natalensis individuals carrying the A-II mitochondrion and concluded that LASV may spread to the rest of the A-II matrilineage range (which extends up to eastern Democratic Republic of Congo; Fig 1; [43]) [39]. However, the observations in Olayemi et al. (2016) actually appear consistent with the current study: 1) frequencies of matrilineages A-I and A-II (only mitochondria were typed) gradually changed along a west-east axis; 2) LASV and no other arenavirus was found in two localities where A-I predominated over A-II (1/9 and 2/9 genotyped animals carrying A-II, the remainder A-I); 3) a different, Mobala-like arenavirus, but not LASV, was found in localities where A-II mitochondrion predominates (A-II found in 3/3 genotyped animals, and lies east of a locality where A-II was found in 17/19 animals). As the authors note, M. natalensis subtaxa associated with A-I and A-II matrilineages thus likely form a hybrid zone in eastern Nigeria, potentially coinciding with the river Niger. Comparing these findings with our multilocus fine-scale data from Tanzania, we would predict that the locality where LASV was detected in three individuals with A-II mitochondria lies west of the hybrid zone’s centre, across which these A-II mitochondrial copies have introgressed, as is common in contact zones [56] and as we also observed in this study. We would thus predict multilocus (not mitochondrial) genotyping would cluster those three individuals in the “A-I subtaxon”. The convergent spatial patterns in Olayemi et al. (2016) and this study are thus consistent with a general association of the arenaviruses of M. natalensis to particular subtaxa, implying that LASV is restricted to the West African range of the subtaxon corresponding with A-I matrilineage.
However, firstly the potential dispersal barrier effect of the river Niger on M. natalensis nuclear gene flow and LASV transmission should be evaluated as alternative explanation. Secondly, the role of other rodent hosts in the spatial spread of LASV needs to be clarified. It has recently become clear that several strains of LASV may be harboured by rodent species other than M. natalensis, namely the closely related M. erythroleucus and Hylomyscus pamfi [34], and divergent LASV-related strains have been found in distantly related Mus baoulei and Mus cf. setulosus [90].
Two important questions thus remain: (1) is LASV a generalist whereas other African arenaviruses, especially GAIV and MORV, are specialists? This would explain why LASV is found in a wide array of species, including humans, but fails to explain why the distribution of LASV appears largely bordered by the distribution of M. natalensis A-I matrilineage. Therefore: (2) Do hosts other than M. natalensis A-I subtaxon contribute to the long-term persistence of LASV in nature?
M. erythroleucus has been found carrying LASV in an area just outside of M. natalensis’ range (coastal Guinea) and where human Lassa fever cases have also been reported–implying that either humans or M. erythroleucus managed to import LASV and establish a transmission chain without involvement of M. natalensis [34]. On the other hand, it is clear that these importations have not (yet) occurred very far outside of the M. natalensis A-I matrilineage range, despite the continuous distribution of M. erythroleucus, H. pamfi and of course humans through to other parts of Africa. It therefore seems that either: 1) M. erythroleucus and H. pamfi are not able to sustain a long-term persistent LASV transmission, similar to the situation in humans [27, 28]. Such less efficient transmission could e.g. be due to differences in intra-host infection dynamics and population ecology of these species in comparison to M. natalensis. 2) LASV is sustainably transmitted in M. erythroleucus and H. pamfi populations, and LASV’s distribution in reality, and unnoticed, expands throughout the ranges of M. erythroleucus and H. pamfi (though note that at least M. erythroleucus is known to be subdivided into subtaxa similar to M. natalensis [91]). Perhaps they carry LASV strains less pathogenic to humans elsewhere; such a situation has previously been postulated to explain the absence of reported Lassa fever patients in regions in Mali where a particular LASV strain has only recently been found to be common in M. natalensis [32]. 3) LASV strains in M. erythroleucus and H. pamfi are the result of adaptive host switching events that occurred only recently, and have yet to expand through the rest of the rodents’ ranges.
It is clear more surveillance of arenaviruses in rodents, especially in the area bordering the Lassa fever endemic area, is needed to fully answer the pressing question: should we expect LASV and its associated diseases to emerge in the rest of Africa?
|
10.1371/journal.pmed.1002079 | Exercise Training and Weight Gain in Obese Pregnant Women: A Randomized Controlled Trial (ETIP Trial) | The effectiveness of exercise training for preventing excessive gestational weight gain (GWG) and gestational diabetes mellitus (GDM) is still uncertain. As maternal obesity is associated with both GWG and GDM, there is a special need to assess whether prenatal exercise training programs provided to obese women reduce the risk of adverse pregnancy outcomes. Our primary aim was to assess whether regular supervised exercise training in pregnancy could reduce GWG in women with prepregnancy overweight/obesity. Secondary aims were to examine the effects of exercise in pregnancy on 30 outcomes including GDM incidence, blood pressure, blood measurements, skinfold thickness, and body composition.
This was a single-center study where we randomized (1:1) 91 pregnant women with a prepregnancy body mass index (BMI) ≥ 28 kg/m2 to exercise training (n = 46) or control (standard maternity care) (n = 45). Assessments were done at baseline (pregnancy week 12–18) and in late pregnancy (week 34–37), as well as at delivery. The exercise group was offered thrice weekly supervised sessions of 35 min of moderate intensity endurance exercise and 25 min of strength training. Seventeen women were lost to follow-up (eight in the exercise group and nine in the control group). Our primary endpoint was GWG from baseline testing to delivery. The principal analyses were done as intention-to-treat analyses, with supplementary per protocol analyses where we assessed outcomes in the women who adhered to the exercise program (n = 19) compared to the control group. Mean GWG from baseline to delivery was 10.5 kg in the exercise group and 9.2 kg in the control group, with a mean difference of 0.92 kg (95% CI −1.35, 3.18; p = 0.43). Among the 30 secondary outcomes in late pregnancy, an apparent reduction was recorded in the incidence of GDM (2009 WHO definition) in the exercise group (2 cases; 6.1%) compared to the control group (9 cases; 27.3%), with an odds ratio of 0.1 (95% CI 0.02, 0.95; p = 0.04). Systolic blood pressure was significantly lower in the exercise group (mean 120.4 mm Hg) compared to the control group (mean 128.1 mm Hg), with a mean difference of −7.73 mm Hg (95% CI −13.23, −2.22; p = 0.006). No significant between-group differences were seen in diastolic blood pressure, blood measurements, skinfold thickness, or body composition in late pregnancy. In per protocol analyses, late pregnancy systolic blood pressure was 115.7 (95% CI 110.0, 121.5) mm Hg in the exercise group (significant between-group difference, p = 0.001), and diastolic blood pressure was 75.1 (95% CI 71.6, 78.7) mm Hg (significant between-group difference, p = 0.02). We had planned to recruit 150 women into the trial; hence, under-recruitment represents a major limitation of our results. Another limitation to our study was the low adherence to the exercise program, with only 50% of the women included in the intention-to-treat analysis adhering as described in the study protocol.
In this trial we did not observe a reduction in GWG among overweight/obese women who received a supervised exercise training program during their pregnancy. The incidence of GDM in late pregnancy seemed to be lower in the women randomized to exercise training than in the women receiving standard maternity care only. Systolic blood pressure in late pregnancy was also apparently lower in the exercise group than in the control group. These results indicate that supervised exercise training might be beneficial as a part of standard pregnancy care for overweight/obese women.
ClinicalTrials.gov NCT01243554
| Maternal obesity is associated with increased risk of several adverse pregnancy outcomes.
The aim of our study was to investigate whether exercise training during pregnancy can reduce gestational weight gain and prevent negative health outcomes, such as gestational diabetes mellitus and high blood pressure, among overweight/obese pregnant women.
Ninety-one overweight/obese pregnant women were randomly allocated to an exercise group or a control group in early pregnancy (starting in pregnancy week 12–18).
Women in the exercise group were asked to attend supervised sessions of combined endurance and resistance training three times weekly.
Women in both groups gained on average about 10 kg. In late pregnancy, there seemed to be fewer women in the exercise group with gestational diabetes mellitus, and the exercising women had lower systolic blood pressure compared to those in the control group.
Providing an exercise program to overweight/obese pregnant women did not reduce gestational weight gain compared to standard pregnancy care.
Exercise training might reduce the incidence of gestational diabetes mellitus and lower systolic blood pressure in late pregnancy in this population.
| Maternal obesity is a risk factor for adverse pregnancy outcomes, such as gestational diabetes mellitus (GDM) [1], gestational hypertension, preeclampsia, need for cesarean delivery, and large for gestational age [2–4]. Because the prevalence of overweight and obesity among reproductive-age women is increasing, effective preventive strategies are urgently needed.
Excessive gestational weight gain (GWG) is also associated with negative obstetric outcomes [1,5,6]. The 2009 Institute of Medicine (IOM) guidelines on GWG suggest that underweight women (body mass index [BMI] ≤ 18.5 kg/m2) should gain 12.5–18.0 kg during pregnancy; normal weight women (BMI 18.5–24.9 kg/m2), 11.5–16.0 kg; overweight women (BMI 25.0–29.9 kg/m2), 7.0–11.5 kg; and obese women (BMI ≥ 30.0 kg/m2), 5.0–9.0 kg [7]. Overweight and obese women are about two times more likely than normal weight women to exceed these recommendations [8]; thus, there is a special need to find feasible and effective interventions to reduce GWG in women with a high BMI.
Previous research on clinical effects of lifestyle interventions during pregnancy in overweight/obese women has shown conflicting results [9–14]. Most studies have assessed the combined effect of physical activity and dietary guidance. To our knowledge, there are only three previous randomized controlled trials (RCTs) [14–16] assessing the isolated effects of exercise training in pregnancy on GWG and clinical outcomes in overweight and obese women. These studies found no significant difference in GWG between exercise and control groups. However, one study was limited by a small study sample (n = 12) [14], and one study reported results from only a subgroup [15].
Few studies exist on GDM prevention via exercise training in obese women [17–19], and to our knowledge no previous RCT has shown that GDM can be prevented by exercise training as the sole intervention [14,18,20,21]. However, according to a recent meta-analysis [22], structured physical exercise programs during pregnancy decrease the risk of GDM. Hence, there is still a need to establish the potential effects of exercise training on GDM prevention, and especially so in overweight/obese women.
To address the shortcomings in the research on effective prevention of GWG and of GDM, our aim was to assess whether regular supervised exercise training could reduce GWG and improve clinical outcomes, compared to standard maternity care, in women with a prepregnancy BMI of 28 kg/m2 or more.
The study was approved by the Regional Committee for Medical and Health Research Ethics (REK midt 2010/1522) and registered in ClinicalTrials.gov (NCT01243554). The Exercise Training in Pregnancy (ETIP) trial was a single-center, parallel-group RCT of regular exercise training during pregnancy compared to standard maternity care in women with prepregnancy BMI ≥ 28 kg/m2. The study protocol has been published previously [23]. The trial was performed at the Norwegian University of Science and Technology (NTNU) and St. Olavs Hospital, Trondheim University Hospital, in Trondheim, Norway.
We made the following changes to the protocol after trial commencement: body composition was measured using air displacement plethysmography starting 28 June 2011, to improve assessments of body composition. The time limit for completed baseline testing and inclusion into the study was changed from gestational week 16 to gestational week 18 on 15 November 2012, and we changed the inclusion criteria for BMI from ≥30 to ≥28 kg/m2 on 22 March 2013. We changed the time limit for inclusion and the BMI criteria to increase recruitment into the trial. All changes were reported and approved by the Regional Committee for Medical and Health Research Ethics. The procedures followed in the ETIP study were in accordance with ethical standards of research and the Helsinki Declaration.
At recruitment, women received written information, and they signed informed consent on behalf of themselves and their offspring before participation and randomization. Inclusion criteria were prepregnancy BMI ≥ 28 kg/m2, age ≥ 18 y, gestational week < 18, and carrying one singleton live fetus at 11–14 wk ultrasound scan. The participants had to be able to come to St. Olavs Hospital for assessments and exercise classes. Exclusion criteria were high risk for preterm labor, diseases that could interfere with participation, and habitual exercise training (twice or more weekly) in the period before inclusion. Women were recruited through invitations sent along with notices for routine ultrasound scan at the hospital, and additionally through Google advertisements. The women received infant food worth US$65. The participants in this study gave written informed consent to publication of their case details.
The exercise group was offered, in addition to standard maternity care, exercise sessions at the hospital three times weekly, from baseline (gestational week 12–18) until delivery. The exercise sessions were supervised by a physical therapist and were in accordance with the recommendations from the American College of Obstetricians and Gynecologists [24]. Each session lasted 60 min and consisted of treadmill walking/jogging for 35 min (endurance training) and resistance training for large muscle groups and the pelvic floor muscles for 25 min. The intensity of the endurance training was set to ~80% of maximal capacity (corresponding to Borg scale 12–15) [25]. The resistance training consisted of squats, push-ups, diagonal lifts on all fours, and oblique abdominal crunches, with three sets of ten repetitions of each exercise separated by a 1-min rest between sets. Participants also did three sets of the “plank exercise” for 30 s. We adjusted the program according to each woman’s strength level. The pelvic floor exercises consisted of three sets of ten repetitions of pulling the pelvic floor up and holding the contraction for 6–8 s.
In addition, the women were asked to follow a 50-min home exercise program at least once weekly (35 min of endurance training and 15 min of strength exercises) and to do daily pelvic floor muscle exercises. We registered adherence to the supervised exercise program, and the participants reported their home exercise in a training diary. The participants received a weight gain curve showing recommended weight gain throughout pregnancy in accordance to 2009 IOM guidelines [7], and were encouraged to compare their own weight gain with this curve. The women were invited to attend one motivational interview session [23], either individually or in a group, during the intervention period.
The control group received ordinary maternity care by their midwife, general practitioner, and/or obstetrician. The Norwegian national directions for standard maternity care among healthy pregnant women at the time the study was conducted included offering of an ultrasound examination by gestational week 18 and providing information about healthy eating and healthy lifestyle [26]. The women in the control group were asked to continue their normal daily activities and were not discouraged from exercising on their own.
All participants underwent the same test protocol at baseline (gestational week 12–18) and at late pregnancy (gestational week 34–37). In addition, the hospital personnel measured the women’s body weight immediately before delivery.
Our primary outcome measure was GWG calculated as the difference between weight at baseline and weight at delivery. Maternal body weight at baseline, in late pregnancy, and before delivery was measured with a calibrated electronic scale (Seca 770, Medema, Norway) to the nearest 0.1 kg, with the participant wearing indoor clothing, without shoes. If the hospital staff did not have time to measure the women’s weight right before delivery, we used women’s self-reported weight at the time of delivery to calculate the outcome measure.
Secondary outcome measures were BMI, body composition, physical activity level, skinfold thickness, blood pressure, various blood tests, incidence of GDM, and incidence of maternal hypertension in late pregnancy. Height was measured at baseline with a wall-mounted Seca 222 stadiometer. BMI was calculated as weight in kilograms divided by the square of height in meters. Systolic and diastolic blood pressure were measured on the right arm after 15 min of supine resting using a CASMED 740 MAXNIBP (CAS Medical Systems). We used the average of three measurements taken at 2-min intervals. Skinfold thickness was measured on the right side of the body at the sites subscapular, biceps, and triceps, using a Harpenden Skinfold Caliper (Holtain). We used the average of three measurements for each site. Body composition was measured using air displacement plethysmography (BOD POD, COSMED). The participant entered the BOD POD wearing only underwear and a swim cap. Physical activity level was measured by a questionnaire where the participants reported their frequency, duration, and intensity of weekly physical activity.
After a 10-h fast, we drew venous blood for fasting plasma glucose and other blood measurements. The participants then drank 75 g of glucose dissolved in 2.5 dl of water, and blood was drawn again after 2 h (120-min plasma glucose). According to the study protocol [23], GDM was to be diagnosed by the 2009 WHO definition: fasting plasma glucose ≥ 7.0 mmol/l and/or 120-min plasma glucose ≥ 7.8 mmol/l [27]. However, in 2013 WHO, in collaboration with the International Association of Diabetes and Pregnancy Study Groups (IADPSG), endorsed adjusted diagnostic criteria for classification of GDM: fasting plasma glucose ≥ 5.1 mmol/l and/or 120-min plasma glucose ≥ 8.5 mmol/l [28]. GDM is therefore reported here by both definitions. Plasma glucose, high-sensitivity C-reactive protein (CRP), total cholesterol, high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, triglycerides, HbA1c, ferritin, and hemoglobin were measured using a Roche Modular P. We assessed insulin with ELISA (IBL International) using a DS2 ELISA processing system (Dynex Technologies). All assays were performed according to the manufacturer’s instructions. The inter- and intra-assay coefficients of variation were 2.1% and 1.5% for glucose, 3.8% and <1% for high-sensitivity CRP, 2.5% and 0.9% for total cholesterol, 2.8% and 0.8% for HDL cholesterol, 2.4% and 0.8% for LDL cholesterol, 2.9% and 0.9% for triglycerides, and 5.3% and 9.5% for insulin. Homeostatic assessment of insulin resistance (HOMA2-IR) was calculated as [glucose × insulin]/22.5 [29].
Sample size was calculated based on prior studies [30,31] using a 6-kg clinically relevant difference in mean weight gain between the exercise and the control group, from baseline to delivery. According to this, a two-sided independent sample t-test with a 5% level of significance, a standard deviation of 10, and a power of 0.90 gave a target study population of 59 in each group. Dropout was estimated at 15%; therefore, we aimed to include 150 women.
After baseline assessments, the participants were randomly allocated 1:1 to the intervention or the control group. Allocation was done using a computer random number generator developed and administrated at the Unit for Applied Clinical Research, NTNU. The randomization had varying block sizes, with the first, the smallest, and the largest block defined by the computer technician at the Unit for Applied Clinical Research. The investigators enrolling the patients (K. K. G. and T. M.) got the allocation results on screen and by e-mail after registration of each new participant into the study and did not have the full randomization list available.
Weight measurement at delivery and blood analyses were done by personnel blinded for group allocation. All other assessments and intervention administration were done non-blinded. The statistician conducting the statistical analyses was blinded for group allocation.
The trial and the principal analyses were based on intention to treat. All available data were used at all time points. We also performed, as described in the original protocol, per protocol analyses including only the women in the exercise group who adhered to the exercise protocol [23]. Baseline data were tested for normality and analyzed by an independent sample t-test and by Fisher’s exact test.
The outcome measurements were analyzed in accordance to the treatment arm to which patients were randomized, regardless of nonadherence. The effect of treatment on the primary and secondary outcomes was assessed with mixed linear models for continuous outcomes and mixed logistic models for dichotomous outcomes. For the primary outcome, the effect of time and treatment was taken as a fixed effect having the levels baseline, training late pregnancy, control late pregnancy, training delivery, and control delivery. For the secondary outcomes, the effect of time and treatment was taken as a fixed effect having the levels baseline, training late pregnancy, and control late pregnancy. Due to randomization, no systematic differences between groups at baseline were assumed. To account for repeated measurements, participant ID was included as a random effect. The analyses were performed using R version 2.13.1, Stata version 13.1, and IMB SPSS Statistics 22. All results are given as mean values with 95% confidence intervals, and p-values less than 0.05 were considered significant. We did supplementary analyses of GWG where we adjusted for gestational age at delivery.
Per protocol analyses [23] including only the women in the exercise group who adhered to the exercise protocol were performed on both primary and secondary outcomes. Adherence to the exercise protocol was defined as (1) attending ≥ 42 organized exercise sessions, (2) attending ≥ 28 exercise sessions + performing ≥ 28 home exercise sessions, or (3) performing ≥ 60 home exercise sessions. The exercise had to be ≥50 min of either aerobic or strength training to count as a home session.
Fig 1 outlines the flow of participants during the trial.
Recruitment started on 20 September 2010 and was continued until 1 March 2015. The final data collection date for the primary outcome measure was 20 June 2015. The aim of our study was to include 150 participants, but enrollment was stopped on 1 March 2015 at 91 randomized participants, due to the prolonged time for inclusion and fewer eligible participants than expected. Table 1 shows the baseline characteristics of the participants.
There were no significant differences between groups at baseline, except from mean fasting glucose (4.6 mmol/l in the exercise group, 5.0 mmol/l in the control group; p = 0.02). Table 2 shows the model-based analyses for the continuous primary and secondary outcomes. The mean number of weeks from inclusion to delivery was 23.3 (range 10–28) in the exercise group and 24.7 (range 19–30) in the control group. Mean gestational age was 39.5 wk (range 27–42 wk) in the exercise group and 39.4 wk (range 37–42) in the control group.
We found no significant differences in GWG between the exercise group and the control group (Table 2). Body weight at delivery was self-reported by five women in the exercise group and four women in the control group. The proportion of women exceeding the IOM guidelines for recommended GWG was similar in the two groups (Table 3). Adjusting for gestational age in the analyses did not affect the GWG comparison between groups significantly (mean difference 0.56, p = 0.67).
In late pregnancy, two women (6.1%) in the exercise group and nine women (27.3%) in the control group had developed GDM according to the WHO 2009 definition [27], with a statistical difference between groups (p = 0.04; Table 3). According to the WHO/IADPSG 2013 definition of GDM [28], there was no significant difference between the groups (Table 3). There was no significant difference in fasting glucose, 120-min glucose, insulin, or HbA1c level between the groups (Table 2).
In late pregnancy we found a significantly lower systolic blood pressure (p = 0.006) in the exercise group compared to the control group (Table 2). There were no significant differences in other secondary outcome measures (Tables 2 and 3).
The proportion of women reporting to be physically active for at least 30 min each day in late pregnancy was equal in the two groups: 61% in the exercise group and 66% in the control group (p = 0.73). The proportion of women reporting regular exercise training in late pregnancy was significantly higher in the exercise than in the control group: 77% and 23%, respectively (p < 0.01).
In the exercise group, 50% of the women fulfilled the training intervention as described in the study protocol [23]. In the per protocol analyses, we found no significant difference in weight gain and mean weight at delivery between the per protocol exercise group and the control group (S1 Table). Resting systolic and diastolic blood pressure were significantly lower in the per protocol exercise group (115.7 mm Hg/75.1 mm Hg) compared to the control group (128.1 mm Hg/80.2 mm Hg), with p = 0.001 and p = 0.02, respectively. A tendency toward lower incidence of GDM (5.9% in the per protocol exercise group, 27.3% in the control group, p = 0.11) and maternal hypertension (11.1% in the per protocol exercise group, 21.2% in the control group, p = 0.14) was seen in the per protocol exercise group (S2 Table).
No adverse events were reported during the exercise training or study assessments (Table 4).
We found no difference in GWG between women randomized to an exercise training program versus standard maternity care, but found an apparent reduction in the incidence of GDM and lower systolic blood pressure in late pregnancy among the women randomized to the exercise training program. In the per protocol analyses including only the women who had adhered to the exercise program (n = 19), exercise training also seemed to reduce diastolic blood pressure in late pregnancy.
Our findings of no difference in GWG and body composition between groups are in line with several other clinical trials on overweight or obese pregnant women [14–16,32–34]. However, a systematic review by Sui and Dodd [20] that included 216 participants (five randomized trials) found that supervised exercise interventions were associated with lower GWG among overweight or obese pregnant women. But the trials included in this systematic review differed with respect to the type and duration of exercise, and a clinically relevant difference in weight gain was not precalculated. Also Barakat et al. [35] and Haakstad and Bø [36] found significantly lower GWG among women who participated in supervised exercise during pregnancy. However, these two studies included women from all weight classes, and their findings might not translate specifically to overweight/obese women. We can only speculate about why there was no difference in GWG between the two groups in the ETIP study. The proportion of women whose self-reported activity level fulfilled the recommended 30 min of daily physical activity in late pregnancy was higher in the exercise group, but some of the women in the control group exercised on their own. Only 50% of the women in the exercise group adhered to the exercise protocol as prescribed a priori [23]. A possible effect of regular exercise during pregnancy may have been missed in our study due to the relatively low adherence to the training protocol. Protocol adherence is a challenge in all exercise studies. We tried to improve adherence by offering motivational talks throughout the intervention period, as well as adjusting the training times so that more women would be able to attend. The low adherence may have been due to pregnancy symptoms such as tiredness and nausea, limited previous experience with exercise training, or difficulties in prioritizing time for exercise. Furthermore, the intervention protocol might have been too comprehensive for these women. Further studies should carefully consider how exercise adherence can be obtained in this population.
Although the exercise training program in our study followed the current recommendations for exercise in pregnancy, it is possible that the exercise frequency and/or intensity of our program were not sufficient to affect the outcome measures related to weight gain. As our study population had a relatively low fitness level, the amount of energy spent during the exercise sessions was rather low (~400 kcal/session) and was probably not sufficient to affect the energy balance significantly. It is also possible that some of the women in the exercise group compensated for energy expenditure during the exercise sessions either by decreasing their physical activity level during the remaining time of the week [37] or by increasing their energy intake [38]. According to three recent meta-analyses [32,39,40], interventions combining physical activity and diet have proven effective in reducing GWG in overweight and obese women. We did not include any dietary advice or intervention in our study, and probably exercise training alone is not sufficient to reduce GWG in this population.
Changes in body composition throughout pregnancy might be an important determinant of glucose metabolism. Few studies have assessed body composition changes after exercise in pregnancy. Our findings of no significant differences between groups in body composition in late pregnancy are in line with a recent RCT on the effects of a 16-wk moderate intensity cycling program in overweight and obese pregnant women [33].
Our finding of an apparently lower incidence of GDM according to the WHO 2009 definition [27] among the women in the exercise group is in line with a recent meta-analysis of 13 RCTs [22] that concluded that structured moderate intensity exercise programs during pregnancy decrease the risk of GDM. However, two previous Cochrane reviews, one on exercise as the sole intervention [19] and one on both diet and exercise interventions [41], concluded that there is no clear GDM risk reduction after exercise training. Nobles et al. [42] randomized 251 women with increased risk of GDM to either exercise training or a comparison health and wellness group and found no reduction in GDM risk after exercise, in line with another previous review [20]. The recently published DALI Lifestyle Pilot study [43] found that women with BMI ≥ 29 kg/m2 randomized to a healthy eating intervention had significantly lower fasting glucose and 2-h insulin concentrations than women in an exercise only group. In contrast to the DALI study, our results indicate that exercise training alone may be sufficient to prevent glucose intolerance in overweight or obese pregnant women. An important difference between the DALI study and ours is that the exercise training was supervised in our study.
Using the WHO/IADPSG 2013 definition [28] of GDM in the ETIP study, the number of GDM cases increased in both groups, and there was no longer a significant difference between the groups. The WHO/IADPSG 2013 definition is mainly based on the HAPO study (2008) [44], which found strong associations between glucose levels below the WHO 2009 diagnostic definitions and adverse outcomes for both mother and child. However, a retrospective cohort study [45] that included 1,892 women diagnosed with GDM according to the WHO/IADPSG 2013 definition found a significantly higher risk for adverse pregnancy outcomes in those who also would be diagnosed as having GDM according to the WHO 2009 definition.
Despite the difference between the exercise and control groups in GDM incidence, we found no differences between the groups at late pregnancy in glucose levels, insulin, or HOMA2-IR. One possible reason for this finding is that women with high risk of GDM may respond differently to exercise training than women with lower risk [46], such that the average glucose and insulin levels are not sufficiently affected to obtain a difference between groups.
We found a significantly lower systolic blood pressure among the women in the exercise group in late pregnancy, compared to the women in the control group. Diastolic blood pressure did not differ between groups in late pregnancy in the intention-to-treat analysis, but was significantly lower in the exercise group in the per protocol analysis. High blood pressure in pregnancy is associated with increased risk for preeclampsia [47] and thus is important to prevent. To our knowledge, only one previous RCT [33] has studied the effect of exercise training in pregnancy on exact blood pressure measurements among overweight/obese women. Seneviratne et al. [33] found no effect of exercise training on blood pressure in late pregnancy. Other studies that have assessed the effects of exercise on maternal hypertension risk have assessed hypertension as a dichotomous variable [34,35,39]. Some of these studies found no effect of exercise [17,34], but one study [35] found a reduced incidence of maternal hypertension after exercise training. The latter study included both normal weight, overweight, and obese women. Although fewer women in the exercise group than in the control group had hypertension in late pregnancy in our study, the difference was not statistically significant. Further studies are needed to ascertain whether exercise training can prevent hypertensive pregnancies in overweight/obese women.
The ETIP study had few exclusion criteria, and the participants were representative of Norwegian women with BMI ≥ 28 kg/m2 regarding age, parity, and education. However, participants had to have time available for the testing and training. The exercise group was offered training sessions at day and evening times. It is also possible that the participants volunteering for the ETIP study were extra aware of the possible beneficial effects of exercise training in pregnancy and thus were motivated to participate in our trial.
Obese women have elevated risk of GDM and maternal hypertension; thus, finding effective prevention strategies is highly relevant. The study revealed no adverse events related to moderate physical activity during pregnancy. The effect of exercise training to reduce weight gain may most likely be improved with additional dietary interventions. During the study we experienced difficulties in motivating the women in the exercise group to adhere to the training program, despite supervised training sessions at St. Olavs Hospital, training sessions at different times during the week, and individually adjusted exercises. We think further studies should evaluate how supervised exercise programs for obese women can be implemented in the health care system, as well as how to obtain good adherence to such programs.
In our study, exercise training was the only intervention provided. This makes it easier to assess the isolated effects of exercise on pregnancy outcomes. The training program being standardized and supervised makes it easy to reproduce. Furthermore, we had thorough recording of exercise adherence as well as physical activity levels in the two groups. The primary outcome measure (GWG) was assessed by personnel blinded for group allocation. We also regard the assessment of body composition with the gold standard method of air displacement plethysmography as a strength.
The main limitation of the trial was the reduced statistical power because we were able to include only 2/3 of the 150 participants estimated in the power calculation. We analyzed 30 different secondary outcomes among a limited number of women, and thereby increased the risk for detecting differences between groups by chance, and making type 1 errors. Furthermore, only 50% of the participants in the exercise group performed the exercise training program per protocol, which makes it more difficult to detect possible effects of the intervention. However, adherence to exercise in the ETIP study was similar to that in most of the comparable clinical studies. Care must be taken in interpreting the results from the per protocol analysis. Such analyses could be selection biased if the reasons influencing compliance with the exercise training program are associated with prognostic factors [48].
In this trial we did not observe a reduction of GWG or an improvement in body composition among overweight/obese women who were offered supervised exercise training during pregnancy. However, exercise training seemed to reduce the incidence of GDM as well as systolic blood pressure in late pregnancy. As exercise adherence is a major challenge in this population, there is a special need to find methods to reduce participant attrition in future studies.
|
10.1371/journal.ppat.1007500 | Desialylation of platelets induced by Von Willebrand Factor is a novel mechanism of platelet clearance in dengue | Thrombocytopenia and platelet dysfunction are commonly observed in patients with dengue virus (DENV) infection and may contribute to complications such as bleeding and plasma leakage. The etiology of dengue-associated thrombocytopenia is multifactorial and includes increased platelet clearance. The binding of the coagulation protein von Willebrand factor (VWF) to the platelet membrane and removal of sialic acid (desialylation) are two well-known mechanisms of platelet clearance, but whether these conditions also contribute to thrombocytopenia in dengue infection is unknown. In two observational cohort studies in Bandung and Jepara, Indonesia, we show that adult patients with dengue not only had higher plasma concentrations of plasma VWF antigen and active VWF, but that circulating platelets had also bound more VWF to their membrane. The amount of platelet-VWF binding correlated well with platelet count. Furthermore, sialic acid levels in dengue patients were significantly reduced as assessed by the binding of Sambucus nigra lectin (SNA) and Maackia amurensis lectin II (MAL-II) to platelets. Sialic acid on the platelet membrane is neuraminidase-labile, but dengue virus has no known neuraminidase activity. Indeed, no detectable activity of neuraminidase was present in plasma of dengue patients and no desialylation was found of plasma transferrin. Platelet sialylation was also not altered by in vitro exposure of platelets to DENV nonstructural protein 1 or cultured DENV. In contrast, induction of binding of VWF to glycoprotein 1b on platelets using the VWF-activating protein ristocetin resulted in the removal of platelet sialic acid by translocation of platelet neuraminidase to the platelet surface. The neuraminidase inhibitor oseltamivir reduced VWF-induced platelet desialylation. Our data demonstrate that excessive binding of VWF to platelets in dengue results in neuraminidase-mediated platelet desialylation and platelet clearance. Oseltamivir might be a novel treatment option for severe thrombocytopenia in dengue infection.
| Dengue is the most common arbovirus infection in the world. A decrease in the number of blood platelets is an almost universal finding in severe dengue. Binding of the coagulation protein von Willebrand factor (VWF) and loss of sialic acid residues from the platelet membrane are two main mechanisms of clearance of senescent platelets under non-pathological conditions. Here, we show that platelets from patients with acute dengue have bound more VWF and have lost sialic acid from their membrane. Sialic acid can be cleaved by the enzyme neuraminidase. We show that neuraminidase activity in the plasma is not increased and that neither dengue virus itself nor nonstructural protein 1, a protein secreted by dengue virus, cleave sialic acid from the platelet membrane. In contrast, binding of VWF to platelets results in translocation of neuraminidase to the platelet membrane and subsequent cleavage of sialic acid. This process could be inhibited by the neuraminidase inhibitor oseltamivir, a commonly used anti-influenza drug. Altogether, our results indicate that VWF binding to platelets is increased in dengue infection, leading to the removal of sialic acid and platelet clearance. Oseltamivir may prevent this process and thus represent a novel treatment option for low platelet numbers in dengue infection.
| Dengue is the most common arboviral infection in the world with an estimated number of 390 million annual cases, of which 96 million manifests with symptomatic disease [1]. A subset of patients with symptomatic infections develops potentially life-threatening complications in which bleeding and vascular plasma leakage are the most common [2]. To date, there is no curative therapy for dengue and clinical observation and treatment of complications remain the core principles of dengue management.
Thrombocytopenia is an early and consistent feature of dengue virus infection [3–6] and dengue complications are usually preceded by a rapid drop in platelet count [2]. Traditionally known for their key role in hemostasis, platelets are nowadays well known to have important additional functions, including regulation of inflammation and host defense [7–9] and preservation of endothelial integrity [10], especially under inflammatory conditions [11].
Circulating platelets of dengue patients are activated and excessive platelet activation may lead to platelet exhaustion [12–15], which likely contributes to thrombocytopenia and dengue complications. Other well-known mechanisms of increased platelet clearance under non-pathological conditions are the binding of von Willebrand factor (VWF) to platelets and loss of sialic acids from the platelet membrane [16]. VWF is a multimeric protein that is present in plasma, platelets, and endothelial cells. Endothelial activation results in the release of VWF from Weibel-Palade bodies into the blood circulation. Its primary function is platelet adhesion and aggregation at sites of vessel injury [17]. The antibiotic ristocetin induces binding of VWF to platelets and may cause thrombocytopenia, and as a consequence, it was removed from clinical practice [18]. We have previously shown that the plasma of children with severe dengue contains increased levels of circulating VWF in an active–platelet binding–conformation and increased VWF proteolysis [19]. Importantly, enhanced VWF binding to platelets—as seen in von Willebrand disease (VWD) type 2B - was shown to induce phagocytosis of VWF/platelet complexes by macrophages in the liver and spleen and promote thrombocytopathy by inhibition of platelet integrin αIIbβ3, the platelet fibrinogen receptor [20, 21]. Increased platelet phagocytosis, as well as impaired platelet αIIbβ3 function, have also been described in patients with dengue [15, 22].
Most platelet glycoproteins, including the extracellular domains of Glycoprotein 1bα (GPIbα; ‘the major VWF receptor’), are decorated with N-glycans and O-linked glycans, which are covered by sialic acids [23, 24], a family of monosaccharides with a 6-carbon backbone. Loss of terminal sialic acids, which can be mediated by human sialidase neuraminidase 1 (NEU1) or exogenous sialidase, exposes β-galactose and β-N-acetyl-D glucosamine (β-GlcNAc), which can be recognized by macrophages or the hepatic asialoglycoprotein receptor (ASGPR; also called the Ashwell–Morell receptor) [25–28]. Recently, binding of VWF to GPIbα under physiological shear was shown to induce platelet desialylation [29].
The aim of our study was to investigate these platelet clearance processes during the course of the disease in patients with dengue and their relationship with thrombocytopenia and platelet reactivity. We first show that dengue is associated with increased binding of VWF to platelets. In a follow up clinical study, we also show that circulating platelets have markedly reduced surface sialic acid. Both VWF binding and desialylation were associated with thrombocytopenia. In a set of ex vivo studies, we then show that VWF binding to platelets leads to NEU-1 mediated desialylation, which can be inhibited by the sialidase inhibitor oseltamivir.
Two observational clinical studies were performed in Indonesian patients with acute dengue and conducted according to the principles expressed in the Declaration of Helsinki. In study 1, the binding of VWF to circulating platelets was investigated and related to thrombocytopenia and platelet reactivity. In study 2, sialic acid on the platelet surface was determined. Both studies were exploratory in nature and therefore no sample size calculation was performed.
The first clinical study was approved by the local Medical Ethical Committee of the Medical Faculty of Padjadjaran University, Hasan Sadikin General Hospital. The second clinical study was approved by the local Medical Ethical Committee of the Faculty of Medicine, Diponegoro University, and Kartini Hospital. In both clinical studies, written informed consent was obtained before enrolment from all patients and healthy controls, and in the case of pediatric subjects, a parent or guardian provided written informed consent on behalf of the child.
Statistical analyses were performed using GraphPad Prism 5 software. Data are presented as geometric mean with 95% confidence intervals unless stated otherwise. The Kruskal Wallis test was used to determine significance between multiple groups. Differences between individual groups were analyzed with the Mann-Whitney test or Student’s T-test. Correlations were tested using either the Spearman correlation test (for non-normally distributed data) or Pearson correlation test (normally distributed data). A P value <0.05 was considered a statistically significant difference. Flow cytometry data were analyzed with Beckman Coulter Kaluza software, version 1.2. Data on transferrin glycosylation were analyzed with Agilent Mass Hunter Qualitative Analysis Software B.04.00.
In Bandung, a total number of 40 patients with acute dengue and 10 healthy adult controls were enrolled. Three patients were children aged ≤14 years old. Samples collected were divided into four groups based on the day of fever; day 1–3 (n = 20), day 4–6 day (n = 54), day 7–13 (n = 24) and day >13 days (n = 24). In Jepara, 40 patients with dengue were enrolled, together with 15 patients with non-dengue illness (acute gastroenteritis n = 9, acute respiratory tract infection n = 6) and 25 healthy controls. Ten and three of the patients with dengue or non-dengue febrile illness respectively were children. Clinical characteristics are given in Table 1. Hemorrhagic manifestations were common in both dengue cohorts (35% in Bandung and 38% in Jepara) and manifested predominantly as skin or mucosal bleeding. Ultrasonography was performed routinely in the Bandung cohort during the critical phase showing ascites, pleural effusion or a thickened gallbladder in 33 (82.5%) of the enrolled patients.
Fig 1A and 1B shows platelet numbers in the Bandung and Jepara cohorts. Platelets numbers in healthy controls were not determined. As expected, platelet numbers were lowest on day 4–6 after onset of fever, which roughly corresponds with the critical phase of dengue. Fig 1C–1F shows data of platelet activation and reactivity to ex vivo stimulation with the platelet agonist ADP. Compared with healthy controls, circulating platelets of dengue patients were activated on day 1–3 and day 4–6 after fever onset, as shown by increased expression of P-selectin and fibrinogen binding to integrin αIIbβ3. In conjunction with this, ex vivo reactivity to stimulation with ADP was significantly reduced on these days compared with later phases and healthy controls, indicating platelet dysfunction. Similar findings were found for platelet reactivity towards the platelet agonist TRAP (S1 Fig). These findings confirm our earlier findings that platelets in patients with dengue are activated, but also functionally impaired [15].
Plasma levels of active VWF levels are increased in patients with dengue [19], but it is unknown whether this results in higher binding of VWF to platelets. We, therefore, determined the binding of VWF to platelets in the Bandung cohort using flow cytometry. We observed a significantly (P<0.001) higher binding of VWF to circulating platelets of patients in the acute or critical phase of dengue compared to healthy controls (Fig 2A; gating strategy shown in S2 Fig). When plasma VWF was further activated by ex vivo incubation of whole blood with different concentrations of ristocetin, VWF-platelet binding increased. Higher values were obtained in the febrile and critical phase than in the convalescence phase and in healthy controls (Fig 2B). Median plasma VWF antigen (VWF:Ag) concentrations were approximately three fold higher in patients in the acute phase than in controls (median, (IQR); 27.9 μg/ml (21.6–38.8 μg/ml) vs. 8.4 μg/ml (4.4–14.1 μg/ml); P< 0.01) and remained elevated throughout the convalescent phase (Fig 2C). Plasma concentrations of active VWF, which means VWF in which the platelet binding epitope on the A1 domain is exposed, were two-fold higher in the acute phase than in controls (median (IQR) 629 ng/ml (461–860 ng/ml) vs. 300 ng/ml (151–319 ng/ml); P<0.01) (Fig 2D). Levels of ADAMTS-13, an enzyme that cleaves prothrombotic ultra-large VWF, was significantly lower in patients around the critical phase (day 4–6 after fever onset) compared with the convalescent phase (day >13) and healthy controls (S3 Fig).
Platelet-VWF binding was inversely correlated with platelet numbers in enrolment samples (Pearson R = -0.59; P< 0.001) (Fig 2E). Increased VWF-platelet GP1b interaction was reported to be associated with inhibition of platelet αIIbβ3 [21], but we found no correlation between platelet-VWF binding and TRAP- or ADP-induced platelet fibrinogen binding (Fig 2F, data high dose TRAP shown). There was also no correlation between platelet-VWF binding and plasma VWF levels (R = -0.09; P = 0.68), active VWF levels (R = 0.08; p = 0.69) or ADAMTS-13 levels (R = -0.19; P = 0.39) (S3 Fig). These results show for the first time that VWF binding to platelets is a feature of dengue which may contribute to dengue-associated thrombocytopenia.
Finally, we compared platelet parameters between patients with and without bleeding, as well as those with and without plasma leakage. No significant differences in platelet count, platelet-VWF binding, plasma VWF and active VWF levels were found between subjects with or without bleeding complications and between those with or without plasma leakage, except for a significantly higher plasma VWF in the in the plasma leakage group (S4 Fig).
Next, in the Jepara cohort, we determined whether there was also loss of sialic acid from the platelet membrane in dengue infection. Platelet surface sialic acid was measured in patients in the critical or early recovery phase of dengue (median 7 days after fever onset) using the lectins SNA and MAL-II, which primarily binds α-2,6-sialoglycans and MAL-II α-2,3-sialoglycans, respectively. The median (IQR) platelet count at the time of measurement was 57 (37–81) x109/L. Binding of both lectins to sialic acid residues was significantly reduced in patients with dengue, compared with patients with non-dengue febrile illness and healthy controls (Fig 3A and 3B; gating strategy shown in S5 Fig). In 12 participants, SNA and MAL-II binding was also measured in the convalescence phase when platelets numbers had normalized. Both platelet SNA and MAL-II had increased significantly compared with the early dengue phase with median MFI values for SNA of 762 (663–857) vs. 537 (439–752; P = 0.003; Wilcoxon signed rank test) and for MAL-II of 6246 (5171–7188) vs. 5096 (4435–6004; P = 0.001). In the early dengue phase, platelet count correlated significantly with SNA binding (Spearman R 0.48; P = 0.002), but not with MAL-II binding (R 0.28; P = 0.08). In addition, there was also no correlation between SNA or MAL-II binding and levels of VWF:Ag (SNA R = -0.2; P = 0.25 and MAL-II R = 0.13; P = 0.46) or of active VWF (SNA R = -0.2; P = 0.25 and MAL-II R = 0.13; P = 0.46). We compared dengue patients with and without bleeding in the Jepara cohort and did not find statistically significant differences in platelet sialic acid expression or platelet reactivity (S6 Fig).
Next, we investigated mechanisms underlying desialylation. First, we tested whether plasma sialidase activity was increased in acute dengue with a functional sialidase activity assay. As shown in Fig 3C, sialidase activity in the plasma of acute dengue patients was not increased in comparison to the other groups. Further proof that platelet desialylation in dengue infection does not result from increased plasma sialidase activity was provided by mixing PRP from healthy Dutch volunteers with PPP (1:1 volume ratio) from randomly selected patients with dengue (sample from acute and convalescent phase), non-dengue febrile illness or healthy controls (n = 6 each group). As shown in Fig 3D, no differences in platelet desialylation across the groups were observed. Finally, to further confirm these results, we performed glycoprofiling of intact plasma transferrin using mass spectrometry in five dengue patients with the highest plasma sialidase activity in the functional assay [35]. Loss of sialic acid was calculated using the ratio of the undersialylated trisialo-glycosylated transferrin form and this ratio was below the upper reference of normal (<5%) in all five patients (Table 2).
Other possible mechanisms responsible for removal of sialic acid from platelets in dengue infection were further explored in a series of in vitro experiments. A recent study showed that dengue virus NS1 protein is able to induce the expression of sialidase and heparanase in endothelial cells, leading to desialylation of the endothelial glycocalyx layer [38]. We, therefore, investigated whether NS1 also induces desialylation of platelets. Incubation of washed platelets or PRP with different concentrations of NS1 up to four hours did, however, not reduce binding of SNA or MAL-II lectins to sialic acid residues, nor did it increase the binding of RCA to galactose residues or surface expression of Neu1 on platelets indicating that NS1 did not induce platelet desialylation (Figs 4A, 4B and S7A). As expected, desialylation was observed using purified neuraminidase from C. perfringens as a control. NS1 did induce platelet activation, as indicated by increased P-selectin expression (Fig 4C). Next, we investigated whether DENV itself would induce removal of sialic acid. DENV is known to activate platelets [12], which we confirmed in our experiments by finding increased platelet P-selectin expression (Fig 4D). However, removal of sialic acid, as evidenced by a reduction in the binding of SNA or MAL-II, or increased binding of RCA lectin, which is a lectin that binds to desialylated β-galactose residues, was not observed following incubation of washed platelets with DENV2 for 3 or 6 hours (Fig 4E, 4F and 4G; only data at 3 hours are shown).
Next, because our clinical data showed that platelet-VWF binding is increased in dengue infection, we explored whether VWF binding to platelets could be responsible for desialylation of platelets. To induce VWF binding to platelets, we exposed PRP of healthy volunteers to increasing concentrations of ristocetin, which indeed resulted in a dose-dependent increase in VWF-platelet binding (Fig 5A). In the same time, surface sialic acids were reduced as evidenced by a reduction in binding of the lectins SNA and MAL-II. These effects were dependent on VWF binding to GPIbα as blocking the VWF with anti-GPIbα antibodies prevented desialylation (Fig 5B and 5C). VWF-mediated platelet desialylation was confirmed using washed platelets and purified VWF whereby addition of ristocetin resulted into exposure of galactose residues (detected by RCA lectin binding) in a dose response manner (S7B Fig). We further show that ristocetin-induced VWF-platelet binding increases the expression of Neu1 on the platelet surface (Fig 5D). Neu1 is stored in platelet lysosomes and the Neu1 expression on the membrane correlated strongly (R 0.95; P = 0.01) with the lysosomal marker CD63 (Fig 5E). Previous studies showed that the neuraminidase inhibitor oseltamivir is able to inhibit Neu1 activity in platelets [40]. We confirmed these findings by showing that ristocetin induces the binding or RCA lectin to platelets (Fig 5F), which can be inhibited by oseltamivir acid in a concentration as low as 1μM, which is equivalent to plasma levels of the drug [41] (Fig 5G). Lower concentrations of oseltamivir resulted into inconsistency in the RCA lectin binding.
In this study, we describe a new mechanism of platelet clearance and thrombocytopenia in dengue infection. The release of VWF by activated endothelial cells results in excessive VWF binding to platelets, leading to loss of surface sialic acid. We show for the first time that a dengue infection induces a marked increase in VWF binding to circulating platelets, which was inversely related to platelet number. We further show for the first time that circulating platelets in dengue patients have lost surface sialic acid. This occurred in the absence of increased plasma sialidase activity. We further investigated the mechanisms underlying platelet desialylation in dengue infection in vitro and identified increased VWF binding to platelets as the most likely mechanism. Platelets are important in preservation of endothelial integrity [10]. We speculate that the course of events in dengue infection is that endothelial cell activation in the setting of dengue infection leads to release and activation of VWF. The resulting VWF-mediated platelet clearance may in turn contribute to the vascular leakage syndrome that is characteristic of severe dengue.
We and other groups have previously shown that patients with acute dengue have higher VWF plasma levels [19, 42]. Under normal circumstances, VWF does not bind to GPIbα on platelets, but does so only after a conformational change exposing its active A1 domain. Active VWF can be measured using a specific nanobody, and here, we confirm our earlier findings that the amount of active VWF in the circulation is markedly increased in dengue infection [19]. Our present study adds to these earlier data that we now also show directly, using a novel flow cytometry assay, that more VWF has bound to circulating platelets. It thereby supports the notion that VWF is important in the etiology of dengue-induced thrombocytopenia, as binding of VWF to platelets is expected to result in platelet clearance. It is also in concordance with our earlier observations that VWF:Ag is being consumed in children with severe dengue [19]. In contrast to our current findings, VWF:Ag levels in these children admitted to the ICU with severe dengue showed an increasing trend towards discharge, while levels of VWF:propeptide decreased. Because VWF:propeptide is secreted in equimolar amounts as VWF:Ag, this time course was suggestive for VWF consumption. In the current study, VWF consumption was probably less pronounced as most participants had dengue disease without severe complications.
Another important finding of our study was that circulating platelets in dengue patients contained less surface sialic acid. Removal of sialic acid exposes terminal β-galactose and β-GlcNAc resulting in accelerated platelet clearance via Ashwell–Morell receptor [25–28]. Removal of 8–10% of platelet sialic acid was reported to be sufficient to cause platelet clearance from the circulation [43]. Recently, VWF binding to platelets by botrocetin or by platelet refrigeration was shown to induce unfolding of the mechanosensory domain of the GPIbα subunit, leading to GPIb-IX–mediated signaling in the platelet, including calcium mobilization, phosphatidylserine exposure, and likely trafficking of Neu1 to the platelet surface and subsequent desialylation of platelet glycoproteins [29],[44]. We confirm these findings in vitro by demonstrating that induction of VWF binding to platelets with ristocetin results in increased expression of Neu1 on the platelet surface and platelet desialylation. We cannot exclude that ristocetin may execute a VWF-independent effect on platelet desialylation, as previous studies suggested that ristocetin has direct effects on GPIbα and other platelet receptors [45, 46]. However, our findings that ristocetin did not induce platelet desialylation in the presence of GPIbα blocking antibodies or in washed platelets in absence of VWF supports the conclusion that the effects of ristocetin on platelet desialylation are through induction of VWF binding to platelet GP1bα. In addition, the snake venom protein botrocetin, which induces VWF binding to platelets but does not bind GPIbα, also triggers platelet desialylation [44].
Our present study also supports findings from previous studies that platelets are activated during a DENV infection [12, 15]. Despite our observation that activating platelets ex vivo using ADP did not result in removal of sialic acid from the platelet membrane, we cannot exclude that in vivo platelet activation results in trafficking of Neu1 to the membrane where it may desialylate GP1b, leading to increased binding of VWF to the platelet.
We also investigated other possible mechanisms of platelet desialylation in dengue infection. NS1 was recently shown to induce sialidase in endothelial cells, leading to degradation of endothelial bound sialic acid [47], but we did not find platelet desialylation upon exposure of platelets to NS1. In addition, dengue virus does not express neuraminidase, in contrast to other viruses as influenza or bacteria as Streptococcus pneumoniae. Incubation of platelets with cultured dengue virus did induce platelet activation, as previously reported [12–14], but did not induce platelet desialylation. This was further supported by the fact that sialidase activity in plasma of dengue patients was not increased, that plasma of dengue patients failed to induce desialylation of platelets of healthy volunteers and that plasma proteins (transferrin) of dengue patients were not desialylated. The latter is in contrast to an earlier study by Rajendiran et al. [48], who reported increased desialylation of plasma proteins in patients with dengue.
In vitro, VWF-induced platelet desialylation could be circumvented by the neuraminidase inhibitor oseltamivir acid, which is widely used worldwide in the treatment of influenza infections. Interestingly, over the past years, different authors have reported that oseltamivir may increase platelet counts in conditions such as immune thrombocytopenic purpura [49, 50], sepsis [51] and suspected influenza [52]. We are currently carrying out a phase 2 randomized clinical trial to study the effect of oseltamivir on thrombocytopenia and plasma leakage in dengue infection (ISRCTN35227717). In addition, antagonists of VWF, such as the anti-VWF aptamer ARC1779 are nowadays available. ARC1779 was able to reverse thrombocytopenia in VWD type 2B, a condition characterized by excessive binding of VWF to platelets [53]. This reinforces the importance of VWF in platelet clearance and supports the possible use of VWF antagonists as adjunctive therapy for dengue-associated thrombocytopenia.
Our present study confirms our earlier finding that acute dengue infections are not only associated with thrombocytopenia but also platelet dysfunction with hyporeactivity to ex vivo activation [15]. This platelet dysfunction, which may contribute to bleeding complications, may be a consequence of profound platelet activation in the circulation resulting in secondary loss of function. Additional proof for the occurrence of platelet exhaustion in dengue infection was recently provided by a quantitative proteomics study of platelet contents showing exhaustion of the granule-stored PF4/CXCL4 [54]. Gain of function mutations in VWF, as seen in von Willebrand disease type 2B were shown to impair activation of the αIIbβ3 integrin. However, this mechanism appears to be less important in dengue infection, as there was no inverse correlation of platelet-VWF binding with platelet-fibrinogen binding in our cohort.
Different limitations of our studies should also be acknowledged. First, VWF binding to platelets and platelet sialic acid were measured in different cohorts, preventing us to analyze a direct correlation between both parameters. We were able to correlate platelet SNA and MALII lectin binding with plasma levels of active VWF, but this did not show a significant correlation. In our opinion, however, this does not disprove the hypothesis that excessive platelet-VWF binding is responsible for platelet desialylation in dengue infection, because VWF binding to platelets did not correlate with active VWF plasma levels in our cohort. Second, it was not possible to measure the expression of Neu1 on the platelet membrane in the patient samples. Third, the proposed mechanisms for platelet desialylation were analyzed using in vitro studies and samples from healthy individuals, rather than dengue-infected patients due to limited availability of patient material.
In conclusion, acute dengue infections induce the binding of VWF to platelets, which results in the removal of sialic acids from the platelet surface by the actions of endogenous neuraminidase. Neuraminidase inhibitors such as oseltamivir might represent a novel therapeutic option for dengue-associated thrombocytopenia.
|
10.1371/journal.pmed.1002725 | No causal effects of serum urate levels on the risk of chronic kidney disease: A Mendelian randomization study | Studies have shown strong positive associations between serum urate (SU) levels and chronic kidney disease (CKD) risk; however, whether the relation is causal remains uncertain. We evaluate whether genetic data are consistent with a causal impact of SU level on the risk of CKD and estimated glomerular filtration rate (eGFR).
We used Mendelian randomization (MR) methods to evaluate the presence of a causal effect. We used aggregated genome-wide association data (N = 110,347 for SU, N = 69,374 for gout, N = 133,413 for eGFR, N = 117,165 for CKD), electronic-medical-record-linked UK Biobank data (N = 335,212), and population-based cohorts (N = 13,425), all in individuals of European ancestry, for SU levels and CKD. Our MR analysis showed that SU has a causal effect on neither eGFR level nor CKD risk across all MR analyses (all P > 0.05). These null associations contrasted with our epidemiological association findings from the 4 population-based cohorts (change in eGFR level per 1-mg/dl [59.48 μmol/l] increase in SU: −1.99 ml/min/1.73 m2; 95% CI −2.86 to −1.11; P = 8.08 × 10−6; odds ratio [OR] for CKD: 1.48; 95% CI 1.32 to 1.65; P = 1.52 × 10−11). In contrast, the same MR approaches showed that SU has a causal effect on the risk of gout (OR estimates ranging from 3.41 to 6.04 per 1-mg/dl increase in SU, all P < 10−3), which served as a positive control of our approach. Overall, our MR analysis had >99% power to detect a causal effect of SU level on the risk of CKD of the same magnitude as the observed epidemiological association between SU and CKD. Limitations of this study include the lifelong effect of a genetic perturbation not being the same as an acute perturbation, the inability to study non-European populations, and some sample overlap between the datasets used in the study.
Evidence from our series of causal inference approaches using genetics does not support a causal effect of SU level on eGFR level or CKD risk. Reducing SU levels is unlikely to reduce the risk of CKD development.
| Epidemiological studies have shown strong correlations between serum urate (SU) levels and chronic kidney disease (CKD) risk.
Elevated SU levels are often found in patients with CKD, but it is not clear whether high serum urate is a cause of kidney disease or just a common co-occurrence.
Previous studies examining whether SU levels had a causal effect on CKD were limited due to not having large enough samples to detect a true causal relationship if it existed and/or had limitations related to the methodology.
Several clinical trials have been started that aim to use urate-lowering medication to prevent CKD.
To determine whether SU level has a causal effect on CKD, we used a methodology known as Mendelian randomization to test whether genetic variants known to increase SU level also increased the risk of CKD.
We used multiple datasets to perform Mendelian randomization analyses, which included meta-analyses performed across multiple population-based cohorts, 4 individual population-based cohorts, and the large electronic-medical-record-linked UK Biobank.
Across all datasets, we found no significant causal connection between SU level and risk of CKD.
Our findings do not support a causal role of SU level in CKD.
Lower SU levels would be unlikely to translate into reduced risk of CKD.
| Approximately 10% of the global population has chronic kidney disease (CKD) [1,2], which can result in end-stage renal disease, associated with shortened life expectancy and requirement for dialysis or kidney transplantation [3]. There are limited therapeutic options for CKD, with management predominantly focused on control of blood pressure, diabetes, and complications. Hence, there is an intense search for novel therapeutic targets.
Observational studies have consistently shown strong positive associations between serum urate (SU) levels and the risk of CKD [4,5]; however, whether the relation is causal remains unknown. Speculated mechanisms for the potential impact of SU levels on CKD have included nitric oxide and renin-angiotensin pathways [6], stimulation of the renin-angiotensin system [7], and vascular smooth muscle cell proliferation [8]. Furthermore, findings in induced-hyperuricemia rodent models have suggested a causal role of urate in hypertension and associated renal pathophysiology [9]. However, these findings are difficult to directly translate to humans as rodents have considerably lower urate levels owing to functional uricase [9].
As effective medications to lower SU levels are available, the potential causal role of SU level in CKD has become one of the most investigated targets for a renoprotective agent. As such, clinical trials of xanthine oxidase inhibitors (allopurinol or febuxostat) for the endpoint of CKD progression/development are currently underway [10].
Genetic epidemiology can be used to evaluate the causality of risk factors with respect to potential endpoints of interest. This approach, called Mendelian randomization (MR), posits that causality can be inferred because the alleles of a particular exposure-associated genotype are assigned randomly at conception. This can minimize the bias that can occur by confounding and reverse causation in conventional observational studies [11]. Genotypes can be used as a genetically determined lifetime exposure of interest and can be tested for a causal effect on outcomes of interest. For example, MR studies have found that low-density lipoprotein cholesterol is causally related to the risk of coronary artery disease, whereas high-density lipoprotein cholesterol is not [12].
Our objective was to evaluate whether genetic data are consistent with a causal effect of SU levels on estimated glomerular filtration rate (eGFR) and risk of CKD. We performed a series of MR analyses of SU level and eGFR and CKD using genetic variants influencing SU level as the exposure without the influence of confounders.
Our study approach was composed of several complementary components (Fig 1). We first performed a test of heterogeneity to detect the presence of pleiotropy [13] in urate-associated single nucleotide variants (SNVs) identified in a large meta-analysis of genome-wide association (GWA) studies of SU (see S1 Text) [14]. We then conducted 7 distinct MR analyses to evaluate the potential causal role of SU level in eGFR and CKD risk. Since we found significant heterogeneity (see Results), indicating potential pleiotropy, these 7 analytic approaches were specifically chosen to be robust to heterogeneity and pleiotropy in our MR analysis (see S1 Text and below for details). As a positive control endpoint, we assessed whether the same genetic instrumental variables showed the known causal effect of SU on gout. To assess the effect of varying disease endpoint definitions and to control for potential artifacts produced by meta-analyses of heterogeneous populations, we conducted MR analyses using the same SNVs in the UK Biobank, as well as individual-level MR analyses based on 4 population-based cohorts (Atherosclerosis Risk in Communities [ARIC] [15], Coronary Artery Risk Development in Young Adults [CARDIA] [16], Cardiovascular Health Study [CHS] [17], and Framingham Heart Study [FHS] [18]) (Fig 1).
We examined 26 SNVs strongly associated with SU level identified in a GWA meta-analysis study of 110,347 participants of European ancestry conducted by the Global Urate Genetics Consortium (S1 Table) [14].
For MR analyses for eGFR and CKD endpoints, we retrieved GWA study summary statistics for eGFR and CKD from a published meta-analysis conducted by the Chronic Kidney Disease Genetics Consortium (CKDGen) [19]. We used summary statistics for eGFR values calculated from serum creatinine, available in up to 133,413 participants of European ancestry, and for CKD status defined as eGFR < 60 ml/min/1.73 m2, available in up to 12,385 cases and 104,780 controls.
For MR analyses for gout (as a positive control endpoint), we retrieved GWA study summary statistics (effect sizes and standard errors) for gout from a published meta-analysis in 2,115 cases and 67,259 controls of European ancestry from the Global Urate Genetics Consortium [14].
Our additional datasets consisted of electronic-medical-record-linked UK Biobank data (N = 335,212), and 4 population-based cohorts (ARIC, CARDIA, CHS, and FHS; N = 13,425 total). In the UK Biobank, based on recommendations from UK Biobank (http://www.ukbiobank.ac.uk/wp-content/uploads/2018/03/ukb_genetic_data_description_v3.txt), we excluded samples that belonged to any of the following categories: outliers in heterozygosity and missing rates, putative sex chromosome aneuploidy, self-reported non-white British ancestry, and related individuals. Excluded related individuals were defined as 1 individual in each pair with relatedness up to the third degree. In total, 335,212 individuals of white British ancestry remained for analyses. For samples from the UK Biobank, we treated the following ICD-10 codes as indicative of CKD: N03 (chronic nephritic syndrome), N05 (unspecified nephritic syndrome), N17.1 (acute kidney failure with acute cortical necrosis), N17.2 (acute kidney failure with medullary necrosis), N18 (chronic kidney disease), N19 (unspecified kidney failure), N26.9 (renal sclerosis, unspecified), N25.0 (renal osteodystrophy), R80.2 (orthostatic proteinuria, unspecified), I12 (hypertensive chronic kidney disease), I13 (hypertensive heart and chronic kidney disease), Z99.2 (dependence on renal dialysis), Z94.0 (kidney transplant status), and Z49 (encounter for care involving renal dialysis). In total, this produced 5,615 cases. We performed logistic regression for each of the 26 SU-associated SNVs to generate association summary statistics for CKD using this case definition.
In the 4 population-based cohorts (ARIC, CARDIA, CHS, and FHS), we conducted individual-level analyses using a genetic risk score (GRS) of the 26 SNVs associated with SU level. eGFR was recalculated in the 4 prospective cohorts using the Modification of Diet in Renal Disease (MDRD) 4-variable equation. CKD cases were defined as individuals who had an eGFR less than 60 ml/min/1.73 m². A description of the 4 cohorts is provided in the S2 Text.
First, we tested our 26 instrumental SNVs for heterogeneity/pleiotropy using a published test for heterogeneity, the MR-PRESSO [13] global test (see S1 Text for details). We noted the presence of significant heterogeneity/pleiotropy. Accordingly, we selected 7 MR analyses designed to be robust to the presence of heterogeneity in instrumental variables: (i) inverse variance weighted least squares (WLS) regression with a random effects model [20], (ii) MR-Egger regression [21,22], (iii) and (iv) weighted and unweighted median tests [22], (v) and (vi) weighted and unweighted mode-based estimates, and (vii) WLS regression after removing outliers identified by the MR-PRESSO outlier test [13]. For all methods, we used the effect size for the outcome variable (eGFR or CKD) as the response variable, the effect size for the exposure variable (SU) as the predictor variable, and, for weighted methods, the inverse square of the standard error for the outcome variable (eGFR or CKD) as the weight. We used the same 7 methods for analysis of the UK Biobank data. For details of all these methods, see S1 Text. To test the robustness of our analysis to the choice of SNVs, we conducted a leave-one-out analysis, removing each urate-specific SNV separately from the WLS regression test. We also conducted an analysis specifically excluding 2 SNVs: rs12498742 in the SLC2A9 gene and rs2231142 in the ABCG2 gene. These are the 2 SNVs with the most significant effects on SU, and are also the only 2 SNVs identified in the SU GWA analysis as having significant sex-specific effects after applying multiple test correction [14].
For our individual-level analyses in the 4 population-based cohorts (ARIC, CARDIA, FHS, and CHS), we calculated the weighted GRS per individual based on the number of risk alleles for the SNVs and the effect size of these SNVs on SU level based on summary results from the Global Urate Genetics Consortium [14]. Individual-level analyses in these cohorts were performed using linear regression when the outcome was a continuous trait or logistic regression when the outcome was a dichotomous trait, after adjustment for age and sex. An inverse variance weighted meta-analysis was performed across the 4 cohorts. To account for relatedness in the FHS cohort, a linear mixed-effects kinship model implemented in the coxme package [23] was used. To account for relatedness in the ARIC cohort, 66 individuals with known family relationships were removed. The other cohorts are not known to have kinship between individuals. To account for possible nonlinearity in the causal relationship between SU and eGFR, the weighted GRS per individual was stratified into 100 strata to calculate localized average causal effect (LACE), and a fractional polynomial fit was performed [24]. The fractional polynomial fit was performed using the mfp package [25].
Epidemiological associations between SU level and eGFR were assessed using linear regression, and between SU level and CKD using logistic regression, adjusting for age and sex in the 4 population-based cohort studies.
We performed analytic power calculations using mRnd [26] based on the strength of observed epidemiological associations of SU with CKD in the 4 population-based cohorts (odds ratio [OR] = 1.34–1.74 per 1 mg/dl [59.48 μmol/l] SU), the fraction of variance in SU explained by each SNV in the Global Urate Genetics Consortium meta-analysis (0.5% to 3%), and the numbers of cases and controls in the CKDGen meta-analysis (12,385 cases and 104,780 controls).
All statistical analyses were performed using R (R Project for Statistical Computing, Vienna, Austria) or Python 2.7 (Python Software Foundation, Wilmington, DE, US).
We utilized 26 SNVs associated with SU level identified from the GWA study for SU [14] for our MR analyses (S1 Table). We detected significant heterogeneity when testing for a causal effect of SU on both CKD and eGFR (both P < 10−6 for MR-PRESSO global test), which may indicate the presence of pleiotropy. Accordingly, we selected 7 MR approaches considered to be appropriate for instrumental variables displaying potential pleiotropy: a standard WLS regression analysis with a random effects model, 5 alternative MR methods designed to be robust to heterogeneity, and a WLS regression analysis after removing 4 SNVs identified as outliers for the effect of SU on CKD and 11 SNVs identified as outliers for the effect of SU on eGFR (S2 Table). All 7 approaches detected no causal relationship between SU level and eGFR level (Figs 2A and 3A). Similarly, there was no causal effect detected on CKD risk (Figs 2B and 3B). We observed no heterogeneity/pleiotropy after removal of outliers in the MR test of the effect of SU on CKD (P = 0.06); heterogeneity/pleiotropy in the eGFR analysis was still present but reduced (P = 0.004). We also performed leave-one-out analysis with the WLS regression method (S1 Fig), as well as an analysis excluding the 2 SNVs with the most significant effects on SU (SLC2A9 rs12498742 and ABCG2 rs2231142), which are also known to have sex-specific effects (S2 Fig). Both analyses showed no causal effect.
As a positive control, we repeated the same procedure using the same 26 SNVs associated with SU to test for a causal effect of SU on gout. Similarly to the analyses of eGFR and CKD, we observed significant heterogeneity among the 26 SNVs (P < 10−6 for MR-PRESSO global test). However, unlike for eGFR and CKD, all 7 approaches showed a highly significant causal effect of SU on gout (P < 10−3 for all tests; S3 Fig). We observed no heterogeneity/pleiotropy after removal of outliers in the MR test of SU’s effect on gout (P = 0.09). This result serves as a positive control of our approach as it is consistent with the known causal role of SU in gout [27].
To account for varying definitions of the CKD endpoint, we performed an additional analysis using a clinical definition of kidney disease based on electronic medical records in the UK Biobank. In total, we identified 5,615 CKD cases and 329,597 controls using this definition. We applied the same 7 MR analyses to the 26 SU-associated SNVs. We observed no significant causal relationship of SU level with kidney disease (P > 0.05 for all analyses; S4 Fig).
To account for limitations of using aggregated GWA data, such as heterogeneity among study populations in the meta-analyses or unaccounted for population stratification within the sample, we performed an individual-level analysis on 4 population-based cohorts. Within these cohorts, we observed a highly significant association of a urate-specific GRS composed of the 26 urate-associated SNVs with SU (beta = 1.06; 95% CI = 1.00 to 1.13; P = 7.06 × 10−211). However, the GRS was not associated with either eGFR (beta = −0.42; 95% CI = −1.05 to 0.20; P = 0.18) or CKD risk (OR = 1.05; 95% CI = 0.89 to 1.23; P = 0.59) (Table 1). We repeated the same analysis after stratifying by sex and age, and found similar results (S3 Table). We also tested for a nonlinear causal relationship using a fractional polynomial fit [24]. The nonlinear model does not appear to fit the data better than the linear model (P = 0.49), and still produces a null result for a causal effect of SU on CKD (P = 0.34).
We also specifically examined rs12498742 in the SLC2A9 gene, which encodes the GLUT9 transporter (for glucose and urate) in the renal proximal tubule and is the largest contributor to genetic control of SU level, explaining 3% of variance [14]. This SNV was strongly associated with SU (beta = 0.37, P < 10−700) but neither with eGFR (beta = 0.002, P = 0.06) nor with CKD (OR = 0.99, P = 0.53).
These null associations contrasted with conventional epidemiological association analyses. In a meta-analysis of the same 4 population-based cohorts (ARIC, CARDIA, CHS, and FHS), we observed that an equivalent (1 mg/dl) increase in SU level was associated with reduced eGFR level (beta = −1.99; 95% CI −2.86 to −1.11; P = 8.08 × 10−6) and increased risk of CKD (OR = 1.48; 95% CI 1.32 to 1.65; P = 1.52 × 10−11), after adjustment for age and sex (Table 2).
We assessed the statistical power of our MR study given the sample size and the variance in SU explained by the 26 SNVs we used as instrumental variables. We calculated that our MR analyses would have greater than 99% power to detect a statistically significant effect at an alpha rate of 5%, if causality between SU and CKD were present at the strength indicated by observational epidemiology (OR = 1.5 per 1 mg/dl of SU) (Table 3).
In this study, we investigated a potential causal role for SU level in the development of CKD using a series of complementary MR analyses. Despite previous observational study findings that SU levels were strongly associated with the risk of incident CKD [4,5], our MR analyses found no evidence for a causal role of SU level for eGFR level or incident CKD. In contrast, our positive control MR analysis demonstrated that SU level was causal for the risk of gout, which is consistent with a previous study that showed similar results [28].
Unlike previous studies based on smaller sample sizes, our power calculations show that our study is sufficiently powered to assess a causal relationship between SU level and CKD. One study that examined only eGFR (not CKD) reported that increased SU levels due to SNVs in urate transporter genes were associated with increased eGFR only in men (which is opposite to the expected epidemiological association) [29]. However, this study did not account for pleiotropy, which may explain these unexpected findings. The second MR analysis reported null findings for both eGFR and CKD; however, the study was not sufficiently powered to assess causal relationships for CKD and also did not account for pleiotropy [30]. A third study observed no causal effect of SU levels on renal function in 3,734 Chinese individuals but observed significant effects in some subpopulations: females, individuals under 65 years, individuals with normal eGFR levels, current smokers, and individuals with high fasting glucose levels [31]. This study similarly did not account for pleiotropy and did not directly test for CKD. Importantly, the sample size of the current study is much larger (N > 400,000) and therefore has higher power than the previous studies above. Our failure to detect by MR the expected epidemiological association of SU with CKD/eGFR is therefore not due to lack of power. We suggest instead that it is due to our MR analyses being robust to unknown confounders and reverse causation, which can create positive correlations in observational epidemiological studies.
Clinical trial data on lowering SU levels for preventing CKD progression to date have been conflicting [32,33]. Moreover, no randomized controlled trials have been conducted using targeted interventions to lower SU levels for the prevention of incident CKD, although trials evaluating the role of xanthine oxidase inhibitors in disease populations such as patients with early CKD or type 1 diabetes are ongoing. Furthermore, to date, 2 randomized controlled trials among adolescents with hyperuricemic pre-hypertension or stage-1 hypertension found that urate-lowering therapy lowered blood pressure, a strong risk factor for CKD [34,35], whereas a similarly designed trial among adults did not find such a benefit [36]. Nevertheless, these studies did not address the causal role of urate reduction in the prevention of incident CKD in population-based participants, which would be the relevant context of our study. Furthermore, the potential benefit of xanthine oxidase inhibitors could exclusively come from reducing oxidative stress by inhibiting superoxide generation, rather than from SU reduction [37]. Our findings suggest that SU reduction alone would not result in prevention of incident CKD, consistent with a recent study that showed that the initiation of allopurinol in patients with gout was not associated with a change in CKD risk [38]. Finally, our results are relevant to the impact of lowering urate and do not rule out the potential benefit of superoxide reduction resulting from xanthine oxidase inhibition [37].
Potential limitations of this study and, in particular, MR deserve comment. First, MR analysis requires suitable SNVs to act as instrumental variables, and selection of inappropriate or unrepresentative instrumental variables may undermine the validity of the study. Here, we used the SNVs significantly associated with SU in a previously published GWA study. We also performed an additional analysis using a single SNV with a strong effect on SU and a well-understood mechanism (rs12498742 in the SLC2A9 gene), as well as an analysis excluding both this SNV and a second large-effect SNV (rs2231142 in the ABCG2 gene). Second, an assumption of MR is that the SNVs used as instrumental variables are not subject to “horizontal pleiotropy,” meaning that the SNVs should not have pleiotropic effects on the outcome outside of the target biomarker or risk factor. We have used MR approaches designed to be robust to horizontal pleiotropy to account for this. Third, our MR findings cannot predict with absolute certainty that therapeutics lowering SU levels will not result in a lowering of CKD risk, since the effects of genetic variation may not be exactly the same as the effects of a therapeutic intervention. Nevertheless, studies have suggested that genetic findings can predict the effect of drugs on disease outcomes [39]. Fourth, the study is focused on individuals of European ancestry; therefore, it is unclear whether our results can be generalized to non-European populations. Fifth, there are some overlapping samples between the SU and CKD GWA datasets (61% of samples in the SU GWA dataset, 50% in the eGFR GWA dataset, and 43% in the CKD GWA dataset), which can affect the accuracy of MR tests [40]. However, this would only affect our study by producing a false positive result, and we observe only negative results. Furthermore, we also performed an MR analysis with non-overlapping datasets (the SU GWA dataset and the UK Biobank dataset for CKD), and found the same negative result. Sixth, the SU and CKD GWA studies both aggregated heterogeneous populations with differing distributions of age, sex, and other potentially important features, and it is possible that the differences between the populations mask the effect of SU on CKD. However, these studies did account for age and sex as covariates, and we performed individual-level analyses that found a negative result when stratifying by age and sex. Seventh, some individuals included in the GWA datasets for SU and CKD may have been taking urate-lowering medication, which may distort the relationship between SU and CKD. We were not able to remove these individuals from our sample for the GWA summary statistic analysis. However, we did remove individuals taking urate-lowering medication from our individual-level analyses in the population-based cohorts, and these analyses found a similar result to the GWA summary statistic analysis.
In conclusion, our MR analyses do not support a causal effect of SU level on eGFR or CKD. Our results suggest that lowering SU levels would be unlikely to translate into risk reduction for incident CKD.
|
10.1371/journal.pbio.2003586 | Recurrent excitation between motoneurones propagates across segments and is purely glutamatergic | Spinal motoneurones (Mns) constitute the final output for the execution of motor tasks. In addition to innervating muscles, Mns project excitatory collateral connections to Renshaw cells (RCs) and other Mns, but the latter have received little attention. We show that Mns receive strong synaptic input from other Mns throughout development and into maturity, with fast-type Mns systematically receiving greater recurrent excitation than slow-type Mns. Optical recordings show that activation of Mns in one spinal segment can propagate to adjacent segments even in the presence of intact recurrent inhibition. While it is known that transmission at the neuromuscular junction is purely cholinergic and RCs are excited through both acetylcholine and glutamate receptors, here we show that neurotransmission between Mns is purely glutamatergic, indicating that synaptic transmission systems are differentiated at different postsynaptic targets of Mns.
| Motoneurones (Mns) are the last elements of the networks that command, coordinate, and actuate the most essential of behaviours: movement. Their activation triggers muscle contractions, and the many diseases affecting Mns cause progressive and fatal paralysis. We show here that Mns themselves form an interconnected network and that activity in one segment of the spinal cord can propagate reciprocally to neighbouring segments, thus constituting a positive feedback loop that can amplify the strength of motor output. Remarkably, while Mns excite muscles through release of the neurotransmitter acetylcholine, our data show that the synapses between Mns operate through glutamate, providing a rare example of differentiation of transmitter systems according to the postsynaptic targets.
| Motoneurones (Mns) are the ultimate neural targets of effector commands issued from the central nervous system. Their activity is modulated by an intricate network of interneurones [1] that affect the spatial and temporal distribution of excitation to different motor pools [2]. Mns also receive direct inputs from supraspinal tracts and sensory afferents, and their outputs are not confined to the peripheral muscles but also include excitatory collateral terminals to Renshaw cells (RCs).
Early anatomical studies [3] have shown that Mn axon collaterals have large ramifications that invade the motor nucleus and may form synaptic contacts with other Mns. An early evidence of functional connectivity between Mns was found in the adult cat [4] but was attributed to the presence of gap junctions. The first proof of the existance of chemical synapses between Mns was found in tadpoles [5]. A simlar observation in juvenile rats was attributed to the presence of afferent fibers in the ventral roots (VRs) [6], a possibility that was subsequently ruled out [7, 8]. Recurrent excitation was also described in neonatal mice [9], in which VR stimulation elicited a small postsynaptic response in Mns.
None of the previous studies provided a comprehensive analysis of the extent of recurrent excitation, and they only illustrated a few recordings of small evoked currents. Furthermore, there are contrasting reports on the type of receptors mediating recurrent excitation, with evidence showing sensitivity to either glutamatergic antagonists [6], cholinergic antagonists [7], or both [9]. Here, we perform a systematic study of recurrent excitatory circuitry and demonstrate that recurrent excitation between Mns is strong and it is maintained throughout development into maturity. Our data show that while recurrent excitation between intrasegmental and intersegmental Mns is comparable in size, fast-type Mns receive a 10-fold greater amount of recurrent excitation compared to slow Mns. Under normal physiological conditions, recurrent excitation can override recurrent inhibition, and Mn firing in one spinal segment propagates to neighbouring segments. Remarkably, while acetylcholine and a mixture of acetylcholine and glutamate act at the neuromuscular junction and RC synapses, respectively [8, 9], neurotransmission between Mns is purely glutamatergic.
We performed paired recordings to measure the efficacy of unitary connections in fluorescently labelled Mns innervating gastrocnemius. Simultaneous infrared and confocal imaging was used to identify and patch fluorescent Mns in a dorsal horn–ablated spinal cord (Fig 1A). Strychnine (0.5 μM) and gabazine (3 μM) were applied to block recurrent inhibition. Mns were patched in whole-cell voltage clamp while putative presynaptic cells were stimulated in loose-cell attached configuration (see Materials and methods) until an evoked response was detected in the postsynaptic cell. Fig 1B shows an example of a paired recording with an average evoked current of −34 pA. A location map constructed from 14 out of the 18 recorded pairs (Fig 1C) shows that connected Mns tended to be within 150 μM from one another but with no systematic relationship between distance and size of response (range 11–125 pA).
The rise and decay times of evoked currents (Fig 1D, response size colour coded) were fast, with a median rise time of 0.59 ms and decay time of 3.75 ms. There was, however, no correlation between either of the kinetic parameters and the size of response. In oblique slice preparations (see Materials and methods), we assessed the pharmacology of evoked responses (Fig 1E). The postsynaptic current was fully abolished by bath application of 50 μM D-(-)-2-Amino-5-phosphonopentanoic acid (APV) and 2 μM 2,3-Dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulfonamide disodium salt (NBQX), to block N-Methyl-D-aspartate (NMDA) and α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors, respectively (Fig 1F, top). Identical results were obtained from all 4 pairs tested (Fig 1F, bottom).
Because the tested unitary connections might have represented a specific local subset of the entire population of Mn–Mn synapses, we investigated the pharmacology of currents evoked by VR stimulation, thus pooling responses to all inputs from a given segment (Fig 2A). In the example of Fig 2A–2C, we simultaneously recorded from an RC (Fig 2B, top, red) and an Mn (Fig 2B, bottom, blue). The RC response shown includes a second component originating from a gap junction [10], contacting a neighbouring RC in which VR stimulation evoked an action potential.
Whereas bath application of glutamate antagonists resulted in a reduction of the RC response to approximately 50%, the response in the Mn is completely abolished. The remaining cholinergic component of the response in the RC was blocked by further application of 10 nM methyllycaconitine (MLA) and 5 μM dihydro-β-erythroidine (DHβE) to block α7 and αβ receptors, respectively (Fig 2C). Group data from 16 Mn recordings are illustrated in Fig 2D. The mean latency (±SEM) of responses of 1.60 ± 0.13 ms and the corresponding response jitter, quantified using the standard deviation of the latencies, was 0.08 ± 0.01 ms, consistent with monosynaptic responses to VR stimulation. In all cases, application of glutamate antagonists resulted in complete suppression of evoked currents (Fig 2D).
While the data from Fig 2D were obtained from juvenile mice (P7–14), we performed similar recordings from more mature animals (P15–25) to determine whether pure glutamatergic transmission is preserved throughout development. Fig 2E and 2F shows that the response is fully suppressed by glutamatergic blockade. In 20 Mns recorded in voltage clamp (black, Fig 2G) or in current clamp (blue, to reduce the duration of the stimulus artefact), glutamatergic antagonists entirely suppressed responses, whereas prior cholinergic blockade had no significant effect (n = 6, Wilcoxon sign-rank z = −0.53, P = 0.600). The mean latency of the responses was 1.54 ± 0.22 ms, with a jitter of 0.13 ± 0.03 ms.
We next investigated whether recurrent excitation could propagate across segments in coronal preparations (see Materials and methods) from juvenile mice (P7–14) in which Mns innervating gastrocnemius were labelled. Fig 3A illustrates recurrent excitatory postsynaptic currents (rEPSCs) recorded in L5 (left) and L4 (right) Mns while stimulating the L5 (upper, blue trace) or L4 (lower, red trace) VR. The rEPSC size from 43 recordings from L4 and L5 Mns is plotted against the distance from the L4/L5 border, colour coded to represent responses evoked by L4 (red circles) or L5 (blue circles) VR stimulation (Fig 3B). There were no obvious differences in rEPSC size between the 2 stimulated roots or between L4 and L5 Mns. Comparison of rEPSCs from responses to VR stimulation from the same segment or neighbouring segment showed no significant differences (Fig 3B right, Wilcoxon rank-sum z = 0.61, P = 0.541). Despite the lack of correlation between the size of the current and the position of the recorded Mn with respect to the stimulated VR, we observed a very broad distribution of amplitudes of rEPSCs, with sizes ranging from 60 to more than 5,000 pA (see S1 Fig, showing the rEPSCs recorded in all Mns in which the position was registered.)
In the experiments above, the rEPSCs were pharmacologically isolated by blocking fast inhibitory transmission with strychnine and gabazine, in order to avoid potential bias in the current measurements, due to the opening of the large inhibitory conductances associated with the activation of RCs following VR stimulation. However, because the cord might become hyperexcitable following full block of synaptic inhibition, we tested whether propagation of recurrent excitation across segments was preserved even with intact inhibition. In 3 different preparations, we stimulated either L4 or L5 VRs and recorded the rEPSCs from 21 Mns located in the adjacent L5 or L4 segment (see individual responses in S2 Fig). Amplitudes from experiments performed without block of synaptic inhibition are shown as red (descending) or blue (ascending) crosses in Fig 3B (left). The box and whisker plot (Fig 3B, right) shows that a comparison of intersegmental responses in the presence and the absence of antagonists of inhibition shows no significant differences in size (Wilcoxon rank-sum z = 0.48, P = 0.635).
The mean latency (±SEM) of responses of 1.64 ± 0.09 ms was consistent with monosynaptic activation and not significantly different from those from oblique slices (1.60 ± 0.13 ms, Wilcoxon rank-sum z = −0.90, P = 0.384). Latencies of responses from Mns in neighbouring segments (1.77 ± 0.13 ms) were longer compared to those recorded from within the same segment that was stimulated (1.43 ± 0.09 ms, Wilcoxon rank-sum z = −2.43, P = 0.015). The corresponding response jitters were very small both within (0.08 ± 0.03 ms) and across (0.08 ± 0.01 ms) segments, and a comparison between the two showed no significant difference (Wilcoxon rank-sum z=−0.58, P = 0.559). These results demonstrate that, while the latency of synaptic responses may be greater across segments compared to within segments, they are both mediated by monosynaptic connections between Mns within and across segments.
It has been recently demonstrated that in zebrafish, Mns are electrically coupled with V2a interneurones that in turn project back to motor nuclei with glutamatergic synapses [11]. It is therefore possible that VR-evoked, synchronized antidromic spikes could elicit firing in V2a interneurones, which in turn could give rise to the observed short-latency, low-jitter EPSCs. In order to test this possibility, we used a dorsal horn–ablated preparation taken from mice selectively expressing enhanced green fluorescent protein (EGFP) in V2a interneurones. While stimulating a VR, we performed simultaneous whole-cell recordings from an Mn and cell-attached recordings from a V2a interneurone. At a stimulation intensity that elicited a maximal rEPSC in the Mn, no spikes could be evoked in the V2a interneurones in 3 different preparations (n = 46, V2a interneurones tested). We also performed simultaneous whole-cell recordings of Mns and V2a interneurones, and in n = 9 cells, we could not elicit any synaptic or electrically mediated current from the interneurone (one example of a double recording in each preparation is shown in Fig 4B) throughout the stimulated or adjacent segment of the spinal cord. The position of all recorded cells is overlayed with a picture of a cord in Fig 4A. These results therefore exclude any involvement of V2a interneurones in mediating the rEPSCs.
We next assessed whether the magnitude of recurrent excitation was related to the intrinsic properties of postsynaptic Mns. Two types of Mns were identified according to their firing pattern at rheobase in current clamp recordings [12]. The first type (Fig 3C, left, purple) has high rheobase, produces delayed firing with a pronounced increase in firing rate during positive current application, and is associated with fast-type units. By contrast, the second type (Fig 3C, right, green) has a lower rheobase and immediate firing with little change in spike frequency, characteristic of slow-type units [12]. High rheobase (Fig 3D, left) and accelerating initial firing (Fig 3D, middle) were correlated with the size of rEPSCs, with median values of 1,814 in the delayed-firing cells and 267 in the immediate-firing cells. Comparison between the 2 groups confirmed a significant difference (Fig 3D right, Wilcoxon rank-sum z = 3.72, P<0.001).
Differences between the 2 cell types are also associated with their passive properties, with delayed-firing Mns showing lower resistances (median 22 MΩ) and higher capacitances (median 237 pF) than their immediate-firing counterparts (median resistance 44 MΩ, median capacitance 125 pF), both at statistically significant levels (Wilcoxon rank-sum |z|≥2.94, P≤0.003). Recurrent excitatory responses were recorded from delayed-firing (Fig 3E, purple) and immediate-firing (Fig 3E, green) cells in both voltage clamp (top) and current clamp (bottom). Pooling all cell types together, correlations were observed between the size of response and resistance or capacitance (Spearman |r|≥0.516, P<0.001, Fig 3F).
The presence of strychnine and gabazine during electrophysiological recordings precluded evaluation of whether recurrent excitation could override recurrent inhibition. We therefore conducted calcium imaging experiments in mice selectively expressing GCaMP6s in Mns to evaluate the propagation of recurrent excitation across different segments with recurrent inhibition intact. Fig 5A–5C illustrates a coronal preparation with a suction electrode applied to the L5 VR (Fig 5A, left). Calcium signals were acquired throughout the dorsal motor column of L4 and L5 (Fig 5A, middle), with 146 ms frame interval before, during, and following a train of 3 VR stimulations at 30 Hz. The signal from regions of interest, defined within the outline of Mn somata, was evaluated for the period of acquisition under control conditions, in the presence of 0.5 μM strychnine and 3 μM gabazine, and following application of 50 μM APV and 2 μM NBQX (Fig 5A, right).
While the latency of the rEPSCs recorded electrophysiologically was short (1.64–1.77 ms) and exhibited a very low jitter (0.08 ms), a greater variability in the apparent onset of the calcium signals was observed (Fig 5A, middle). However, analysis of each individual trace confirmed that the latency of responses were always within 150 ms to 300 ms. Because the responses were evoked by a single trial of a train of 3 stimuli applied over a 100-ms period, and images were acquired at a frame interval of 146 ms using a calcium indicator with slow kinetics (approximately 200 ms [13]), a variable shift of 1 to 2 frames was expected.
In control, recurrent excitation evoked spikes in Mns from both L4 and L5 segments, as shown by the running medians and interquartile ranges of the relative fluorescence signal (Fig 5B, red) throughout both segments. Bath application of strychnine and gabazine resulted in substantial amplification of responses throughout the motor column (Fig 5B, green), whereas additional application of glutamatergic antagonists abolished responses from the L4 segment and substantially attenuated those from L5 Mns (Fig 5B, blue). The residual response in L5 Mns reflects antidromic activation. Scattergrams, colour-coded by regions, comparing control responses to those during application of inhibitory antagonists (Fig 5C, left) and additional glutamatergic blockade (Fig 5C, right), confirm that while responses were greater in the lumbar regions closer to the stimulated VR, the relative effects of block of recurrent inhibition—or excitation—were similar throughout L4 and L5.
Group data from 461 Mns from 7 preparations are shown in Fig 5D, comparing responses within and across segments for the 3 conditions. In control, responses were significantly greater within the stimulated segment compared to outside (Fig 5D, left, Wilcoxon rank-sum z = 8.50, P<0.001), and these difference were maintained after block of inhibition (Fig 5D, middle) and excitation (Fig 5D, right) (z≥7.23, P<0.001). Pooling Mns from both segments, blockade of inhibition consistently increased the signal (Wilcoxon sign-rank z−18.59, P<0.001), whereas a significant reduction in signal was observed following additional application of glutamatergic antagonists (z = 18.20, P<0.001). Residual firing was mostly confined to Mns within the stimulated segment through antidromic activation, thus confirming the purely glutamatergic nature of recurrent excitation.
Our experiments show that strong recurrent excitation between Mns is maintained throughout development, and fast-type Mns receive greater recurrent excitation than slow ones. We demonstrate that synaptic transmission between Mns is purely glutamatergic. While it could be argued that the observed small unitary postsynaptic responses (approximately 100 pA) would have little effect on the excitability of Mns whose somata are very large, VR stimulation evoked responses usually exceeding 1 nA, indicating extensive convergence of segmental Mn populations.
Very few studies have examined, in any detail, recurrent excitation between Mns. In both available electrophysiological studies [7, 9], the size of the recurrent EPSCs is not reported, even though the examples shown suggest a small size—of the order of 100 pA—corresponding to the lower bound of our observations. The previous studies were performed on young neonatal animals (P0–P4), while we obtained all our recordings starting from the second week of age. Therefore, our estimates of the size of recurrent excitation is not directly comparable to that reported by [9] and [7] because most of the increase in input conductance and dendritic arborization occurs between P2–P4 and P10–P13 [14]. In the only previous study performed on more mature animals (P10–P20 rats [6]), recurrent EPSPs were reported in the range of 3 to 15 mV, similar to our observations in current clamp. While the conduction velocity reported in this study (0.35–0.96 m/s) would point towards stimulation of unmyelinated fibers, the consistent observation of anti-dromically induced firing strongly suggests that Mn axons were indeed stimulated, and the slow conduction thus most likely reflects incomplete myelination at this developmental stage.
The variation in the magnitude of rEPSCs is associated with Mn classification into delayed and immediate-firing types, with larger responses systematically observed in the delayed-firing, low-resistance, and high-capacitance cells. Our results are therefore consistent with a structural connectivity in which the fast-type larger Mns receive stronger recurrent excitation compared to slow-type smaller cells. This pattern of connectivity suggests that recurrent excitation could play a role in sequential recruitment of fast-type units during motor tasks in which progressively increasing muscular forces are needed. Alternatively, recurrent excitation might represent a closed-loop amplification circuit that reinforces and increases the firing rate preferentially in fast-type Mns and thus rapidly increases muscle contraction strength when required. Distinguishing between these two possibilities would require a full characterization of presynaptic Mns in order to determine whether it is slow or fast Mns that are preferentially connected to fast Mns.
In neonatal animals, VR stimulation can induce fictive locomotion [8] and entrain the spontaneous rhythmic bursting induced by block of inhibition [15]. Furthermore, optogenetic activation or silencing of motor pools alters the frequency and phase of chemically induced fictive locomotion [16]. These effects cannot be explained solely by recurrent excitation and may provide evidence for Mn collaterals contacting unidentified interneurones [17]. A similar finding has been recently reported in zebrafish, for which gap junctions between Mns and V2a interneurones can alter the swimming pattern [11]. In our experiments however, an involvement of gap junctions between Mns and V2a interneurones seems unlikely because in none of our recordings could we detect spikes in identified V2a interneurones following VR stimulation.
It is possible, however, that Mns synapse onto other, so far unidentified interneurones that, in turn, could project back to Mns. While the existence of such a disynaptic pathway is possible, it would not account for the large, constant latency responses observed both within and across segments. Indeed, in our electrophysiological recordings during pharmacological blockade of recurrent inhibition, we often observed a late disynaptic component. At present, we cannot ascertain whether the disynaptic current results from premotor interneurons or from orthodromic activation of Mn pools that were not antidromically activated by VR stimulation. In either case, such recruitment implies the existence of a positive-feedback amplifying circuit whose tendency to reverberate may be suppressed by recurrent inhibition.
The recurrent excitation characterised in the present study includes predominantly a monosynaptic component, and this is evidenced by three observations. First, the connectivity between Mn pairs must have been monosynaptic. Second, the latency of responses within (1.43 ± 0.09 ms) and between (1.77 ± 0.13 ms) lumbar segments were within the time-scale of neurotransmission through only a single synapse. Finally, the response jitters within (0.08 ± 0.03 ms) and between (0.08 ± 0.01 ms) segments were very small and virtually identical. These observations are only consistent with a monosynaptic connectivity between Mns both within the same segment and across neighbouring segments. While the occurrence of synaptic projections between Mns crossing spinal segments may be regarded as unusual, it is perfectly compatible with the known rostrocaudal distribution of Mn dendritic trees, which may span over 1 millimetre in juvenile mice with little or no change into adulthood [14].
A glutamate receptor–dependent effect on Mn EPSPs evoked by VR stimulation has been reported previously [6], but it was attributed to afferent fibres within the root [18], a possibility now excluded by subsequent labelling studies [8]. A previous study has reported a purely cholinergic response to VR stimulation in a small proportion (2/9) of Mns [9]. Across all electrophysiological recordings of the present study, however, there was not a single instance of a cholinergic component. The origin of such a discrepancy may result from differences in maturity because in the previous study [9] neonatal mice (P0-P4) were used, while our experiments were performed on mice of at least partially weight-bearing age (P7–P25).
Neurotransmission between Mns is purely glutamatergic, yet following normal maturation, the neuromuscular junction is solely cholinergic [19] and synaptic transmission of recurrent collaterals onto RCs is mixed with both cholinergic and glutamatergic components [20]. This remarkable dissociation demonstrates a differentiation of neurotransmission systems on the basis of the different postsynaptic targets of Mns. However, the presence of vesicular glutamate transporters in Mn collateral terminals is still controversial. Immunohistochemistry and in situ hybridization studies have reported the expression of the vesicular glutamate transporter 2 (VGlut2) in some Mn terminals onto RCs that are either positive [9] or negative [21] for the vesicular acetylcholine transporter. These respective findings indicate either coexistence or segregration of cholinergic and glutamatergic transmission of Mns onto RCs.
Others, however, have not detected the presence of VGlut2—or any other vesicular glutamate transporter—in Mn terminals [8, 22]. It is possible that such discrepencies arise from undetectable albeit functional expression levels of VGlut2. Another possibility is the existence of an unidentified vesicular glutamate transporter [8, 22]. This hypothesis is supported by the presence of glutamate-releasing C-fibres in the dorsal horn that are nevertheless negative for all known vesicular glutamate transporters [23, 24]. Because many Mn terminals may contain more aspartate than glutamate [25], it has been proposed that the released neurotransmitter could be aspartate. However, aspartate alone cannot activate AMPA receptors that mediate responses of RCs [26] or of the Mns characterised in the present study. Glutamate therefore remains the most likely candidate.
All experiments were carried out in accordance with the Animal (Scientific Procedures) Act (Home Office, UK, 1986) and were approved by the UCL Ethical Committee, under project licence number 70/7621. Intramuscular injections were performed under inhaled isofluorane anaesthesia and by a surgical procedure supervised and approved by the Veterinary Surgeon named by the Home Office, UK. Before being euthanized, animals were administered terminal anaesthesia via intraperitoneal injection of a mixture of ketamine and xylazine (80 mg/Kg and 10 mg/Kg, respectively).
Experiments were performed on preparations obtained from male or female mice bred using a C57BL/6J background. For electrophysiological experiments with simultaneous recordings from Mns and RCs, a transgenic strain—in which the EGFP is expressed under the control of the promotor of the neuronal glycine transporter GlyT-2 [27]—was used to label glycinergic interneurones. Simultaneous recordings from Mns and V2a interneurones were performed on mice expressing EGFP under the control of the Chx10 transcription factor [28].
Following anaesthesia by intraperitoneal injection of a mixture of ketamine/xylazine (80 mg/kg and 10 mg/kg, respectively), both juvenile and mature mice were decapitated and the spinal cord dissected in normal icecold aCSF containing (in mM) 113 NaCl, 3 KCl, 25 NaHCO3, 1 NaH2PO4, 2 CaCl2, 2 MgCl2, and 11 D-glucose (same solution was used for recording). The spinal cord was then glued onto an agar block and affixed to the chamber of a vibrating slicer (HM 650V, Microm, ThermoFisher Scientific, UK). We used a slicing solution containing (in mM) 130 K-gluconate, 15 KCl, 0.05 EGTA, 20 HEPES, 25 D-glucose, 3 kynurenic acid, and ph 7.4 with NaOH [29]. For cutting oblique slices, the cord was glued to an agar block cut at a 45-degree angle, with the ventral side facing the direction of the blade [20].
For coronally sliced preparations in which the dorsal horns were ablated, the cord was glued horizontally with the ventral surface facing upwards. A blade was used to transect the cord at the L1–L2 boundary at an angle that allowed visualization of the exact position of the central canal under a dissection microscope. The vibratome blade was then aligned to the central canal, and the ventral portion of the cord was sliced away from the dorsal part. Alignment with the central canal was essential to ensure a consistent dorsoventral level of the ablation across different preparations and to retain the dorsal motor nuclei near the cut surface of the tissue.
Identical procedures were used for juvenile (P7–14) and mature (P15–25) animals. For older animals, we routinely cut the first slice within 8 minutes following decapitation. Because spinal cord preparations are extremely sensitive to anoxia, especially prior to slicing, we found that minimizing the time to obtain the first slice consistently resulted in viable preparations with healthy Mns [20].
All recordings from postsynaptic Mns were performed with a Molecular Devices Axopatch 200B amplifier, filtered at 5 kHz and digitized at 50 kHz. Patch pipettes were pulled to resistances in the range of 0.8–2 MΩ when filled with (in mM) 125 K-gluconate, 6 KCl, 10 HEPES, 0.1 EGTA, 2 Mg-ATP, pH 7.3 with KOH, and osmolarity of 290 to 310 mOsm. During voltage clamp recordings, Mns were clamped at −60 mV with series resistances in the range of 2 to 10 MΩ compensated by 60% to 80%.
During paired recordings, loose cell–attached stimulation was used to evoke spikes in putative presynaptic Mns using an ELC-03X (NPI Instruments, Bauhofring, Germany) amplifier and a 4–5 MΩ pipette filled with normal aCSF [30]. Within each field of view (240 μm × 240 μm), typically up to 40 Mns could be visualized, but in order to avoid excessive mechanical disturbances of the tissue while recording from the postsynaptic target, only those located within the first 150 from the surface could be tested for connections. Typically, 1 Mn for every 40 tested was connected to the recorded cell.
VR stimulation was delivered to evoke rEPSCs in Mns using a glass suction electrode whose tip was cut to correspond with the size of the VR [31]. The stimulation intensity was increased until the size of the rEPSC remained constant, typically at 5× threshold. In order to exclude direct stimulation of the ventral white matter, VRs were only used if they were of sufficient length to afford no possible physical contact between the slice and suction pipette. This was tested before and after each recording by confirming that side-to-side movement of the suction pipette resulted only in movement of the root and not the slice. The position of the stimulating electrode for all the dorsal horn–ablated preparations is shown for the experiments performed in the presence and absence of block of synaptic inhibition in Supplementary S1 Fig and S2 Fig, respectively, and ranged between 200 μm and 1500 μm from the edge of the glass pipette to the point of entry of the VR.
For measuring the size of the excitatory response in some Mns, for which it was necessary to prevent action potentials, cells were hyperpolarised below their resting membrane potential. Measurements of synaptic current and potentials from these recordings were adjusted to their predicted value at −60 mV, assuming a reversal potential of 0 mV for excitatory conductances. All electrophysiological experiments were performed in the presence of 0.5 μM strychnine and 3 μM gabazine except where stated otherwise. Where indicated, excitatory receptors were blocked using APV, NBQX, MLA, or DHβE.
In order to label Mns innervating the ankle flexor gastrocnemius muscle, intramuscular injections were performed 2 to 5 days prior to recording. Inhalant isofluorane was used for the induction and maintenance of anaesthesia. Traction was applied to the lower limb, and an incision was made through the skin and deep fascia overlying the muscle. A Hamilton syringe loaded with a glass needle was used to inject 1 μl of CTB-Alexa-Fluor-555 (0.2% in 1× phosphate buffer saline) into the middle of the muscle belly over a period of at least 1 minute. The skin was closed by suture using a buried stitch before cessation of anaesthesia and recovery.
Calcium imaging experiments were performed on animals selectively expressing the genetically encoded calcium indicator GCaMP6s in Mns. These mice were generated by crossing mice expressing Cre under the control of choline-acetyltransferase (ChAT-Cre; JAX mouse line number 006410) with animals with the gene expressing GCaMP6 flanked by a flox-Stop cassette (JAX mouse line number 028866 [32]). Upon recombination with Cre, GCaMP6 is selectively expressed in Mns and other cholinergic cells of the offspring. Because the only other population of ChAT-positive lumbar spinal cells are the cholinergic partition neurones located around the central canal, there was no ambiguity in the identification of Mns from their basal GCaMP6 fluorescence and position within the motor nuclei.
Dorsal horn–ablated coronal slice preparations (P9–12) were used for imaging experiments to visualise the dorsal motor nuclei in the L4 and L5 segments containing Mns innervating tibialis anterior, gastrocnemius, and peroneus longus close to the cut surface. A laser-scanning confocal unit (D-Eclipse C1, Nikon, UK) with a diode laser (λ=488 nm, power output from optic fibre 3–5 mW) was used to locate and record calcium signals from Mns reaching a depth of approximately 100 μm from the surface. Fields of 128 × 64 pixels (pixel size 1.38 μm and dwell 7.2 μs) were scanned with a frame interval of 146 ms over different regions throughout the dorsal motor column. Trains of 3 stimulii at 30 Hz were delivered to the VR (L4 or L5) while images were being acquired from at least 1 s before the onset of the first stimulus pulse. For each field, calcium signals were acquired for a total of 35 frames corresponding to approximately 5 s, and the position of each field was recorded.
Posthoc analysis was performed to quantify Mn responses. Within each field, single Mns were identified by their fluorescence, and regions of interests were defined by the contour profiles of their somata. The time course of excitation was measured using the change in mean fluoresence following stimulation divided by the baseline average. In some cases, slow drifts in fluorescence was corrected by fitting an exponential to the initial trace before the stimulus. Changes in fluorescence exceeding 2 standard deviations of the baseline noise were measured over a 1 s window following VR stimulation.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.